UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

E-scape : interactive visualization of single cell phylogenetics and spatio-temporal evolution in cancer Smith, Maia A 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2016_november_smith_maia.pdf [ 27.7MB ]
JSON: 24-1.0313405.json
JSON-LD: 24-1.0313405-ld.json
RDF/XML (Pretty): 24-1.0313405-rdf.xml
RDF/JSON: 24-1.0313405-rdf.json
Turtle: 24-1.0313405-turtle.txt
N-Triples: 24-1.0313405-rdf-ntriples.txt
Original Record: 24-1.0313405-source.json
Full Text

Full Text

E-scape: Interactive visualization of single cellphylogenetics and spatio-temporal evolution incancerbyMaia A. SmithB.Sc. Biological Sciences, Simon Fraser University, 2014a thesis submitted in partial fulfillmentof the requirements for the degree ofMaster of Scienceinthe faculty of graduate and postdoctoralstudies(Bioinformatics)The University of British Columbia(Vancouver)August 2016© Maia A. Smith, 2016AbstractCancers evolve over time and space, producing a dynamic, heterogeneousmixture of related cells. Reconstructing the evolution of each cancer re-quires sequencing tumour cells and processing resulting data with novelcomputational and statistical methods. These advances have led to numer-ous insights, both clinical and biological, but the ability for a biologist tointeract with these results across an experimental workflow remains limited,with expert intuition often injected only through cumbersome iterations ofdata analysis. Here we describe E-scape, a visualization tool suite enablinginteractive analysis of cancer heterogeneity and evolution. The suite includesthree tools: TimeScape and MapScape for visualizing population dynamicsover time and space, respectively, and CellScape for visualizing evolution atsingle cell resolution. The tools integrate phylogenetic, clonal prevalence,mutation and imaging data to generate intuitive views of a cancer's evolu-tion.iiPrefaceThis dissertation contains my original work, with contributions from manymembers of the Shah Lab for Computational Cancer Biology.Sohrab Shah provided the motivation for the development of this work.Cydney Nielsen contributed much mentorship for the design of visual-ization tools presented in Chapters 4 and 3.Hossein Farahani, Fong Chun Chan, and both Andrew McPherson andAndrew Roth contributed the user needs for the visualization tools describedin Chapter 4, 3 and 5, respectively.Daniel Machev provided the technical guidance for JavaScript develop-ment.The visualization tool in Chapter 5 was applied in McPherson, Andrew,et al. “Divergent modes of clonal spread and intraperitoneal mixing in high-grade serous ovarian cancer.” Nature genetics (2016).All research contained within the dissertation has been submitted forpublication.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixGlossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . xiiDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Clonal evolution in cancer . . . . . . . . . . . . . . . . . . . . 11.2 Current techniques for measuring clonal evolution . . . . . . . 31.3 The need for visualizing clonal evolution data . . . . . . . . . 51.4 Existing visualization standards for single cell analyses . . . . 71.5 Existing visualization standards for temporal clonal evolution 81.6 Existing visualization standards for spatial clonal evolution . 91.7 E-scape tool suite overview . . . . . . . . . . . . . . . . . . . 92 Computational workflow . . . . . . . . . . . . . . . . . . . . . 11iv2.1 Software architecture of the E-scape tool suite . . . . . . . . . 112.2 Sharing E-scape visualizations . . . . . . . . . . . . . . . . . . 132.3 Runtime execution of E-scape tools . . . . . . . . . . . . . . . 132.4 Code availability . . . . . . . . . . . . . . . . . . . . . . . . . 143 Visualization of temporal clonal evolution . . . . . . . . . . 153.1 An overview of TimeScape . . . . . . . . . . . . . . . . . . . . 153.2 Colouring the clonal phylogeny . . . . . . . . . . . . . . . . . 203.3 Representing small clonal populations . . . . . . . . . . . . . 213.4 X-axis spacing of emergent clones . . . . . . . . . . . . . . . . 223.5 Vertical genotype layouts . . . . . . . . . . . . . . . . . . . . 243.6 Perturbation events . . . . . . . . . . . . . . . . . . . . . . . . 264 Visualizing clonal evolution at single cell resolution . . . . 274.1 An overview of CellScape . . . . . . . . . . . . . . . . . . . . 274.2 Vertically ordering the single cells in the phylogeny and heatmap 344.3 Targeted mutation ordering . . . . . . . . . . . . . . . . . . . 355 Visualization of spatial clonal evolution data . . . . . . . . 405.1 An overview of MapScape . . . . . . . . . . . . . . . . . . . . 405.2 The power of multi-clone selection to highlight clonal evolu-tion in a single patient . . . . . . . . . . . . . . . . . . . . . . 445.3 Concurrent visualization of temporal and spatial clonal evo-lution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455.4 Automatic radial layout of samples . . . . . . . . . . . . . . . 485.5 Automatic image cropping . . . . . . . . . . . . . . . . . . . . 485.6 Handling minor clones . . . . . . . . . . . . . . . . . . . . . . 505.7 Distribution of clones within the cellular aggregate . . . . . . 506 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536.1 Significance and contribution . . . . . . . . . . . . . . . . . . 536.2 Limitations and future improvements . . . . . . . . . . . . . . 546.3 Potential applications . . . . . . . . . . . . . . . . . . . . . . 546.4 Final word . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55vBibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56A Data used in figures . . . . . . . . . . . . . . . . . . . . . . . 61A.1 Learning the clonal prevalences for prostate cancer patientA21 (Gundem et al. [1]) . . . . . . . . . . . . . . . . . . . . . 61A.2 Mutational information for prostate cancer patient A21 (Gun-dem et al. [1]) . . . . . . . . . . . . . . . . . . . . . . . . . . . 62A.3 Time-series clonal prevalences for ovarian cancer patient 7(McPherson and Roth et al. [2]) . . . . . . . . . . . . . . . . . 62A.4 Acute myeloid leukemia (Ding et et al. [3]) . . . . . . . . . . 63A.5 Learning the phylogenetic tree for triple-negative breast can-cer patient (Wang et al. [4]) . . . . . . . . . . . . . . . . . . . 63B Input requirements and examples for E-scape tools . . . . 64C Supplemental links . . . . . . . . . . . . . . . . . . . . . . . . 75C.1 Interactive MapScape of metastatic ovarian cancer patient 1from McPherson and Roth et al. [2] . . . . . . . . . . . . . . 75C.2 Interactive MapScape of metastatic ovarian cancer patient 3from McPherson and Roth et al. [2] . . . . . . . . . . . . . . 75C.3 Interactive MapScape of metastatic ovarian cancer patient 7from McPherson and Roth et al. [2] . . . . . . . . . . . . . . 75C.4 Interactive TimeScape of metastatic ovarian cancer patient 7from McPherson and Roth et al. [2] . . . . . . . . . . . . . . 76C.5 Interactive MapScape of metastatic prostate cancer patientA21 from Gundem et al. [1] . . . . . . . . . . . . . . . . . . . 76C.6 Interactive MapScape of metastatic prostate cancer patientA22 from Gundem et al. [1] . . . . . . . . . . . . . . . . . . . 76C.7 Interactive CellScape of the xenograft SA501 single cell datafrom Eirew et al. [5] . . . . . . . . . . . . . . . . . . . . . . . 76C.8 Interactive TimeScape of AML patient Patient 933124 fromDing et al. [3] . . . . . . . . . . . . . . . . . . . . . . . . . . . 76C.9 Interactive TimeScape of AML Patient 933124 from Ding etal. [3], created with the centred genotype layout . . . . . . . . 76viC.10 Interactive TimeScape of AML Patient 933124 from Ding etal. [3], created with the spaced genotype layout . . . . . . . . 77C.11 Interactive TimeScape of triple negative breast cancer patientfrom Wang et al. [4] . . . . . . . . . . . . . . . . . . . . . . . 77C.12 Interactive TimesScapes of transformed follicular lymphomapatients from Kridel and Chan et al. [6] . . . . . . . . . . . . 77viiList of TablesTable B.1 Input requirements for CellScape. Available in the CellScapehelp menu. . . . . . . . . . . . . . . . . . . . . . . . . . . . 64Table B.2 Input requirements for TimeScape. Available in the TimeScapehelp menu. . . . . . . . . . . . . . . . . . . . . . . . . . . . 67Table B.3 Input requirements for MapScape. Available in the Map-Scape help menu. . . . . . . . . . . . . . . . . . . . . . . . 69Table B.4 Examples for E-scape tools. Available in the help menus. . 71viiiList of FiguresFigure 1.1 Schematic workflow for obtaining clonal and single cellevolution data. . . . . . . . . . . . . . . . . . . . . . . . . 4Figure 1.2 Existing visualizations for temporal clonal evolution. . . . 8Figure 2.1 Software architecture of E-scape tool suite. . . . . . . . . 12Figure 3.1 An overview of TimeScape. . . . . . . . . . . . . . . . . . 16Figure 3.2 TimeScape clicking functionality. . . . . . . . . . . . . . . 18Figure 3.3 The power of TimeScape to compare clonal dynamicsacross multiple patients. . . . . . . . . . . . . . . . . . . . 19Figure 3.4 TimeScape can easily plot extensive time-series data. . . . 20Figure 3.5 Method for colouring the clonal phylogeny. . . . . . . . . 20Figure 3.6 Minor clone expansion. . . . . . . . . . . . . . . . . . . . 21Figure 3.7 X-axis spacing of emergent clones. . . . . . . . . . . . . . 23Figure 3.8 Vertical genotype layouts in TimeScape. . . . . . . . . . . 25Figure 4.1 CellScape of single cell copy number data - triple-negativebreast cancer patient from Wang et al. [4]. . . . . . . . . . 28Figure 4.2 The phylogenetic views avaiable within CellScape. . . . . 29Figure 4.3 The selection tool. . . . . . . . . . . . . . . . . . . . . . . 30Figure 4.4 The tree trimming tool. . . . . . . . . . . . . . . . . . . . 31Figure 4.5 An overview of time-series CellScape. . . . . . . . . . . . 33Figure 4.6 Inspection of the heatmap in CellScape. . . . . . . . . . . 34Figure 4.7 Hierarchical clustering method for ordering targeted mu-tations in a heatmap - xenograft SA501, Eirew et al. [5]. . 36ixFigure 4.8 Hierarchical clustering method for ordering targeted mu-tations in a heatmap - ovarian cancer patient 2, McPher-son and Roth et al. [2]. . . . . . . . . . . . . . . . . . . . 37Figure 4.9 Hierarchical clustering method for ordering targeted mu-tations in a heatmap - ovarian cancer patient 9, McPher-son and Roth et al. [2]. . . . . . . . . . . . . . . . . . . . 38Figure 5.1 An overview of MapScape. . . . . . . . . . . . . . . . . . 41Figure 5.2 Evolutionary progression viewable in a single patient. . . 44Figure 5.3 TimeScape and MapScape applied to a single patient. . . 46Figure 5.4 Automatic radial layout of samples. . . . . . . . . . . . . 48Figure 5.5 Automatic cropping of anatomical images. . . . . . . . . . 49Figure 5.6 Comparing methods for distributing clones within a cel-lular aggregate. . . . . . . . . . . . . . . . . . . . . . . . . 52Figure A.1 Calculating clonal prevalence given a clonal phylogenyand mutation cluster cancer cell fractions. . . . . . . . . . 61xGlossaryamplicon sequencing: Next-generation sequencing performed at specificsites in the genome.bezier curve: A parametric curve often used in computer graphics to rep-resent a smooth curve in two dimensions.clone: A group of cells with similar genotypes and a common ancestry.d3.js: A JavaScript library for the development of custom web-based datavisualizations.JavaScript Object Notation: A data format suitable for JavaScript soft-ware development.tumour cellularity: The proportion of cancer cells in a tumour sample.variant allele frequency (VAF): The frequency of the variant (mutated)allele in all sequenced reads.whole genome sequencing: Next-generation sequencing performed onthe whole genome.xiAcknowledgmentsI would like to thank Sohrab P. Shah for his great support and encourage-ment during my studies. I would also like to thank my committee membersTamara Munzner and Carl Hansen for guiding the production of qualitywork from the standpoints of both visualization and application. An addi-tional thank you to Steven Jones, who chaired my defense. Thanks also toSharon Ruschkowski and Carolyn Lui for their contributions on the man-agerial side of my studies. Of course, I would also like to acknowledge theCanadian Institute of Health Research (CIHR) and the National Sciencesand Engineering Research Council (NSERC) for funding my studies.xiiDedicationTo my family and friends, for their unrelenting support during my academicpursuits.xiiiChapter 1IntroductionCancer genomics datasets are challenging to work with due to their com-plexity, volume and imperfect nature. This work is focused on providingvisualization tools that address these challenges for the effective analysis ofevolution in cancer.1.1 Clonal evolution in cancerCancer cell populations, or clones, emerge through branched evolutionaryprocesses, changing in prevalence due to variation in fitness and neutral drift.Advances in high-throughput sequencing and computational methods [7] [8]have enabled the quantification of clonal dynamics, leading to an increasedunderstanding of how cancer evolution impacts