CS224W Final Project Report

Gene interaction networks have seen limited utility in drug discovery settings. By synergistically leveraging two new datasets, we propose a framework for more effective therapeutic interventions based on biologically realistic data. First, we develop and characterize a new gene-inhibition network, which is based on single gene inhibition data. Our network displays a previously uncharacterized in-degree structure which is more representative of core biological and evolutionary principles than prior findings. We develop and validate a novel greedy algorithm, which is focused on maximizing disease state reduction while minimizing off-target effects and the number of required therapeutics. Our greedy algorithm takes an NP-complete problem and identifies a high-accuracy solution in linear time. We observe that more gene knockout data is necessary before our approach will deliver clinically actionable results.