Deep Learning Model

About

Background: In small molecule drug discovery pipelines, identification of mechanism of action and therapeutic potential of compounds with known bioactivity is a time-intensive process, often taking years to arrive at a conclusion. Current methods focus on a single therapeutic indication (i.e. antibacterial, non-steroidal anti-inflammatory, anticancer) which are typically hypothesis-driven, intuitive, and slow. Therefore, there is a need for improved predictive tools in identifying possible mechanisms of action and therapeutic utilities. Gene expression studies of a wide variety of cell types have revealed that various cell systems have complex unique fingerprint responses to chemical perturbants. Researchers at McMaster have developed a deep learning model which predicts the mechanism of action of chemical compounds based on such microbial fingerprint responses. This densely connected network uses unique microbial responses to chemical inputs to map unknown (test set) compounds to a well-annotated database (training set) of compounds. By utilizing a sensitive microorganism reporter, the biological activity inherent in any drug may be detected, even if it is not an antibiotic.