Pairwise Molecular Representation Learning for Improved and Explainable Property Prediction
Using tools and intuition straight from the chemist’s toolbox, namely a technique called Matched Molecular Pair Analysis, Tom MacDougal’s project seeks to improve the accuracy and explainability of activity predictions. By using a molecular design technique that is familiar to chemists, it is possible to appeal to the chemist’s intuition directly. Instead of having a system that instructs them what to do like robots, we hope to have one where the chemist and the predictive system can work cooperatively.
Keywords: Machine learning, deep learning, graph representation learning, drug discovery, matched molecular pair analysis