treatment resistance, cancerprogression and metastasis.Three landmark papers from 2012 showcase the ability of these novelmethods to advance the field of cancer research through the quantificationof clonal dynamics. Shah et al. [9] were the first to show that treatment-naivetriple-negative breast cancers present a spectrum of clonal diversity, fromhighly heterogeneous to clonally pure. Nik-Zainal et al. [10] reconstructedthe evolutionary histories of breast cancers and articulated a pattern ofclonal evolution whereby minimal clonal expansion precedes the accumula-tion of genomic aberrations in a single clone, enabling its expansion and1enlarging the tumour to a diagnostic level. Ding et al. [3] reconstructed theevolutionary histories of acute myeloid leukemias to discover two patternsof clonal evolution contributing to relapse, where a clone that is either (i)dominant or (ii) minor at diagnosis will accumulate mutations and expandat relapse.While the aforementioned studies infer clonal architecture from bulktumour samples, other studies have harnessed the power of single cell se-quencing to directly observe clonal structure and dynamics in cancer [5, 11,12, 13, 14]. In 2011, Navin et al. [15] presented a method for resolving copynumber profiles of single cells, and demonstrated its utility in the inferenceof clonal evolution. In the two cancers they study, the distances betweenobserved clones and their root suggest the occurrence of punctuated ratherthan gradual clonal evolution at the copy number level. In 2015, Wanget al. [4] published a single cell sequencing approach suitable for single-cellresolution detection of mutations. They apply this approach to two breastcancer patients and show the gradual accumulation of point mutations gen-erating tumour heterogeneity in these patients.Although a single tumour sample can reveal a great deal about the evo-lutionary history of a cancer, we have seen that cancers can be highly hetero-geneous, and thus, one tumour sample cannot possibly provide the completestory. To gain a more comprehensive view of cancer evolution, some studieshave employed multi-region sequencing to investigate the spatial heterogene-ity within a single cancer. For instance, Gerlinger et al. [16], de Bruin etal. [17], McPherson and Roth et al. [2] and Bashashati et al. [18] discoveredthe presence of spatial heterogeneity in all studied patients with metastaticrenal-cell carcinoma, nonsmall cell lung cancer, metastatic ovarian cancerand high-grade serous ovarian cancer, respectively. These studies reinforcethat one tumour sample cannot fully represent the clonal composition itsoriginating tumour, nor any related metastases.As a growing number of studies address the impact of clonal evolution ontherapeutic success and cancer progression, the field experiences an equiv-alent acceleration in the development and application of novel techniquesfor measuring clonal evolution. Lentiviral lineage tracking, for instance, is2a novel method for directly tracking clonal dynamics in cancer by followingthe replication of single cells (e.g. Kreso et al. [19]). Another recent devel-opment enables the simultaneous detection of both single cell transcriptomeand genome data [20] [21], a combination bringing the possibility of measur-ing the intraclonal diversity of gene expression. The spatial heterogeneity ofcancer is commonly acknowledged, and methods such as in situ PCR enablethe researcher to detect genomic heterogeneity in spatially varying regionsof a tissue sample (e.g. Raub et al. [22]).1.2 Current techniques for measuring clonal evo-lutionIn order to interpret the spectrum of genomic diversity within a cancer, sim-plifying assumptions are applied to the data, grouping cells into operationalunits (clones) assumed to exhibit shared phenotypic properties. As pheno-type is difficult to measure directly, mutations in the genome are used assurrogate clonal marks, providing an accessible approach to grouping cellswith shared genotype.Figure 1.1 summarizes the process of obtaining clonal genotype, preva-lence and phylogeny from patient samples. The first step in this process isbulk or single cell sequencing performed on the whole genome (whole genomesequencing) or at specific sites in the genome (amplicon sequencing). Thesequencing reads are aligned to a reference genome, and mutations are calledwhere reads differ from the reference. Mutations occur with varying frequen-cies in a sample, and each mutation therefore carries an associated variantallele frequency (VAF): the frequency of the variant allele in all sequencedreads. Given whole genome sequencing data, copy number profiles may alsobe inferred.3Sequence Whole genome AmpliconInfer clones,prevalences andphylogenies0.20.50.3Clonal prevalence, phylogenySingle cell phylogenyObtain genomicprofiles Copy Number MutationGGAGATTATCCCCAGGAGATTATCCCCACCAGAATATCCCCA024Obtain sample(s)Bulk Single cellSampling dimension Space TimeSample typeAlign reads...GACCTATCCGAATATATAGCCCCCCGGGCGCGCGCTTTAA...GCTTTAA......GACCTATCCGAATATATAGCCCCCCGGGCGCGCGCTTGGGCGCGCGCTTTAA...Reference genomeSequenced readsFigure 1.1: A schematic workflow for obtaining clonal and single cell evo-lution data. At the start of the workflow, the patient's tumours aresampled over multiple spatial regions and/or time points. Each sam-ple, processed as single cells or bulk, is sequenced at specific mutationsites or over the whole genome. The sequencing data is aligned to areference genome, at which point the genomic profiles (copy numbervariants or targeted mutations) may be obtained. Subsequently, com-putational methods may utilize the genomic profiles as input to inferthe phylogeny of clones or single cells, as well as the prevalence of eachclone.For single cell data, the next stage in the inference process involves group-ing single cells into clonal populations using novel statistical methods suchas the Single Cell Genotyper [23]. From these methods, the clonal preva-4lences are directly observable. Furthermore, the single cell and establishedclonal genotypes may be used to infer single cell and clonal phylogenies,respectively, using maximum parsimony.When samples are sequenced in bulk, the process of clonal inferenceinvolves the assumption that mutations with similar cellular prevalences(the fraction of tumour cells with a particular mutation) occur at the samebranch in the clonal phylogeny. The VAF differs from cellular prevalencedue to the influences of copy number variation and tumour cellularity (theproportion of tumour versus normal cells in the sample). Correcting for theseinfluences and grouping mutations by their cellular prevalences representsthe next stage of clonal inference. From the mutation clusters, statisticalmethods such as CITUP [7] infer the clonal prevalences and phylogeny.Once clonal prevalences and phylogenies are established, the evolution-ary analysis of cancer proceeds by tracking the abundances of clonal popu-lations in time [5, 24, 25] and/or anatomic space [2, 1, 26].1.3 The need for visualizing clonal evolution dataClonal evolution datasets are complex, voluminous, and imperfect. Fortu-nately, these properties can each be addressed by visualization.The complexity of clonal evolution datasets presents two challenges. Thefirst challenge lies in extracting meaningful information from the many datatypes required to describe an individual cancer. These include anatomicaldiagrams and images, tree structures of single cell and clonal phylogeneticrelationships, and tables of genomic aberrations and clonal prevalences. Ef-fective visualization integrates these data types, enabling an intuitive andcomplete appreciation of the clonal dynamics occurring within the patient.The second challenge of complex biological datasets is to involve biomedicalinvestigators in the analysis process. Due to the statistical nature of thesedatasets, insights are most often driven by computational scientists whoimpose objective interpretations, leading to quantitatively justified conclu-sions often under theoretical assumptions. Visualization provides a conduitfor disease-focused biomedical investigators to engage with genomic datasets5and offer complementary insights in a multidisciplinary process of data anal-ysis and inference.The challenges arising from voluminous datasets are those of efficientaccessibility and interpretation. These challenges will only be exacerbatedin the future as data volumes continue to increase. For instance, recent sin-gle cell experiments range from measuring on the order of 50 pre-selectedmutations, to genome wide assays of hundreds to thousands of cells [2, 27].Moreover, each cell potentially contains hundreds or thousands of copy num-ber segments. Automated visualization systems can efficiently display largedatasets, quickly guiding the eye to patterns in need of interpretation.The imperfections within clonal evolution datasets can arise from multi-ple sources throughout the data acquisition and processing pipeline. If theseimperfections are often left unnoticed, the analyst may misconstrue noise assignal. Visualization systems can reveal imperfections that are otherwisewell disguised. In addition to separating signal and noise, visualization sys-tems can validate data processing approaches and/or expose the need forimprovements.An effective visualization design can provide many benefits, but often-times, even the most intelligent static visualization cannot fully accommo-date the user's needs. Incorporating interactivity, however, enables the in-jection of human intuition which can enhance the discovery process. Forinstance, Google Maps provides the best data-inspired route from origin todestination, but is unaware of locations to avoid—such as this afternoon'sstreet-blocking parade—or seek out—such as the neighbourhood rose gar-den that just came into bloom. By enabling the user to interact with andalter the map, however, Google Maps empowers the user to enhance theirexperience above Google’s initial efforts.61.4 Existing visualization standards for single cellanalysesCurrently, a single published tool exists for visualizing single cell data.Ginkgo [28] displays a single cell phylogeny, and on request, the copy numberprofile of each cell. However, Ginkgo is limited in several important ways:it lacks the capacity to display all single cell genomes of a dataset for directcomparison; it does not support views across temporal or spatial dimensions;and it cannot support targeted mutation representations of cells.71.5 Existing visualization standards for temporalclonal evolutionabFigure 1.2: Two existing visualizations for temporal clonal evolution.(a) Figure 5e from Findlay et al. [29] shows a manually-created diagramof temporal clonal evolution.(b) The fishplot R package [30] shows an automated visualization oftemporal clonal evolution.In the case of temporal clonal dynamics, a standard, manually-created rep-resentation has emerged in the literature (e.g. Fig. 1.2a), depicting eachclone emerging from its ancestor and changing in prevalence as a function oftreatment interventions. While this representation is effective for commu-nicating the concept of temporal clonal evolution, its manual developmentis time-consuming and prohibits integration into automated, reconfigurableanalysis pipelines.A recent software development, the fishplot R package [30] (Fig. 1.2b),automates this representation as a static figure. However, the package is8limited in three ways. Firstly, the mutational information is inaccessiblefrom the figure, and thus, alternative means are required to access muta-tional elements characterizing each clone. Secondly, the emergence of clonesis centered with respect to their ancestors—this splits clones on the verticalaxis and thereby negatively influences the visual perception of clonal preva-lence. Thirdly, quantitative clonal prevalence values are absent from theview, confining this figure to a qualitative representation.1.6 Existing visualization standards for spatial clonalevolutionMulti-region sequencing studies are becoming increasingly common, but fewvisualization standards exist for spatially-varying data describing metastaticcancers. Two recent studies of metastatic cancer visually represent the clonaldynamics of metastasis using a series of anatomical diagrams, phylogeniesand matrices [1] [16]. However, an improved visual design would representall relevant data in a single comprehensive figure, reducing the need to men-tally relate the elements of multiple figures. Another study of pancreaticcancer approaches the ideal visualization design by presenting a clonal phy-logeny annotated with mutational information and partitioned by site [26].However, the direct integration of both an anatomical diagram and clonalprevalence information would enhance reader intuition about the spatialdistribution and dominance of clonal populations.1.7 E-scape tool suite overviewHere we present the Evolutionary Landscapes (E-scape) tool suite includingthree automated visualization tools for the analysis of temporal and spa-tial evolution in cancer at clonal and single cell resolution. The