Predicting cancer drug effectiveness with deep learning artificial intelligence.

Despite advances in molecular technologies, application of precision medicine in clinical oncology remains difficult. Although widely used, single-gene biomarkers are imperfect predictors for the effective administration of targeted therapy. Meanwhile, nonspecific cytotoxic medications lack established biomarkers to guide their usage, yet they remain first-line chemotherapy for many patients. As the formulary of precision medications and tissue-agnostic treatments expands, there is a pressing and growing need for sensitive and specific companion diagnostic testing. The effective application of powerful computational techniques holds great potential for assisting clinicians and patients in the navigation of ever-increasingly complex treatment decisions. To address this challenge, we applied state of the art deep learning techniques and modern machine learning frameworks to develop a deep neural network autoencoder for learning latent representations of integrated omics data from cancer cell lines. We used these representations to build predictive models of drug sensitivity. We evaluated the effectiveness of these models using a variety of preclinical and clinical data to assess potential for translational impact. This research is significant in three primary ways. First, we developed a novel data-driven approach to precision medicine. Second, we demonstrated the potential for this approach to improve clinical outcomes relative the current standard of care. Third, we demonstrated that this approach not only optimizes therapeutic efficacy in preclinical cancer models, but is also generalizable to real, clinical tumors.