cfOncoXpress: Tumor gene expression prediction from cell-free DNA whole-genome sequences

Abstract

Cell-free DNA (cfDNA) fragments in the plasma capture cellular nucleosomal profiles since nucleosome-protected regions escape enzymatic degradation while nucleosome-depleted regions can not. We developed cfOncoXpress, a machine learning framework that uses fragmentation patterns to predict oncogene expression from cfDNA WGS. cfOncoXpress incorporates gene copy number aberrations inferred from cfDNA, including those associated with extrachromosomal DNA. Its application in prostate and breast cancers shows it can predict tumor subtype based on expression of signature genes and activated pathways. cfOncoXpress shows superior performance relative to other state-of-the-art methods and can be used to predict tumor gene expression when tissue biopsies are infeasible.