tools areuser-friendly, freely accessible, and easily integrated into web documents forsharing. Launched from R as htmlwidgets [31], the output web-based visu-alizations support interactivity with the data through brushing and linking9of views in the web browser. This human interaction can in turn facilitateinsights into (1) the genomic drivers of branched evolution, (2) clonal fit-ness and metastatic potential, (3) treatment sensitivity and resistance and(4) metastatic progression.10Chapter 2Computational workflow2.1 Software architecture of the E-scape tool suiteWe developed a novel software platform for visualizing clonal evolutiondatasets, schematically represented in Figure 2.1. The platform consistsof two main components: a data handling component developed in R [32](version 3.2.2), and a web component written in JavaScript (version 1.7).Using an R-based platform facilities integration with other popular pack-ages in the bioinformatics community, while JavaScript provides a route topowerful web based graphing libraries such as d3.js [33] (version 3.5.6). Aninterface between the two components, the R htmlwidgets framework [31](version 0.6), enables the production of html output (optimized for Chromeversion 52.0.2743) from user input in R.The internal software architecture of each E-scape tool consists of thefollowing layers. In the data handling component, the R htmlwidget re-ceives inputs in the form of data tables and strings. Once data validationis complete, the component's data processing layer harnesses the power ofR to perform computationally intensive data processing tasks, such as con-verting thousands of copy number segments into a pixel matrix. The datais then transformed into a list of JavaScript Object Notation elements—adata format suitable for JavaScript software development—in preparationfor migration to the web component. The web component receives this list11∆   ●  ●  ●    Clonal phylogeny ∆   ●  ●  ●*   Targeted mutations∆   ●  ●    Clonal prevalences∆   ●    Image reference∆   ●     Tumour sample locations∆   ●*   Copy number variants∆   ●  Single cell phylogeny ∆    ●    Annotations (genotype, time point)MapScapeTimeScapeCellScape● Required input● Optional inputUser prepares inputs Input Type∆ Data table∆ StringOnly provide one*User calls E-scape htmlwidgetData sent to Web componentVisualization displayed in browserUser may share visualizationse.g. R> timescape(clonal_prev = my_prev, tree_edges = my_edges) - SVG/PNG download button - Call htmlwidget from RMarkdown, then knit to htmld3.js library translates data into visual elementsInput validation and computationally intensive processingData handling component (R)Data restructured for d3.js library useWeb component (JavaScript)Figure 2.1: A schematic of the E-scape tool suite software architecture.Within the data handling component, the user prepares the inputs,whose content and data type will vary depending on the desired vi-sualization. The user then calls the desired E-scape htmlwidget. Atthis point, the user input is validated and any computationally inten-sive processing will occur before the data is sent to the web component.The data is then restructured for use in d3.js [33] functions which trans-late the data into visual elements displayed in the user's browser. Theuser may share a static visualization by directly pressing the downloadbuttons on the visualization. Alternatively, the user may share an in-teractive visualization by calling the htmlwidget from an RMarkdowndocument, then knitting the document to html.12and further transforms the data into a format suitable for d3.js, a JavaScriptlibrary for the development of custom web-based data visualizations. d3.jsthen performs two functions: (i) mapping data elements to their final visualpositions and forms and (ii) defining user interactivity. The resultant inter-active, shareable visualization is displayed in the user's browser. Note thatinput-specific warnings will be issued in the R console, and visualization-specific warnings will be issued in the browser console.2.2 Sharing E-scape visualizationsAll E-scape visualizations may be exported as a static PNG or SVG byclicking the appropriate button on the top panel. In order to share theinteractive visualizations directly, the user may call the TimeScape, Map-Scape or CellScape tools from within an RMarkdown document, which willproduce a shareable html document. Note that multiple interactive visual-izations may be shared on the same html document.2.3 Runtime execution of E-scape toolsTo run the E-scape tools, the user must have R (version >= 3.2.2) installedon their computer. The user then runs the following commands:i n s t a l l . packages (” dev too l s ”)l i b r a r y ( dev too l s )i n s t a l l b i t b u c k e t (”MOBCCRC/<tool name>”)l i b r a r y (<tool name>)To view the input requirements (Fig. 2.1 and Tables B.1, B.2, B.3)and example code (Table B.4) for each tool, the user runs the followingcommand:?<tool name>To run the example code for each tool, the user runs the following com-mand:13example(<tool name>)2.4 Code availabilityTimeScape, MapScape and CellScape are open source and available ongithub at:• https://bitbucket.org/MO BCCRC/timescape• https://bitbucket.org/MO BCCRC/mapscape• https://bitbucket.org/MO BCCRC/cellscape14Chapter 3Visualization of temporalclonal evolution3.1 An overview of TimeScapeA dynamic clonal composition results from the evolutionary processes thatact on a single tumour over time. Viewing this temporal clonal evolution re-quires specialized visualization beyond static bar or scatter plots, which cancapture changes in clonal prevalence over time, but cannot capture (i) thehierarchical relationship of clones to one another, (ii) the tumour-shrinkingeffects of treatment interventions, nor (iii) the mutational content under-lying the clonal dynamics. We therefore built TimeScape (Fig. 3.1) whichcaptures all these elements for the analysis of temporal clonal evolution data.To the right of the main plot is the clonal phylogeny, and beneath themain plot is the mutation table—these features provide evolutionary and ge-nomic contexts, respectively. The plot axes show clonal prevalence verticallyand time horizontally. To preserve clonal hierarchy within the plot, clonesare nested according to their branched descendance (e.g. Fig. 3.1a). Ateach data-associated time point (denoted by a faint white line), the heightof each clone accurately reflects its proportionate prevalence. These rigiddata points form the anchors for bezier curves—parametric curves often usedin computer graphics to represent a smooth curve in two dimensions—that15Clonal PrevalenceChemotherapyT0 Diagnosis RelapseTime PointClonalPhylogenyTIMESCAPE SVGPNGSearch:Showing 379 entriesClone Clone ID Chrom Coord1 1 25540211 1 119653321 1 18952534abClonal PrevalenceChemotherapyT0 Diagnosis RelapseTime PointClonalPhylogenyTIMESCAPE SVGPNGFigure 3.1: A TimeScape visualization displaying the clonal evolution as ex-pressed in AML patient 933124 from Ding et al. [3]. For the interactiveTimeScape, see Supplemental link C.8.(a) In the nested view, descendant clones emerge from their ancestors.(b) In the clonal trajectory view, each clone changes in prevalence on itsown track. The clonal hierarchy is maintained by the vertical orderingand horizontal offsetting of clones.visually represent the dynamic transitions between time points. It shouldbe noted that bezier curves may mislead the user, as they are simple ap-proximations for the underlying sparsely-sampled biology. Future researchmodeling patterns of clonal dynamics will clarify the accuracy of these ap-proximations, but growth patterns are likely less gradual than representedin TimeScape.During tumour progression, selective pressures, such as treatment in-16tervention, contribute to clonal dynamics and should be apparent withinthe visualization. In the case of acute myeloid leukemia (AML) Patient933124 published in Ding et al. [3], chemotherapeutic interventions likelyinfluenced in the clonal sweep occurring at relapse (Figure 3.1). To displaysuch tumour-shrinking interventions, plot height encodes tumour size (Fig.3.1a “Chemotherapy” time point).Commonly, a descendant clone will replace its ancestor, creating a falseillusion of ancestral expansion. This illusion can be seen in the TimeScapeof the AML patient (Figure 3.1)—prior to “Relapse”, the mid-blue clonevertically expands to encompass its descendant, the dark-blue clone. Twofeatures of TimeSweep clarify such misconceptions. Firstly, the user maymouseover each clone to display its temporally varying cellular prevalence(as in Fig. 4.5c). Secondly, the nested organization of clones may be inter-actively disabled, revealing a new perspective that assigns each clone to aseparate track while retaining hierarchical intuition through intelligent ver-tical ordering and horizontal offsetting of clones (Fig. 3.1b). Both featuresreveal the disparate prevalences between the ancestor and its descendant.The expansion of a clone indicates high fitness of its acquired mutationalcontent. These mutations are immediately accessible from the mutationtable when filtered by clicking the expanding clone in the clonal phylogeny(e.g. Fig. 3.2a). Furthermore, clicking each mutation will reveal its variantallele frequency across time, potentially reflecting its influence on tumourprogression (e.g. Fig. 3.2b).17abFigure 3.2: The clone- and mutation-clicking functionality in TimeScape,shown on a TimeScape visualization of AML patient 933124 from Dinget al. [3]. For the interactive TimeScape, see Supplemental link C.8.(a) Clicking a clone filters the mutation table to show only those mu-tations emerging in the clicked clone.(b) Clicking a mutation displays its variant allele frequency across time,and highlights the clonal phylogeny branch in which it originated.The automated nature of TimeScape facilitates studies with large samplesizes. TimeScape is particularly useful for revealing trends in the clonal dy-namics of many patients with a common cancer subtype. As a specific exam-ple, Kridel and Chan et al. [6] describe the majority of transformed follicular18lymphomas as changing dramatically in their clonal composition upon histo-logical transformation, regardless of time difference and treatment applica-tion between the pre- and post-transformation samples. TimeScape clearlydisplays this finding by quickly plotting the 15 patients in their study, re-vealing that post-transformation dominant clones originated as minor clonesprior to transformation (Fig. 3.3).Clonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyFigure 3.3: TimeScape visualizations of transformed follicular lymphoma pa-tients from Kridel and Chan et al. [6]. From time point 1 to time point2, there is a clear shift in clonal composition in fourteen out of fif-teen patients, where minor clones at time point 1 come to dominateat time point 2. For an interactive version of these TimeScapes, seeSupplemental link C.12.The automated nature of TimeScape is also valuable for visualizing a sin-gle tumour with many time points, as is common to xenograft studies whereserial transplantations provide longitudinal tracking of clonal dynamics in asingle cancer. Figure 4.5 showcases a TimeScape of the xenograft single celldata published in Eirew et al. [5] which includes three time points, althoughplotting longer xenograft lineages is equally effortless (e.g. Fig. 3.4).19Clonal PrevalenceT0 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10Time PointClonalPhylogenyFigure 3.4: A TimeScape visualization of a simulated dataset with 10 timepoints.3.2 Colouring the clonal phylogenyColour can be a powerful tool for visually grouping similar elements, sep-arating dissimilar elements, and indicating sequential progression from oneelement to another. In CellScape, we use hue to visually distinguish clonallineages, and we use brightness to indicate the accumulation of mutationalcontent along a lineage.marker cloneclonal phylogeny with n marker clones colour wheel split by n coloured clonal phylogenyb cadarker version of ancestorFigure 3.5: The method used to colour the clonal phylogeny.(a) The clonal phylogeny showing n marker clones, where a markerclone is a clone that marks the start of a lineage.(b) n colours from an evenly-split colour wheel.(c) The coloured phylogeny, where marker clones are assigned—indepth-first search order—a hue from the split colour wheel. All otherclones obtain a darker version of their ancestor.The algorithm, summarized by Figure 3.5, proceeds by first finding themarker clones, where a marker clone is a clone that marks the start of a20lineage. The marker clones are assigned a hue from a split colour wheel, andtheir assignment proceeds in depth-first search order—a specific algorithmfor searching a tree structure. This order ensures that genomically similarlineages are similar in colour, and the converse. All non-marker clones ob-tain a darker version of their ancestor. The choice to darken rather thanlighten a descendant clone reflects the accumulation, rather than the loss, ofmutational content as lineages evolve. The user may override these colourdefaults with custom colours.3.3 Representing small clonal populationsIn the study of clonal evolution, it is important to discriminate between thelow-prevalence and absence of a clone.bClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyaClonal PrevalenceT0 T1 T2Time PointClonalPhylogenyFigure 3.6: Low-prevalence clones are expanded to a minimum width, ensur-ing their visibility and enabling the discrimination of clonal minorityand absence.