A Minimal-Input Multilayer Perceptron for Predicting Drug-Drug Interactions Without Knowledge of Drug Structure

Abstract

Predictive models in the drug discovery industry have an essential role that cannot be understated. With the sheer volume of potentially useful compounds that are considered for use, it is becoming more difficult to investigate overlapping interactions between two drugs. Given that recreational drugs lack the rigorous warnings of prescription drugs, it is important for the layperson to know which drugs can and cannot mix. Other methods are necessary to bridge this knowledge gap in the absence of deterministic experimental results for every drug combination. Ideally, such methods would require minimal inputs, have high accuracy, and be computationally feasible. We have not encountered a model that meets all the above criteria. In light of this, we propose a minimal-input multi-layer perceptron that predicts the interactions between two drugs. This model has a great advantage of requiring no structural knowledge of the molecules in question, and instead only uses experimentally accessible chemical and physical properties; in particular, 20 per compound. These 20 were selected as they seemed they may be pertinent to the authors, although given this was an exploratory model, there was not a rigorous method behind the choice. Using a set of known drug-drug interactions and associated properties of the drugs involved, we trained our model on a dataset of about 650,000 entries. We report an accuracy of 0.968 on unseen samples of interactions between drugs on which the model was trained and an accuracy of 0.942 on unseen samples of interactions between unseen drugs. We believe this to be a promising and highly extensible model that has the potential for high generalized predictive accuracy with further tuning.