UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Cell-conditional generative adversarial network Zhang, Xi


With single cell sequencing advances, research has increasingly focused on under-standing cell-specific gene regulation mechanisms. However, single cell sequencing data are often noisy and the amount of sequence obtained from rare cell types small. Simulation can be a powerful approach to aid understanding when data is limited, both because the process used to generate such data can provide mechanistic insights into cell-specific regulation and the data produced can augment analysis methods development. We constructed and optimized a stand-alone cell-conditional GAN (ccGAN) to simulate cell-specific ATAC-seq data. We trained our model on published single cell ATAC-seq (scATAC-seq) data that had been produced with different protocols on embryonic mice forebrain and adult mice brain. The ccGAN generated sequence was correlated in both Transcription Factor (TF) binding motif composition and positional distribution with the experimental scATAC-seq. The ccGAN simulator was able to learn important cell-specific signals amidst noise. The ccGAN architecture holds broad potential for single cell regulatory data simulation beyond ATAC-seq, such as for ChIP-seq or epigenetic properties

Item Media

Item Citations and Data


Attribution-NonCommercial-NoDerivatives 4.0 International