(b) The width of each clone follows directly from its prevalence. In thiscase, it is difficult to visualize clonal emergence of the green lineage,which contains low-prevalence clones.(a) Low-prevalence clones are expanded to a minimum width. Theemergence of all clones in the green lineage is now noticeable.In TimeScape, if the width of each clone follows directly from its preva-lence, it is difficult to discriminate low-prevalence from absence, as shown in3.6a (green lineage). We therefore expand low-prevalence clones (less than0.5% prevalence) to 0.5% of the view height to ensure their visibility, asshown in Figure 3.6b (green lineage). To compensate, the remaining clones21experience a proportionate reduction in height.3.4 X-axis spacing of emergent clonesPrior to each time point in TimeScape, certain clones may emerge. Theprocedure for spacing these emergent clones along the x-axis is summarizedby Figure 3.7. The height of the longest emergent lineage is used to evenlyspace the emergent clones in the previous window of time. If a perturbationevent is specified between two time points, the spacing of emergent cloneswill be achieved after the event but before the next time point.22Clonal PrevalenceChemotherapyT0 Diagnosis RelapseTime PointClonalPhylogenyClonal PrevalenceChemotherapyT0 Diagnosis RelapseTime PointClonalPhylogenyemerging clonelongest emerging lineage3 clones in longest emerging lineage1 clone in longest emerging lineageUntil DiagnosisDiagnosis to RelapseFigure 3.7: The x-axis spacing of emergent clones, shown on a TimeScape vi-sualization of AML patient 933124 from Ding et al. [3]. From time zero(“T0”) until “Diagnosis”, the longest emergent lineage is 3 clones long.We horizontally split this window of time into n + 1 segments, suchthat each of the three clones can emerge equidistantly. Likewise, from“Diagnosis” until “Relapse”, a single clone emerges, and thus, we splitthe window of time into 2 segments—however, there is a “Chemother-apy” perturbation event half-way between “Diagnosis” and “Relapse”,and thus we achieve the spacing after the “Chemotherapy” event.233.5 Vertical genotype layoutsThree genotype layouts—stacked, centred and spaced—are available as auser parameter. Figure 3.8 directly compares the three layouts. The defaultgenotype layout in TimeScape is the stacked layout, which stacks genotypesone on top of the another. The spaced layout stacks the genotypes whileensuring a minimum amount of spacing such that each genotype is verticallysurrounded by its ancestor. The centered layout centers genotypes withrespect to their ancestors.24Clonal PrevalenceChemotherapyT0 Diagnosis RelapseTime PointClonalPhylogenyStacked layoutClonal PrevalenceChemotherapyT0 Diagnosis RelapseTime PointClonalPhylogenyCentered layoutClonal PrevalenceChemotherapyT0 Diagnosis RelapseTime PointClonalPhylogenySpaced layoutFigure 3.8: The three vertical genotype layouts in TimeScape. The samedata (AML Patient 933124 from Ding et al. [3]) is used for eachTimeScape. “Stacked” is the default layout—it stacks genotypes oneon top of another to clearly display genotype prevalences at each timepoint. The “spaced” layout uses the same stacking method while main-taining a minimum amount of space between each genotype. The “cen-tered” layout centers genotypes with respect to their ancestors. Forinteractive views of each TimeScape, see Supplement links C.8, C.10and C.9.The layout choice is a matter of user preference, and depends on thepurpose of the visualization. The stacked layout is most useful for analysisof clonal dynamics, as it most clearly displays clonal prevalences. The spaced25layout maintains most of the clonal prevalence clarity while also accentuatingthe nested nature of clones, and is therefore useful for publication figures.The centered layout is the most qualitative of the three, fully highlightingthe nested nature of clones, and is useful for conceptually communicatingclonal evolution to a lay audience.3.6 Perturbation eventsTo incorporate a perturbation event (Fig. 3.1a “Chemotherapy” time point)into TimeScape, the user specifies the event name, the subsequent timepoint, and the remaining fraction of total tumour content. During the event,the plot height will shrink to the specified fraction.26Chapter 4Visualizing clonal evolutionat single cell resolution4.1 An overview of CellScapeSingle cell data is an emerging and powerful data type for studying clonalevolution. Rather than inferring clonal genotypes and prevalences from bulktissue data, single cell data enables us to directly observe these measure-ments. However, single cell data is often noisy, with missing and low-qualitycontent. Proper visualization can help to discriminate signal from noise, andcan also reveal sources of noise in the upstream experimental and compu-tational workflows. Signal and noise are most easily discriminated in largedatasets, and fortunately, single cell datasets are voluminous in nature. Vo-luminous data comes with the challenge of scalability—a visualization toolmust be able to efficiently process and display these large datasets. More-over, the display of single cell data must capture the evolutionary progressionof cells and the accumulation of mutational content. This requires sophisti-cation beyond hierarchical mutation clustering and dendrograms, neither ofwhich accurately capture the evolutionary nature of the data. Here we de-scribe CellScape, a visualization tool for efficiently and intuitively presentingsingle cell clonal evolution data.271 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 1819202122 X YCNV>=620Clone123HeatmapSingle Cell PhylogenyCELLSCAPE SVGPNGFigure 4.1: A CellScape visualization of triple-negative breast cancer sin-gle cell data from Wang et al. [4]. Supplemental link C.11 shows aninteractive version of this view.To our knowledge, CellScape is the first automated tool linking a singlecell phylogeny with a genomic heatmap. The utility of this design choice isapparent in Figure 4.1, a visualization of triple-negative breast cancer datafrom Wang et al. [4]. On its own, a phylogeny can indicate the hierarchicalrelationship of cells, but cannot convey the genomic underpinning of itsstructure. Analogously, a heatmap alone can relay the genomic diversity ina sample, but cannot express the evolutionary progression from one genometo another. Although heatmaps are often accompanied by a dendrogram,the distinction between a dendrogram and a phylogeny is important: while aphylogeny can account for extant ancestral cells in the sampled population,a dendrogram incorrectly assumes all cells to be leaf nodes. By integratingthe heatmap and phylogeny in an interactive interface, CellScape facilitatesthe navigation and analysis of single cell evolution data.As shown in Figure 4.2, a suite of phylogenetic views offers varying per-spectives on the data. One standard phylogenetic view, the force-directedgraph (e.g. Fig. 4.2c), displays an unrooted phylogeny. Alternatively, rec-tilinear trees (e.g. Fig. 4.2a) provide a direct correlation between nodes28and genomic profiles through vertical alignment (e.g. Fig. 4.1). The phy-logenies may incorporate edge distance information to emphasize genomicheterogeneity (Figs. 4.2b and 4.2d), some of which may be the result oflow-quality or missing values.a b cdFigure 4.2: The phylogenetic views available within CellScape. Data fromxenograft SA501 published in Eirew et al. [5].(a) The traditional phylogeny view, where each single cell is verticallyaligned with its genomic counterpart in the heatmap (not shown).(b) The traditional phylogeny with edge distance information empha-sizes genomic heterogeneity.(c) The phylogeny as a force-directed graph, for an unrooted phylogenyperspective preferred by some analysts.(d) The force-directed graph phylogeny with edge distances highlightsthe intraclonal diversity.The phylogenetic nodes may be interactively related to their genomicprofiles using the selection tool (Fig. 4.3). To use the tool, the user mustclick the tool in the tool bar, then select a section of the heatmap. Thisfeature is especially useful in two situations: (i) when nodes are not verticallyaligned to their genomic profiles, as in the force-directed graph view, and(ii) when the eye requires guidance to vertically discriminate one single cellfrom another within large datasets.29VAF10.50CloneABCDETimepointX1X2X4HeatmapSingle Cell Phylogeny Single Cell Phylogenya b cFigure 4.3: The selection tool at work on the CellScape visualization ofxenograft SA501 published in Eirew et al. [5]. This feature is partic-ularly useful for correlating genomic profiles with their correspondingnodes in the force-directed graph.(a) The selection tool.(b) The CellScape as the user selects the genomic profiles of interest.(c) The CellScape after selection takes place.As mentioned previously, single cell data often contains low-quality ormissing values. Branches plagued with these technical artifacts may beinteractively removed using the tree trimming tool (Fig. 4.4). After selectingthe tool in the tool bar, clicking a branch to prune will remove the nodesand heatmap rows associated with all downstream single cells.30b ca VAF10.50CloneABCDETimepointX1X2X4HeatmapSingle Cell Phylogeny Single Cell PhylogenyFigure 4.4: The tree trimming tool at work on the CellScape visualization ofxenograft SA501 published in Eirew et al. [5]. This feature is especiallyuseful for removing branches with low-quality genomic profiles.(a) The trimming tool.(b) The CellScape as the user hovers over the branch to trim.(c) The CellScape after trimming.CellScape can accommodate both copy number data, as seen in Figure4.1, and targeted mutation data, as demonstrated in Figure 4.5 which showsa CellScape visualization of single cell targeted mutation data publishedin Eirew et al. [5]. The fundamental visual distinction between the twoheatmaps lies in the x-axis—copy number data necessitates a fixed axis,while the order of targeted mutations is arbitrary. We therefore developeda custom hierarchical clustering method (see Section 4.3; Figs. 4.7, 4.8, 4.9)that highlights genomic evolution by sorting mutations to the left or rightof the heatmap based on their ancestral or descendant nature, respectively.aA powerful experimental design to observe clonal dynamics is to samplecell populations over time and compare their relative composition. For ex-ample, serial propagation of patient derived xenograft material facilitatessuch temporal sampling. We designed CellScape to accommodate tem-poral sampling information by automatically generating and appending aTimeScape (see Chapter 3) to the view (e.g. Fig. 4.5). The TimeScapeprovides a holistic context of temporal clonal dynamics while the CellScapepresents a granular view of clonal and temporal constituents. In Figure 4.5,31for instance, the TimeScape facilitates an appreciation of lineage successionand competition: the green and blue lineages compete for dominance untilthe blue lineage displays a selective sweep. The mutational contributions tothis selective sweep are interactively accessible. In Figure 4.5c, the user dis-covers that Clone E, responsible for the selective sweep, contains the largestaccumulation of sequenced mutations; to explore the novel mutational reper-toire of this clone, the user inspects the heatmap (Fig. 4.6). Although CloneE dominates xenograft X4, the preceding time points have a wider clonalspread—in Figure 4.5b, the user inspects xenograft X1 to uncover the widephylogenetic and genomic nature of its comprising cells.32VAF10.50CloneABCDETimepointX1X2X4HeatmapSingle Cell PhylogenyClonal PrevalenceT0 X1 X2 X4Time PointClonalPhylogenyCELLSCAPE SVGPNGTIMESCAPE SVGPNGa bcSingle Cell PhylogenyClonal PrevalenceT0 X1CELLSCAPETIMESCAPE0.96Clonal PrevalenceT0 X1 X2 X4Time PointClonalPhylogenyTIMESCAPE SVGPNGFigure 4.5: A time-series CellScape visualizes the xenograft SA501 singlecell data published in Eirew et al. [5]. For the interactive view, seeSupplemental link C.7.(a) The full time-series CellScape view. The bottom panel displays aTimeScape automatically generated from the single cell data.(b) Mouseover of time point “X1” elicits a response in “X1”-associatednodes and heatmap rows, emphasizing the diversity of cells at this timepoint.