Approved research

Epistasis analysis of cancer susceptibility

Lay summary

The overarching goal of this 3-year project is to develop algorithms to facilitate detection, characterization, and interpretation of non-additive interactions between genome-wide sets of common and rare genetic variants that influence disease susceptibility and to apply these algorithms to identify such interactions associated with different types of cancer, such as breast, bladder, colorectal, and lung cancer. In the typical approach to the study of association between genetic variants and a trait, each genetic variant is treated as an independent unit. It is our working hypothesis that the relationship between genomic variation and phenotypic variation is very complex and that this complexity may be partly due to context-dependent genetic effects that are the result of non-additive variant-variant and variant-environment interactions. We propose to develop an innovative machine learning approach for the detection of disease associated variants that takes into consideration the hierarchical complexity of the genome, i.e. the gene structure and biological interactions between genes and other functional elements that give rise to complex hierarchical interactions. Our goal is to improve the power to detect disease associated genetic complexity as well as our ability to interpret the results because the models will be constructed in a hierarchical manner with constructed features using biology as a guide.