A machine learning framework to reveal systems-level biological signatures by integrating molecular signatures from regulatory and functional multi-omics data

Biological traits and diseases are the result of a system-wide orchestration of dynamic biomolecular interactions, encompassing nucleic acids, proteins and metabolites. Conventional experiments target a single viewpoint corresponding to one biomolecular “omics” data category due to the complexity of analysing entire ensembles of biological molecules. Intuitively, consolidating and viewing a biological system as a whole will deepen our understanding of its regulation, key components and output. We achieved this systems-wide view by operating on molecular abundance data in a raw state to preserve information, and applying novel machine learning and computational linguistics techniques to understand the grammar of the genome.