(c) Mouseover of a clone causes reactive highlighting in the TimeScape,single cell phylogeny, and heatmap.33Figure 4.6: Inspection of the CellScape heatmap shows the Variant AlleleFrequency (VAF) for the corresponding mutation site and single cellID. The CellScape visualization shows xenograft SA501 single cell datapublished in Eirew et al. [5].CellScape can efficiently process and visualize hundreds of cells, andis therefore suitable for most currently available datasets. For instance,visualizing our largest copy number dataset of 371 single cells and 37,580copy number segments takes 5.433 seconds of CPU time on a 2.8 GHz IntelCore i7 Macintosh computer, and less than 5 seconds to load on a Chromebrowser (version 52.0.2743). The SA501 xenograft data from Eirew et al. [5]comprises 106 single cells each with 55 mutations, and takes 0.373 seconds ofCPU time on the same machine, and less than one second to load on the sameChrome browser. Thus, CellScape does not require external machinery, andany biomedical researcher can operate the tool on their personal computer.4.2 Vertically ordering the single cells in the phy-logeny and heatmapIn CellScape, the vertical plane is divided equally such that each observedcell occupies a unique vertical position. The vertical ordering of cells pro-ceeds according to the phylogeny. Sibling cells are vertically ordered based34on the size of their descendant lineage, and ancestors are centered verticallywith respect to their descendants. Some of these ancestors are indicatedby an empty circle—these are latent nodes, or inferred nodes that are notsampled. Any leaf nodes or peripheral lineages void of genomic informationare removed from the phylogeny—in this situation, console warnings willindicate node removal.4.3 Targeted mutation orderingThe order of targeted mutations is arbitrary, and can be used as a powerfulmeans of communicating the accumulation of mutations across the evolu-tionary course of cancer. The order is computed by Algorithm 1, whichextracts the mutation order from a hierarchically clustered matrix of VAFsfor each mutation site and single cell. Prior to clustering (performed by the“hclust” function in the “stats” R package, version 3.3.1), VAFs less than0.05 are temporarily defined as -10, forcibly separating mutation sites withfew and many variants upon clustering. Given the resultant mutation order,if the mean mutation VAF is increasing rather than decreasing, the order isreversed, ensuring that ancestral mutations (high mean VAF) are on the leftof the matrix, and newly acquired mutations (low mean VAF) are on theright of the matrix. Although this algorithm is a target for future improve-ments, in all tested cases it performs better than a traditional clusteringmethod (Figs. 4.7, 4.8 and 4.9). The user may override the computed orderwith a custom mutation order.35VAF10.50Heatmap VAF10.50CloneABCDETimepointX1X2X4Heatmapa bTraditional hierarchical clustering method Custom hierarchical clustering methodFigure 4.7: A comparison of tradition mutation clustering and custom clus-tering described by Algorithm 1. The data comes from xenograft SA501published in Eirew et al. [5].(a) Using a traditional mutation clustering algorithm, the order of mu-tations in the heatmap is unintuitive.(b) Using the custom clustering method, the mutation order progressesfrom ancestral to novel.36VAF10.50HeatmapOm1Traditional method Custom methodOm2VAF10.50HeatmapVAF10.50HeatmapROv1VAF10.50Heatmap VAF10.50HeatmapROv2VAF10.50HeatmapVAF10.50HeatmapVAF10.50HeatmapFigure 4.8: A comparison of tradition mutation clustering and custom clus-tering described by Algorithm 1. The data comes from ovarian cancerpatient 2, McPherson and Roth et al. [2]. Abbreviations: Om = Omen-tum; ROv = Right Ovary.37Traditional method Custom methodLOv1VAF10.50HeatmapLOv2VAF10.50HeatmapVAF10.50HeatmapOm1VAF10.50HeatmapOm2ROv1VAF10.50HeatmapVAF10.50HeatmapVAF10.50HeatmapVAF10.50HeatmapVAF10.50HeatmapVAF10.50HeatmapFigure 4.9: A comparison of tradition mutation clustering and custom clus-tering described by Algorithm 1. The data comes from ovarian cancerpatient 9, McPherson and Roth et al. [2]. Abbreviations: LOv = LeftOvary; Om = Omentum; ROv = Right Ovary.38Algorithm 1 Order mutations1: procedure getMutationOrder(T , A)2: o← hclust(T )$order . obtain mutation order from hierarchicallyclustered mutation VAFs3: Ao ← A ordered by o4: s← slope of Ao5: if s > 0 then . ensure mutations proceed from high to low meanVAF6: reverse(o)7: end if8: return o9: end procedurewheren: the number of mutations in the heatmapc: the number of single cells in the heatmapM = {m1,m2,m3, ...,mn}: the unsorted set of n mutationsA = {a1, a2, a3, ..., an}: the average VAF for each of the n mutationsV = (vij): an n by c matrix of VAFsT = (tij): an n by c matrix of modified VAFs, wheretij =−10 if vij < 0.050.5 if vij ≥ 0.05 & vij < 0.951 if vij ≥ 0.95(4.1)39Chapter 5Visualization of spatial clonalevolution data5.1 An overview of MapScapeThe tumours composing a metastatic cancer are clonally related to one an-other, but often present significant spatial heterogeneity, both within andbetween tumours. The data describing this spatial clonal evolution is oftenmultifaceted, consisting of clonal prevalences, a clonal phylogeny, mutationalcontent and an anatomical image or diagram. To appreciate the clonal dy-namics occurring in a single patient, the analyst must associate one datasource with another. This necessitates a visualization design that integratesthese data sources and presents them to the analyst in an intuitive manner.We built MapScape (Fig. 5.1) as an integrative and intuitive visualizationtool for the analysis of spatial clonal evolution data.The main view displays the clonal composition of tumours at a varietyof anatomic locations in the patient, and is radially divided to display eachtumour sample separately. Two views of each sample form a comprehen-sive understanding of its clonal composition. Within each view, clones aredistinguished by colour, as described in Section 3.2. The first view, a cel-lular aggregate inspired by Eirew et al. [5], visually displays the prevalenceof each clone. The second view shows a skeleton of the patient's clonal40MAPSCAPE SVGPNGClonalPhylogenyAnatomyPhylogeneticClassificationpuremonophyleticpolyphyleticHFJDAICEGSearch:Showing 83 entriesClone Clone ID Chrom Coord Ref Alt Validation Status2 1 172571221 C T not-validated2 17 2275690 G A validated2 11 75599940 A G validated4 1 29639195 C A not-validatedd efga bcClonalPhylogenyAnatomyOm1SBwlSBwlE4HFJDAICEG0.460.470.560.270.04Search:Showing 83 entriesClone Clone ID Chrom Coord Ref Alt5 12 57556206 G A1 20 32267636 C T5 5 434261 G A1 11 63398851 C APhylogeneticClassificationpuremonophyleticpolyphyleticLFTB4LOvB2ClonalPhylogenyAnatomyOm1SBwlSBwlE40.970.120.65HIMAPSCAPE SVGPNGClonalPhylogenyAnatomyPhylogeneticClassificationpuremonophyleticpolyphyleticLFTB4LOvB2ApC1ROvA4ROv4ROv3ROv2ROv1RFTA16 Om1SBwlSBwlE4Search:Showing 194 entriesClone Clone ID Chrom Coord Gene Name Effect Impact Nuc ChangeA 7 100658929 MUC12NON_SYNONYMOUS_CODINGMODERATE G>CC 15 101599691 LRRK1 UPSTREAM MODIFIER T>CB 11 102856321 INTERGENIC MODIFIER G>TFigure 5.1: (Caption on following page.)phylogeny while highlighting only those clones present in the sample. Allsamples are centrally connected to their anatomic sources on a user-providedimage, which may be a drawing (as in Fig. 5.1d) or medical image (as in41Figure 5.1: An overview of MapScape.(a) A MapScape visualization of metastatic prostate cancer PatientA21 published in Gundem et al. [1]. For an interactive version of thisview, see Supplemental link C.5.(b) For sample reordering purposes, the user may click and drag a sam-ple. This feature may be used to emphasize effects such as metastaticprogression between samples.(c) Clicking a mutation displays its sample-specific variant allele fre-quencies at each participating tumour sample.(d) A MapScape visualization of metastatic ovarian cancer Patient 1published in McPherson and Roth et al. [2]. For an interactive versionof this view, see Supplemental link C.1.(e) Mouseover of a phylogenetic classification label highlights all tu-mour samples following this classification.(f) Mouseover of a clone emphasizes its presence throughout the viewand displays its prevalence in each sample.(g) Mouseover of a phylogenetic branch highlights the descendant clonesthroughout the view.Fig. 5.1a). In many patients, the anatomic region of interest is a small areawithin a more comprehensive medical image—for this reason, we auto-cropand enlarge the region of interest in the main view, while preserving itscontext in the legend (see Section 5.5).MapScape can serve as a visualization platform for any spatially variantdata consisting of phylogenetically related populations. However, its pri-mary use is for visualizing the evolution of metastatic cancer—an objectivepreviously attempted by Gundem et al. [1]. In their study, they displaymetastatic prostate cancer data using multiple disconnected views. Usingdata from Patient A21 as input to MapScape, we illustrate the advantageof integrating these views in an interactive visualization.The ability of MapScape to simultaneously display within- and between-tumour heterogeneity enables the user to view the polyclonal seeding of thispatient, as seen in Figure 5.1a—sample H likely received clones from samplesI (green clone), D (blue clone) and J (yellow clone).By clicking on mutations in the mutation table, the user can view theirVariant Allele Frequencies in each sample, and all descendant clones in whichthey are present. Figure 5.1c shows the utility of this feature for the discov-42ery of metastasis-to-metastasis spread—metastases D and CEGH are morerelated than either is to the primary tumour (sample F), as they share sim-ilar mutational and clonal content that is absent in the primary tumour.Clicking and dragging samples will reorder them, a feature of MapScapethat can clarify certain events and progressions such as the metastatic spreadfrom one site to another. Figure 5.1b shows the user reordering samples tofollow the metastasis-to-metastasis progression.In McPherson and Roth et al. [2], MapScape was employed to commu-nicate the complex heterogeneous clonal composition of metastatic ovariancancer and the two potential modes of intraperitoneal metastatic spread.In Patient 1, monoclonal seeding from the ovary to the peritoneum may bediscovered through additional interactive features of MapScape.The interactive exploration of tumour phylogenetic classes facilitatesthe hypothesis generation surrounding metastatic progression. Polyphyleticsites, or sites with a branched clonal phylogeny, are sites of high diversityand may offer potential for metastasis. For instance, in Patient 1, the leftovary (Fig. 5.1e) is potentially the primary tumour since it contains all nec-essary ancestral clones for metastatic progression: the grey clone seeds theright fallopian tubule, the yellow clone seeds the right ovary, and the greenclone seeds the remaining metastases.Mouseover of a phylogenetic clone will highlight the clone's presencethroughout the view, and can immediately guide the eye to the seeded metas-tases. For instance, mouseover of the green clone (Fig. 5.1f) displays theproportionate presence of this seeded ancestor in each metastasis at the timeof sampling.Once seeded, a metastasis follows its own evolutionary progression. Mouseoverof a phylogenetic branch will highlight its descendant clones throughout theview, enabling the user to track a single evolutionary lineage. For instance,mouseover of the phylogenetic branch preceding the seeded green clone dis-plays its descendant lineage (Fig. 5.1g).435.2 The power of multi-clone selection to highlightclonal evolution in a single patientClonalPhylogenyAdnxLFTC1LOvC5ROv1ROv2ROvA7RFTA2Om1OmF2CDSB1ClnE1Early development, prevalent clonesLate development, migrated clonesLate development, private clonesabcClonalPhylogenyAdnxLFTC1LOvC5ROv1ROv2ROvA7RFTA2Om1OmF2CDSB1ClnE1ClonalPhylogenyAdnxLFTC1LOvC5ROv1ROv2ROvA7RFTA2Om1OmF2CDSB1ClnE1Figure 5.2: (Caption on following page.)44Figure 5.2: Three multi-clone selections follow the progression of evolutionin Patient 3 published in McPherson and Roth et al. [2]. For the inter-active view, see Supplemental link C.2.(a) Early development clones that are prevalent in all samples of thepatient.(b) Late development clones that have migrated to one or more sitesin the patient.(c) Late development clones that are private to individual sites in thepatient.Within a MapScape visualization, clicking multiple clones in the clonal phy-logeny will highlight their presence throughout the view. This multi-cloneselection feature empowers the user to visually group clones with similarproperties. Figure 5.2, for instance, shows the use of this feature to map theevolutionary progression of Patient 3 from McPherson and Roth et al. [2].A subset of clones are present at all eight sampled locations—these clonesdeveloped early and possessed high metastatic potential (Fig. 5.2a). Laterin development, descendant clones emerged and migrated to one or moresites in the patient (Fig. 5.2b). The latest clones in development are privateto individual anatomical locations, their confinement due to a lack of eithermetastatic potential or time (Fig. 5.2c).5.3 Concurrent visualization of temporal and spa-tial clonal evolutionWhen clonal evolution data is present in both spatial and temporal dimen-sions, MapScape and TimeScape can compliment one another during anal-ysis. MapScape, on one hand, can reveal metastatic progression and spatialheterogeneity, but cannot capture the temporal variation of the data, whileTimeScape can expose the global temporal dynamics as evolution progresses,but cannot convey the physical locations of each clone. Together, these twoviews can present both dimensions, enabling a more comprehensive analysisof the patient.45Figure 5.3: (Caption on following page.)46Figure 5.3: Both temporal and spatial visualizations of metastatic ovariancancer Patient 7 from McPherson and Roth et al. [2]. Together, theseviews highlight two different avenues of metastatic progression in thepatient: intraperitoneal spread and hematogenous dissemination.