Towards predicting drug effects using neural networks

Accurate computational prediction of cancer patients’ response to therapies based on their unique molecular and clinical profiles is essential to assist clinicians in making decisions on the most effective and least toxic therapeutic options available. Here we develop a generative model based on a neural network machine learning algorithm, that uses pre-treatment genomic profile (specifically, transcriptome profiling using the L1000 technology) to predict the post-treatment profile. Using the Connectivity map (CMAP) data as a training set we highly accurately predict the post-treatment profile of each certain cell line given post-treatment data of other cell lines. Learning is simultaneously performed over ten different drugs, and may allow predictions of the activity of drug combinations. Analyzing the systems allows direct biological interpretation of different impacts of the drug, identification of side effects, and effective description of the drug’s pharmacodynamics.