Distortion Discovery: A Framework to Model, Spot and Explain Tumor Heterogeneity and Mitigate its Negative Impact on Cancer Risk Assessment

ABSTRACT In a complex system of inter-genome interactions, false negatives remain an overwhelming problem when using omics data for disease risk prediction. This is especially clear when dealing with complex diseases like cancer in which the infiltration of stromal and immune cells into the tumor tissue can affect the degree of its tumor purity and hence its cancer signal. Previous work was done to estimate the degree of cancer purity in a tissue. In this work, we introduce a data and biomarker selection independent, information theoretic, approach to tackle this problem. We model distortion as a source of false negatives and introduce a mechanism to detect and remove its impact on the accuracy of disease risk prediction.