(a) A MapScape visualization displays the samples taken from a varietyof metastases over three time points. The view reveals the metastaticpotential of the right uterus tumour, as it colonizes the brain viahematogenous dissemination, and also colonizes the bowel and rightpelvic mass via intraperitoneal spread. Interactive MapScape availableat Supplemental link C.3.(b) A TimeScape visualization shows the clonal composition at eachsampling time point, highlighting the dramatic global temporal dy-namics in this patient. At “intraperitoneal diagnosis”, a diversity ofclones inhabit the peritoneum. After surgical interventions, the pinkand yellow clones survive; at “intraperitoneal relapse” these clones seedand come to dominate the right pelvic mass and bowel implant. Likelyprior to surgical interventions, hematogenous dissemination of the greenclone from the right uterus results in a clonally-pure “Brain metasta-sis”. Interactive TimeScape available at Supplemental link C.4.As a specific example, Patient 7 from McPherson and Roth et al. [2]provides spatial sampling data from three different time points. Figure5.3 displays the data as both a MapScape and a TimeScape. When com-bined, the views highlight two avenues for metastatic progression within thepatient: intraperitoneal spread and hematogenous dissemination. The in-traperitoneal space at diagnosis contains a diversity of clones (Fig. 5.3a,right ovary, left ovary and right uterus; Fig. 5.3b, time point “Intraperi-toneal diagnosis”). One of these clones colonized the brain via hematoge-nous dissemination likely prior to intraperitoneal surgery (Fig. 5.3a, site“RUtD3” and brain sites; Fig. 5.3b, time point “Brain metastasis”). Af-ter surgical interventions, a few clones remain in the peritoneum, and viaintraperitoneal spread they colonize the right pelvic mass and the bowelimplant (Fig. 5.3a, right pelvic mass and bowel implant; Fig. 5.3b, timepoint “Intraperitoneal relapse”). While these samples represent a fractionof the patient's total tumour content (and perhaps tumour sites), the vi-sualizations nonetheless offer insights into the spatial and temporal clonaldynamics occurring in the patient.475.4 Automatic radial layout of samplesThe radial layout of samples can help or hinder data interpretation. Theautomated layout in MapScape intelligently places samples for the simplestand most intuitive view by ensuring that samples from the same anatomicalsite remain adjacent to one another, and that each sample is maximallyproximal to its anatomical location. This intelligent placement occurs bythe method described in Figure 5.4, which orders samples based on therelationship between the sample's anatomic location and the centre of themain view—the line segment between these two points forms an angle withthe x-axis, and this angle is used to order the samples in a clockwise mannerstarting at the positive x-axis.θxθsample locationview centreangle used for sample layoutFigure 5.4: To automatically calculate the sample layout in MapScape, aline segment is formed between the MapScape view centre (grey circle)and each sample location (red circle). The angle formed between thisline segment and the x-axis is used to order samples clockwise from thepositive x-axis.5.5 Automatic image croppingThe anatomical image or diagram of each patient may be large in compari-son with the cancerous area, and for this reason, we automatically crop andenlarge the region of interest in the main view. Figure 5.5 summarizes thealgorithm for automatically cropping the user-provided anatomical image.We first use the x- and y-axis sample boundaries to calculate the samplecentre. Then, we find the largest radius from the sample centre to each48sample. To ensure a minimum space of 15 pixels between the edge of thecropped image and the sample locations, we extend the radius by a corre-sponding amount. Finally, using the extended radius, we circularly crop theimage.1) 2)3)Algorithm for automatic image croppingtumour samplesample centresample x- and y-axis boundarieslargest radius from sample centre to sampleradius extension (15px when enlarged in main view)Original image4)Figure 5.5: The algorithm for automatically cropping anatomical images. 1)We calculate the centre of the tumour samples using the x- and y-axissample boundaries. 2) We find the largest radius from the sample centreto each tumour sample, and extend this radius by an amount equivalentto 15 pixels when the image is later enlarged in the main MapScapeview. 3) This extended radius is used to draw the encompassing circle.4) The image is cropped.495.6 Handling minor clonesThe statistical processing of clonal evolution datasets often assigns extremelysmall but non-zero clonal prevalences when in reality the clones are absentat particular anatomical sites. To display these minor clones would unnec-essarily complicate the view. By default, therefore, clones with less than1% prevalence are not shown, but warnings will be issued in the console,viewable by opening the browser inspector.The user may wish to display minor clones by setting the“show low prev gtypes” parameter to “TRUE”. In this case, the clones,where minor, will appear as open circles in the site-specific phylogeny. Notethat a clone will only appear in the cellular aggregate if its prevalence ismore than 1/“n cells”, where “n cells” is a parameter specifying the totalnumber of cells to display in the aggregate.5.7 Distribution of clones within the cellular ag-gregateThe visualization of each sample contains a cellular aggregate representa-tion of the tumour. Cellular aggregates are created by Voronoi tessellation,a mathematical process that partitions a plane into polygons such that eachpolygon contains one coordinate from a predetermined set, and the spacewithin each polygon is closer to its containing coordinate than to any othercoordinate within the set. The polygons, or “cells”, are assigned clonalcolours in numbers proportionate to the clonal prevalences within the tu-mour. Algorithm 2 ensures that for a given patient, this colouration adheresto two principles: stability, where spatial regions are preserved (e.g. thegreen clone will always appear on the left-hand side of an aggregate), andcontiguity, where each clone has a maximal clustering coefficient. Each cloneis assigned a starting cell where their colouration will begin. These startingcells are evenly spaced along the aggregate's edge, and remain consistentacross aggregates to ensure stability. To ensure contiguity, the assignmentof colours to cells proceeds in a round-robin fashion, whereby each clone, in50turn, is assigned to the uncoloured cell nearest to its starting cell. Figure5.6 compares this algorithm to a previous iteration, and demonstrates thebenefits brought by maintaining the two principles of stability and contigu-ity, especially given a patient with polyclonal seeding and a complex clonalphylogeny.Algorithm 2 Colour cellular aggregate1: procedure colourCellularAggregate(Q, C, V , T )2: P ← pick |Q| equidistant points ordered along an encompassing circle3: S ← a set of |Q| nulls . to be filled with |Q| starter cells4: for i in 1 : |Q| do5: si ← the v closest to pi6: end for7: J ← a set of |Q| zeros . to keep track of the number of cellscurrently assigned to each clone8: while there are uncoloured cells do9: for i in 1 : |Q| do10: if ji < ti then11: use ci to colour the next uncoloured v closest to si12: ji ← ji + 113: end if14: end for15: end while16: end procedurewherem: the number of clones in the patientQ = {q1, q2, q3, ..., qm}: the set of m clones in the patient, sorted by depth-first search of the clonal phylogenyC = {c1, c2, c3, ..., cm}: the set of m clonal coloursV = {v1, v2, v3, ...}: the set of “cells”, or polygons, in the cellular aggregateT = {t1, t2, t3, ..., tm}: the total number of cells to assign to each of the mclones, given |V | and the prevalence of each clone in the sample51ClonalPhylogenyMethod A Method BFigure 5.6: The comparison of two methods for distributing clones withina cellular aggregate, data from Patient A22 published in Gundem etal. [1]. Method B, described by Algorithm 2, provides a cleaner viewof each sample by maintaining the principles of stability, where spatialregions are preserved, and contiguity, where each clone has a maxi-mal clustering coefficient. Method A is different from Method B inthe following ways. (i) Method A generated a different Voronoi tes-sellation structure for each sample. (ii) Method A randomly allocatedstarting cells for each genotype, and these starting cells are not consis-tent throughout the patient samples. (iii) Method A did not employ around-robin assignment of cells to genotypes, and instead assigned cellsto each clone successively, starting with the lowest-prevalence clone.For the interactive visualization, see Supplemental link C.6.52Chapter 6Conclusion6.1 Significance and contributionThe novel and complex nature of cancer evolution data necessitates the in-volvement of biomedical experts in the inference and analysis process. How-ever, the tractability of this voluminous and increasingly available data typeis limited in the absence of automated and interactive visualization tools.We developed the E-scale tool suite: the first interactive and automatedset of visualization tools for the exploration of clonal dynamics in cancer.E-scape has the potential to aid discoveries that will greatly improve ourunderstanding of cancer progression, metastasis, and treatment resistance.The novel contributions provided by E-scape are manifold. Beyond itscore ability to visually represent clonal evolution data, E-scape holds thecapacity to (i) visually link population-level clonal dynamics with a singlecell perspective, (ii) simultaneously visualize temporal and spatial clonal dy-namics of metastatic cancer patients, and (iii) improve clinical and publicaccessibility of clonal evolution data. We expect that with these contri-butions, E-scape will provide the impetus for novel insights into biologicalphenomena occurring within the cancer patient at the levels of academia,the clinic and general public.536.2 Limitations and future improvementsThe current iteration of E-scape is not without its limitations. Future im-provements include the incorporation of a backend for improved scalabilityas the magnitude of experimental output increases to tens of samples, thou-sands of mutations and millions of single cells. The integration of E-scapewith other plot types is another priority for facilitating alternative granularperspectives on the data. For example, an integrated whole genome browserwould provide a more detailed view of clonal and single cell genotypes, sharp-ening copy number breakpoints and revealing small copy number changes.The incorporation of data quality metrics would facilitate the separation ofsignal from noise. Visualization support for emerging data types, such assingle cell expression data and single cell in situ PCR data, will becomecritical as their availability becomes commonplace in the field.The E-scape tool suite represents, for the most part, the first of its kind.We therefore expect E-scape to be a springboard for further iterations onclonal evolution visualization design. As these visualization designs undergomany evolutions, the field will settle on the most powerful and intuitiverepresentations of clonal evolution. Such iterations could involve layoutchanges, representation alterations, additional annotation capabilities andthe incorporation of additional interactive features. These iterations coulddirectly build upon E-scape, as the code is open source.6.3 Potential applicationsWe anticipate many applications for E-scape in a variety of communities.Within the lab, E-scape is effective for data analysis and interpretation, asthoroughly explored in the body of this thesis. In the broader scientificcommunity, E-scape is valuable for communicating research findings, givenits capability to produce publication-ready figures that straightforwardlypresent complex research. In the clinic, E-scape visualizations may aid clin-ical decision-making in the future of personalized oncology. By visualizinga metastatic cancer, for instance, the metastatic progression and/or clonal54diversity at different sites may influence treatment decisions. Likewise, visu-alizing the dynamic clonal composition at diagnosis and relapse may explainthe presenting pathology and/or influence the future clinical course for thepatient. Beyond the genomic analysis of cancer patients, E-scape might finduse in other fields of “clonal” biology, such as metagenomics, epigenetics andimmunology.6.4 Final wordE-scape helps to facilitate a multidisciplinary analysis process whereby com-putational biologists, biomedical researchers and clinicians can effectivelyengage with genomic datasets through intelligent interactive and automatedvisualization. As more bright eyes interact with clonal evolution datasets,we approach the goal of understanding, predicting and halting cancer pro-gression.55Bibliography[1] Gunes Gundem, Peter Van Loo, Barbara Kremeyer, Ludmil BAlexandrov, Jose MC Tubio, Elli Papaemmanuil, Daniel S Brewer,Heini ML Kallio, Gunilla Ho¨gna¨s, Matti Annala, et al. Theevolutionary history of lethal metastatic prostate cancer. Nature,520(7547):353–357, 2015.