Novel Target Discovery of Existing Therapies: Path to Personalized Cancer Therapy

Abstract

Discovering new drugs involves tremendous effort and financial resources, often at a significant risk of failed trials. Identifying new targets of existing drugs provides a promising direction, especially for molecular targeted cancer therapies. This paper presents a novel, machine learning, and optimization-based method that identifies potential targets of existing drugs to expand the treatable patient population. The method has the following advantages: (1) It is based on clinical and genomic data from a large national cancer hospital; (2) it incorporates state-of-the-art knowledge of cancer molecular biology and signaling pathways; and (3) it models patient heterogeneity explicitly outside genomics. The output is an ordered list of therapy–target pairs that our algorithm identifies as highly promising to be further tested. The results are highly accurate when validated against known mechanisms of action for existing drugs, where relationships such as pertuzumab–ERBB2, cetuximab–EGFR, and erlotinib–EGFR were independently identified. We found similar results in the external The Cancer Genome Atlas data set. The findings suggest that a data-driven optimization approach to precision cancer medicine may lead to breakthroughs in the drug-discovery process and recommend effective personalized cancer treatments given patient-specific genomic and phenotypic information.