PADME: A Deep Learning-based Framework for Drug-Target Interaction Prediction.

In silico Drug-target Interaction (DTI) prediction is an important and challenging problem in medicinal chemistry with a huge potential benefit to the pharmaceutical industry and patients. Most existing methods for DTI prediction generally have binary endpoints, which could be an oversimplification of the problem. With the advent of deep learning, some deep learning models were devised to solve the DTI prediction problem, but most of them still use binary endpoints, and they are generally unable to handle cold-target problems, i.e., problems involving target protein that never appeared in the training set. We contrived PADME (Protein And Drug Molecule interaction prEdiction), a framework based on Deep Neural Networks, to predict real-valued interaction strength between compounds and proteins. PADME inputs both compound and protein information into the model, so it is applicable to cold-target problems. To our knowledge, we are also the first to incorporate Molecular Graph Convolution (MGC) into the model for compound featurization. We used different Cross-Validation split schemes and different metrics to measure the performance of PADME on multiple datasets (in which we are the first to use ToxCast for such problems), and PADME consistently dominates baseline methods. We also conducted a case study, predicting the interaction between compounds and androgen receptor (AR) and compared the prediction results with growth inhibition activity of the compounds in NCI60, which also gave us satisfactory results, suggesting PADMEu0027s potential in drug development. We expect different variants of PADME to be proposed and experimented on in the future, and we believe Deep Learning will transform the field of cheminformatics.