[2] Andrew McPherson, Andrew Roth, Emma Laks, Tehmina Masud, AliBashashati, Allen W Zhang, Gavin Ha, Justina Biele, Damian Yap,Adrian Wan, et al. Divergent modes of clonal spread andintraperitoneal mixing in high-grade serous ovarian cancer. NatureGenetics, 2016.[3] Li Ding, Timothy J Ley, David E Larson, Christopher A Miller,Daniel C Koboldt, John S Welch, Julie K Ritchey, Margaret A Young,Tamara Lamprecht, Michael D McLellan, et al. Clonal evolution inrelapsed acute myeloid leukaemia revealed by whole-genomesequencing. Nature, 481(7382):506–510, 2012.[4] Yong Wang, Jill Waters, Marco L Leung, Anna Unruh, Whijae Roh,Xiuqing Shi, Ken Chen, Paul Scheet, Selina Vattathil, Han Liang,et al. Clonal evolution in breast cancer revealed by single nucleusgenome sequencing. Nature, 512(7513):155–160, 2014.[5] Peter Eirew, Adi Steif, Jaswinder Khattra, Gavin Ha, Damian Yap,Hossein Farahani, Karen Gelmon, Stephen Chia, Colin Mar, AdrianWan, et al. Dynamics of genomic clones in breast cancer patientxenografts at single-cell resolution. Nature, 518(7539):422–426, 2015.[6] Robert Kridel, Fong Chun Chan, Anja Mottok, Merrill Boyle, PedroFarinha, King Tan, Barbara Meissner, Ali Bashashati, AndrewMcPherson, Andrew Roth, Karey Shumansky, Damian Yap, SusanaBen-Neriah, Jamie Rosner, Maia A. Smith, Cidney Nielsen, Eva Gin,56Adele Telenius, Daisuke Ennishi, Andrew Mungall, Richard Moore,Ryan D. Morin, Nathalie A. Johnson, Laurie H. Sehn, ThomasTousseyn, Ahmet Dogan, Joseph M. Connors, David W. Scott,Christian Steidl, Marco A. Marra, Randy D. Gascoyne, and Sohrab P.Shah. Clonal dynamics shaping histological transformation andprogression in follicular lymphoma clinical histories. ManuscriptSubmitted.[7] Salem Malikic, Andrew W McPherson, Nilgun Donmez, and Cenk SSahinalp. Clonality inference in multiple tumor samples usingphylogeny. Bioinformatics, 31(9):1349–1356, 2015.[8] Andrew Roth, Jaswinder Khattra, Damian Yap, Adrian Wan, EmmaLaks, Justina Biele, Gavin Ha, Samuel Aparicio, AlexandreBouchard-Coˆte´, and Sohrab P Shah. Pyclone: statistical inference ofclonal population structure in cancer. Nature Methods, 11(4):396–398,2014.[9] Sohrab P Shah, Andrew Roth, Rodrigo Goya, Arusha Oloumi, GavinHa, Yongjun Zhao, Gulisa Turashvili, Jiarui Ding, Kane Tse,Gholamreza Haffari, et al. The clonal and mutational evolutionspectrum of primary triple-negative breast cancers. Nature,486(7403):395–399, 2012.[10] Serena Nik-Zainal, Peter Van Loo, David C Wedge, Ludmil BAlexandrov, Christopher D Greenman, King Wai Lau, Keiran Raine,David Jones, John Marshall, Manasa Ramakrishna, et al. The lifehistory of 21 breast cancers. Cell, 149(5):994–1007, 2012.[11] Mona Meyer, Ju¨ri Reimand, Xiaoyang Lan, Renee Head, XuemingZhu, Michelle Kushida, Jane Bayani, Jessica C Pressey, Anath CLionel, Ian D Clarke, et al. Single cell-derived clonal analysis of humanglioblastoma links functional and genomic heterogeneity. Proceedingsof the National Academy of Sciences, 112(3):851–856, 2015.[12] Nicola E Potter, Luca Ermini, Elli Papaemmanuil, GiovanniCazzaniga, Gowri Vijayaraghavan, Ian Titley, Anthony Ford, PeterCampbell, Lyndal Kearney, and Mel Greaves. Single-cell mutationalprofiling and clonal phylogeny in cancer. Genome Research,23(12):2115–2125, 2013.[13] C Bhang Hyo-eun, David A Ruddy, Viveksagar KrishnamurthyRadhakrishna, Justina X Caushi, Rui Zhao, Matthew M Hims,57Angad P Singh, Iris Kao, Daniel Rakiec, Pamela Shaw, et al.Studying clonal dynamics in response to cancer therapy usinghigh-complexity barcoding. Nature Medicine, 21(5):440–448, 2015.[14] L Melchor, A Brioli, CP Wardell, A Murison, NE Potter, MF Kaiser,RA Fryer, DC Johnson, DB Begum, S Hulkki Wilson, et al.Single-cell genetic analysis reveals the composition of initiating clonesand phylogenetic patterns of branching and parallel evolution inmyeloma. Leukemia, 28(8):1705–1715, 2014.[15] Nicholas Navin, Jude Kendall, Jennifer Troge, Peter Andrews, LindaRodgers, Jeanne McIndoo, Kerry Cook, Asya Stepansky, Dan Levy,Diane Esposito, et al. Tumour evolution inferred by single-cellsequencing. Nature, 472(7341):90–94, 2011.[16] Marco Gerlinger, Andrew J Rowan, Stuart Horswell, James Larkin,David Endesfelder, Eva Gronroos, Pierre Martinez, NicholasMatthews, Aengus Stewart, Patrick Tarpey, et al. Intratumorheterogeneity and branched evolution revealed by multiregionsequencing. New England Journal of Medicine, 366(10):883–892, 2012.[17] Elza C de Bruin, Nicholas McGranahan, Richard Mitter, Max Salm,David C Wedge, Lucy Yates, Mariam Jamal-Hanjani, Seema Shafi,Nirupa Murugaesu, Andrew J Rowan, et al. Spatial and temporaldiversity in genomic instability processes defines lung cancerevolution. Science, 346(6206):251–256, 2014.[18] Ali Bashashati, Gavin Ha, Alicia Tone, Jiarui Ding, Leah M Prentice,Andrew Roth, Jamie Rosner, Karey Shumansky, Steve Kalloger,Janine Senz, et al. Distinct evolutionary trajectories of primaryhigh-grade serous ovarian cancers revealed through spatial mutationalprofiling. The Journal of Pathology, 231(1):21–34, 2013.[19] Antonija Kreso, Catherine A O’Brien, Peter van Galen, Olga I Gan,Faiyaz Notta, Andrew MK Brown, Karen Ng, Jing Ma, ErnoWienholds, Cyrille Dunant, et al. Variable clonal repopulationdynamics influence chemotherapy response in colorectal cancer.Science, 339(6119):543–548, 2013.[20] Iain C Macaulay, Wilfried Haerty, Parveen Kumar, Yang I Li,Tim Xiaoming Hu, Mabel J Teng, Mubeen Goolam, Nathalie Saurat,Paul Coupland, Lesley M Shirley, et al. G&t-seq: parallel sequencing58of single-cell genomes and transcriptomes. Nature Methods,12(6):519–522, 2015.[21] Siddharth S Dey, Lennart Kester, Bastiaan Spanjaard, Magda Bienko,and Alexander van Oudenaarden. Integrated genome andtranscriptome sequencing of the same cell. Nature Biotechnology,33(3):285–289, 2015.[22] Christopher B Raub, Chen-Chung Lee, Darryl Shibata, Clive Taylor,and Emil Kartalov. Histomosaic detecting kras g12v mutation acrosscolorectal cancer tissue slices through in situ pcr. AnalyticalChemistry, 88(5):2792–2798, 2016.[23] Andrew Roth, Andrew McPherson, Emma Laks, Justina Biele,Damian Yap, Adrian Wan, Maia A Smith, Cydney B Nielsen,Jessica N McAlpine, Samuel Aparicio, et al. Clonal genotype andpopulation structure inference from single-cell tumor sequencing.Nature Methods, 2016.[24] Dan A Landau, Scott L Carter, Petar Stojanov, Aaron McKenna,Kristen Stevenson, Michael S Lawrence, Carrie Sougnez, ChipStewart, Andrey Sivachenko, Lili Wang, et al. Evolution and impactof subclonal mutations in chronic lymphocytic leukemia. Cell,152(4):714–726, 2013.[25] Kristina Anderson, Christoph Lutz, Frederik W Van Delft,Caroline M Bateman, Yanping Guo, Susan M Colman, HelenaKempski, Anthony V Moorman, Ian Titley, John Swansbury, et al.Genetic variegation of clonal architecture and propagating cells inleukaemia. Nature, 469(7330):356–361, 2011.[26] Shinichi Yachida, Siaˆn Jones, Ivana Bozic, Tibor Antal, RebeccaLeary, Baojin Fu, Mihoko Kamiyama, Ralph H Hruban, James REshleman, Martin A Nowak, et al. Distant metastasis occurs lateduring the genetic evolution of pancreatic cancer. Nature,467(7319):1114–1117, 2010.[27] Charles Gawad, Winston Koh, and Stephen R Quake. Dissecting theclonal origins of childhood acute lymphoblastic leukemia by single-cellgenomics. Proceedings of the National Academy of Sciences,111(50):17947–17952, 2014.59[28] Tyler Garvin, Robert Aboukhalil, Jude Kendall, Timour Baslan,Gurinder S Atwal, James Hicks, Michael Wigler, and Michael CSchatz. Interactive analysis and assessment of single-cell copy-numbervariations. Nature Methods, 12(11):1058–1060, 2015.[29] John M Findlay, Francesc Castro-Giner, Seiko Makino, Emily Rayner,Christiana Kartsonaki, William Cross, Michal Kovac, DannyUlahannan, Claire Palles, Richard S Gillies, et al. Differential clonalevolution in oesophageal cancers in response to neo-adjuvantchemotherapy. Nature Communications, 7, 2016.[30] Christopher A Miller, Joshua McMichael, Ha X Dang, Christopher AMaher, Li Ding, Timothy J Ley, Elaine R Mardis, and Richard KWilson. Visualizing tumor evolution with the fishplot package for r.bioRxiv, page 059055, 2016.[31] Ramnath Vaidyanathan, Kenton Russell, and Inc. RStudio.htmlwidgets for r, 2014.[32] R Core Team. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria,2015.[33] Michael Bostock, Vadim Ogievetsky, and Jeffrey Heer. D3 data-drivendocuments. IEEE Transactions on Visualization and ComputerGraphics, 17(12):2301–2309, December 2011.60Appendix AData used in figuresA.1 Learning the clonal prevalences for prostatecancer patient A21 (Gundem et al. [1])Phylogeny withmutation clustersMutation clustercancer cell fractions (CCF)CCFA: 1.00CCFB: 0.17CCFC: 0.35CCFD: 0.22CCFE: 0.01Clonal prevalence (CP) calculationsCP1 = CCFA - (CCFB + CCFC  + CCFD )CP2 = CCFB CP3 = CCFCCP4 = CCFD - CCFE CP5 = CCFE Phylogeny withclonal prevalences123450.260.170.350.210.01ABACADAADE12345a b c dFigure A.1: Calculating clonal prevalence given a clonal phylogeny and mu-tation cluster cancer cell fractions.(a) Cancer cell fractions (CCF) for each mutation cluster present inthe patient.(b) The patient clonal phylogeny annotated with mutation clusters.This representation emphasizes that the CCF of any mutation cluster(e.g. cluster D) encompasses the prevalence of all its descendant clones(e.g. clones 4 and 5).(c) The prevalence for each clone (e.g. clone 1) is calculated by sum-ming the CCFs of mutation clusters originating in the immediatelydescendant branches (e.g. clusters B, C and D), then subtracting thissum from the CCF of the mutation cluster originating in the immedi-ately ancestral branch (e.g. cluster A).(d) The patient clonal phylogeny annotated with clonal prevalenceswhich sum to 1.61In their paper, Gundem et al. group the clonal and subclonal mutationsby their cancer cell fractions (CCFs), the fraction of tumour cells carryinga specified mutation, in each sample. For each patient, CCFs from samplepairs are plotted against one another to determine the clonal phylogeny forthe patient. For each sample, we combine these two sources of information—the sample-specific CCFs and the patient clonal phylogeny—to infer clonalprevalences in a process summarized by Figure A.1. Similarly to their paper,we assume the infinite sites hypothesis, which postulates an infinite numberof mutation sites, although a mutation may only occur once at any givensite and cannot revert to its previous form. From this assumption followsthe additivity assumption, where mutations in an ancestral branch will bepresent throughout their descendant lineages while novel mutations accumu-late. Therefore, to obtain the prevalence of each clone, we sum the CCFs ofmutation clusters originating in the immediately descendant branches, thensubtract this sum from the CCF of the mutation cluster originating in theimmediately ancestral branch.A.2 Mutational information for prostate cancerpatient A21 (Gundem et al. [1])The mutation information was taken from their Supplementary Informationsection (Supplementary Data, tab “subs”).A.3 Time-series clonal prevalences for ovarian can-cer patient 7 (McPherson and Roth et al. [2])To calculate time-series clonal prevalences, we averaged the prevalence ofeach clone over all samples for each time point.62A.4 Acute myeloid leukemia (Ding et et al. [3])The clonal prevalences and hierarchy of patient UPN 933124 were bothreported in the main text of the paper, and the deep readcounts of somaticmutations were taken from Supplementary Table 5a in their paper.A.5 Learning the phylogenetic tree for triple-negativebreast cancer patient (Wang et al. [4])We learned the phylogenetic tree from the single cell copy number dataobtained directly from the authors. In traditional phylogenetic algorithms,e.g. neighbour-joining, the internal nodes and leaves of the tree are assignedto unobserved ancestors and observed genotypes, respectively. However,we used a novel algorithm that can assign observed genotypes to internalnodes. This is an important capability for single cell cancer data, whereboth descendants and ancestors can be observed in a sample.63Appendix BInput requirements andexamples for E-scape toolsTable B.1: Input requirements for