Open Collections

Data integration, variant aggregation and combined annotation

Banff International Research Station for Mathematical Innovation and Discovery

Majority of human diseases are complex, arising due to a multitude of factors. Identifying these factors is critical to understanding diseases and improving health care, yet it is a very difficult computational problem: low signal-to-noise ratio (only a few variants out of millions are likely to be causal), heterogeneity of reasons (e.g. coding, regulatory, epigenetic), epistasis (gene interaction patterns), etc. We propose to combine two mostly complementary data sources: coding variants and gene expression. These two data sources are responsible for different kinds of protein aberrations. Combining them allows us to survey both coding and regulatory aberrations genome wide without underpowering the model. We developed a biologically motivated hierarchical factor graph model which efficiently combines these two sources of data. We use variant harmfulness and gene interactions as priors, to increase the likelihood of identifying the genes correctly. To our knowledge, this is the first work that takes into account complementarity of exome and gene expression data sources in a principled way, integrating variant harmfulness and gene interaction information in the inference process of the model. Our approach a) allows to integrate different data modalities; b) provides a principled way to aggregate rare (and common) variants; c) improves the power of detecting genes associated with a given disease; d) implicates proteins that have been affected in the population in a variety of ways, rather than solely through the coding DNA sequence. Our extensive simulations confirm that our method has superior sensitivity and precision compared to other methods that aggregate rare variants. We have tested our approach in a large breast cancer dataset as a proof of concept and found that our method is able to identify important breast cancer genes. Interestingly, we find genes that have DNA mutations or coding variants in some patients and gene expression aberrations in other patients, indicating that our method is able to effectively explain the disease in more patients.

Mathematics; Statistics; Biology and other natural sciences

video/mp4

Author affiliation: SickKids Research Institute/ University of Toronto

2016-04-19

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/57676

Unreviewed

http://creativecommons.org/licenses/by-nc-nd/4.0/