CellScape. Available in the CellScapehelp menu.cnv data Data frame (Required if mut data not provided) Singlecell copy number segments data. Note that every sin-gle cell id must be present in the tree edges data frame.Format: columns are (1) String “single cell id” - singlecell id (2) String “chr” - chromosome number (3) Number“start” - start position (4) Number “end” - end position(5) Number “integer copy number” - copy number state.mut data Data frame (Required if cnv data not provided) Single celltargeted mutation data frame. Note that every single cellid must be present in the tree edges data frame. Format:columns are (1) String “single cell id” - single cell id (2)String “chr” - chromosome number (3) Number “coord”- genomic coordinate (5) Number “VAF” - variant allelefrequency [0, 1].64mut order Array (Optional) Mutation order for targeted mutationheatmap (each mutation should consist of a string in theform “chrom:coord”). Default will use a clustering func-tion to determine mutation order.tree edges Data frame Edges for the single cell phylogenetic tree.Format: columns are (1) String “source” - edge source(single cell id) (2) String “target” - edge target (singlecell id) (2) Number (Optional) “dist” - edge distancegtype tree edges DataFrame (Required for TimeScape) Genotype treeedges of a rooted tree. Format: columns are (1) String“source” - source node id (2) String “target” - target nodeid.sc annot Data frame (Required for TimeScape) Annotations (geno-type and sample id) for each single cell. Format: columnsare (1) String “single cell id” - single cell id (2) String“genotype” - genotype assignment (3) String (Optional)“timepoint” - id of the sampled time point. Note: in thecase of time points, they will be ordered alphabeticallyclone colours DataFrame (Optional) Clone ids and their correspond-ing colours Format: columns are (1) String “clone id” -the clone ids (2) String “colour” - the corresponding Hexcolour for each clone id.timepoint title String (Optional) Legend title for timepoint groups. De-fault is “Timepoint”.clone title String (Optional) Legend title for clones. Default is“Clone”.xaxis title String (Optional) For TimeScape - x-axis title. Default is“Time Point”.yaxis title String (Optional) For TimeScape - y-axis title. Default is“Clonal Prevalence”.phylogeny title String (Optional) For TimeScape - legend phylogeny title.Default is “Clonal Phylogeny”.65value type String (Optional) The type of value plotted in heatmap -will affect legend and heatmap tooltips. Default is “VAF”for mutation data, and “CNV” for copy number data.node type String (Optional) The type of node plotted in single cellphylogeny - will affect phylogeny tooltips. Default is“Cell”.display node ids Boolean (Optional) Whether or not to display the singlecell ID within the tree nodes. Default is FALSE.show warnings Boolean (Optional) Whether or not to show any warnings.Default is TRUE.width Number (Optional) Width of the plot.height Number (Optional) Height of the plot.66Table B.2: Input requirements for TimeScape. Available in the TimeScapehelp menu.clonal prev Data Frame Clonal prevalence. Note: timepoints will bealphanumerically sorted in the view. Format: columns are(1) String “timepoint” - time point (2) String “clone id”- clone id (3) Number “clonal prev” - clonal prevalence.tree edges Data Frame Tree edges of a rooted tree. Format: columnsare (1) String “source” - source node id (2) String “target”- target node id.mutations Data Frame (Optional) Mutations occurring at each clone.Any additional field will be shown in the mutation table.Format: columns are (1) String “chrom” - chromosomenumber (2) Number “coord” - coordinate of mutation onchromosome (3) String “clone id” - clone id (4) String“timepoint” - time point (5) Number “VAF” - variant al-lele frequency of the mutation in the corresponding time-point.clone colours Data Frame (Optional) Clone ids and their correspond-ing colours Format: columns are (1) String “clone id” -the clone ids (2) String “colour” - the corresponding Hexcolour for each clone id.xaxis title String (Optional) x-axis title. Default is “Time Point”.yaxis title String (Optional) y-axis title. Default is “Clonal Preva-lence”.phylogeny title String (Optional) Legend phylogeny title. Default is“Clonal Phylogeny”.alpha Number (Optional) Alpha value for sweeps, range [0, 100].67genotype position String (Optional) How to position the genotypes from[“centre”, “stack”, “space”] “centre” - genotypes are cen-tred with respect to their ancestors “stack” - genotypesare stacked such that no genotype is split at any timepoint “space” - genotypes are stacked but with a bit ofspacing at the bottom.perturbations Data Frame (Optional) Any perturbations that occurredbetween two time points, and the fraction of total tu-mour content remaining. Format: columns are (1) String“pert name” - the perturbation name (2) String “prev tp”- the time point (as labelled in clonal prevalence data) BE-FORE perturbation (3) Number “frac” - the fraction oftotal tumour content remaining at the time of perturba-tion, range [0, 1].sort Boolean (Optional) Whether (TRUE) or not (FALSE) tovertically sort the genotypes by their emergence values(descending). Default is FALSE. Note that genotype sort-ing will always retain the phylogenetic hierarchy, and thisparameter will only affect the ordering of siblings.show warnings Boolean (Optional) Whether or not to show any warnings.Default is TRUE.width Number (Optional) Width of the plot. Minimum width is450.height Number (Optional) Height of the plot. Minimum heightwith and without mutations is 500 and 260, respectively.68Table B.3: Input requirements for MapScape. Available in the MapScapehelp menu.clonal prev Data Frame Clonal prevalence. Format: columns are (1)String “sample id” - id for the tumour sample (2) String“clone id” - clone id (3) Number “clonal prev” - clonalprevalence.tree edges Data Frame Tree edges for a rooted tree. Format: columnsare (1) String “source” - source clone id (2) String “target”- target clone id.sample locations Data Frame Anatomic locations for each tumour sample.Format: columns are (1) String “sample id” - id for the tu-mour sample (2) String “location id” - name of anatomiclocation for this tumour sample (3) Number (Optional)“x” - x-coordinate (in pixels) for anatomic location onanatomic image (4) Number (Optional) “y” - y-coordinate(in pixels) for anatomic location on anatomic image.img ref String A reference for the custom anatomical image touse, *** in PNG format ***, either a URL to an imagehosted online or a path to the image in local file system.If unspecified, will use default generic male and femaleimages.clone colours Data Frame (Optional) Clone ids and their correspondingcolours (in hex format) Format: columns are (1) String“clone id” - the clone ids (2) String “colour” - the corre-sponding Hex colour for each clone id.69mutations Data Frame (Optional) Mutations occurring at each clone.Any additional field will be shown in the mutation ta-ble. Format: columns are (1) String “chrom” - chromo-some number (2) Number “coord” - coordinate of muta-tion on chromosome (3) String “clone id” - clone id (4)String “sample id” - id for the tumour sample (5) Num-ber “VAF” - variant allele frequency of the mutation inthe corresponding sample.sample ids Vector (Optional) Ids of the samples in the order yourwish to display them (clockwise from positive x-axis).n cells Number (Optional) The number of cells to plot (forvoronoi tessellation).phylogeny title String (Optional) Legend title for the phylogeny. Defaultis “Clonal Phylogeny”.anatomy title String (Optional) Legend title for the anatomy. Defaultis “Anatomy”.classification title String (Optional) Legend title for the phylogenetic classi-fication. Default is “Phylogenetic Classification”.show warnings Boolean (Optional) Whether or not to show any warnings.Default is TRUE.width Number (Optional) Width of the plot. Minimum width is930.height Number (Optional) Height of the plot. Minimum heightis 700.70Table B.4: Examples for E-scape tools. Available in the help menus.MapScape library(“mapscape”)# clonal prevalencesclonal prev ← read.csv(system.file(“extdata”, “example-Data clonal prev.csv”, package = “mapscape”))# mutationsmutations ← read.csv(system.file(“extdata”, “example-Data mutations.csv”, package = “mapscape”))# locations of each tumour sample on user-provided im-agesample locations ← read.csv(system.file(“extdata”, “exam-pleData sample locations.csv”, package = “mapscape”))# genotype tree edgestree edges ← read.csv(system.file(“extdata”, “example-Data tree.csv”, package = “mapscape”))# image referenceimg ref ← system.file(“extdata”, “example-Data anatomical image.png”, package = “mapscape”)# run mapscapemapscape(clonal prev = clonal prev, tree edges = tree edges,sample locations = sample locations, mutations = mutations,img ref = img ref)71TimeScape library(“timescape”)# clonal prevalencesclonal prev ← data.frame( timepoint = c(rep(“T1”,6), rep(“T2”, 6)), clone id = c(“1”, “6”, “5”, “4”,“3”, “2”, “1”, “6”, “5”, “4”, “3”, “2”), clonal prev= c(“0.0205127”, “0.284957”, “0.637239”, “0.0477972”,“0.00404099”, “0.00545235”, “0.0134362”, “0.00000150677”,“0.00000385311”, “0.000627522”, “0.551521”, “0.43441”))# genotype tree edgestree edges ← data.frame(source = c(“1”, “1”, “6”, “5”, “3”),target = c(“3”, “6”, “5”, “4”, “2”))# clone coloursclone colours ← data.frame( clone id = c(“1”, “2”, “3”, “4”,“5”, “6”), colour = c(“F8766D66”, “B79F0066”, “00BA3866”,“00BFC466”, “619CFF66”, “F564E366”))# perturbations (chemotherapy)perturbations ← data.frame( pert name = c(“Chemo”),prev tp = c(“T1”), frac = c(0.1))# run timescapetimescape(clonal prev = clonal prev, tree edges = tree edges,clone colours = clone colours, perturbations = perturbations,height=400)72CellScapee.g.1library(“cellscape”)# EXAMPLE 1 - TARGETED MUTATION DATA# single cell tree edgestree edges ← read.csv(system.file(“extdata”, “tar-geted tree edges.csv”, package = “cellscape”))# targeted mutationstargeted data ← read.csv(system.file(“extdata”, “tar-geted muts.csv”, package = “cellscape”))# genotype tree edgesgtype tree edges ← data.frame(“source”=c(“Ancestral”,“Ancestral”, “B”, “C”, “D”), “target”=c(“A”, “B”, “C”,“D”, “E”))# annotationssc annot ← read.csv(system.file(“extdata”, “tar-geted annots.csv”, package = “cellscape”))# mutation ordermut order ← scan(system.file(“extdata”, “tar-geted mut order.txt”, package = “cellscape”),what=character())# run cellscapecellscape(mut data=targeted data, tree edges=tree edges,sc annot = sc annot, gtype tree edges=gtype tree edges,mut order=mut order)73CellScapee.g.2library(“cellscape”)# EXAMPLE 2 - COPY NUMBER DATA# single cell tree edgestree edges ← read.csv(system.file(“extdata”,“cnv tree edges.csv”, package = “cellscape”))# cnv segments datacnv data ← read.csv(system.file(“extdata”, “cnv data.csv”,package = “cellscape”))# annotationssc annot ← read.csv(system.file(“extdata”, “cnv annots.tsv”,package = “cellscape”), sep=”’’)# custom clone coloursclone colours ← data.frame( clone id = c(“1”, “2”, “3”),colour = c(“7fc97f”, “beaed4”, “fdc086”))# run cellscapecellscape(cnv data=cnv data, tree edges=tree edges,sc annot=sc annot, width=800, height=475,show warnings=FALSE, clone colours=clone colours)74Appendix CSupplemental linksC.1 Interactive MapScape of metastatic ovariancancer patient 1 from McPherson and Rothet al. [2]http://mo bccrc.bitbucket.org/escape/ITHPatient1.html.C.2 Interactive MapScape of metastatic ovariancancer patient 3 from McPherson and Rothet al. [2]http://mo bccrc.bitbucket.org/escape/ITHPatient3.html.C.3 Interactive MapScape of metastatic ovariancancer patient 7 from McPherson and Rothet al. [2]http://mo bccrc.bitbucket.org/escape/ITHPatient7MapScape.html.75C.4 Interactive TimeScape of metastatic ovariancancer patient 7 from McPherson and Rothet al. [2]http://mo bccrc.bitbucket.org/escape/ITHPatient7TimeScape.html.C.5 Interactive MapScape of metastatic prostatecancer patient A21 from Gundem et al. [1]http://mo bccrc.bitbucket.org/escape/ProstatePatientA21.html.C.6 Interactive MapScape of metastatic prostatecancer patient A22 from Gundem et al. [1]http://mo bccrc.bitbucket.org/escape/ProstatePatientA22.html.C.7 Interactive CellScape of the xenograft SA501single cell data from Eirew et al. [5]http://mo bccrc.bitbucket.org/escape/SA501.html.C.8 Interactive TimeScape of AML patient Pa-tient 933124 from Ding et al. [3]http://mo bccrc.bitbucket.org/escape/AMLStacked.html.C.9 Interactive TimeScape of AML Patient 933124from Ding et al. [3], created with the centredgenotype layouthttp://mo bccrc.bitbucket.org/escape/AMLCentred.html.76C.10 Interactive TimeScape of AML Patient 933124from Ding et al. [3], created with the spacedgenotype layouthttp://mo bccrc.bitbucket.org/escape/AMLSpaced.html.C.11 Interactive TimeScape of triple negative breastcancer patient from Wang et al. [4]http://mo bccrc.bitbucket.org/escape/TNBC.html.C.12 Interactive TimesScapes of transformed fol-licular lymphoma patients from Kridel andChan et al. [6]http://mo bccrc.bitbucket.org/escape/TFL.html.77


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items