Abstract

As access to computing power, open source data and machine learning packages becomes greater, as does the growth of molecular machine learning (MML). Such technology can attempt to provide answers to a wide range of chemistry based questions. Most techniques with good performance are however unable to answers one key question “but why?”. Much of science is achieved through pondering such a question so the ability to provide an answer to this question is something which is strived for within MML . Explainable AI has been introduced into the realm of MML with the goal of giving reasons why. In this work molecular counterfactuals have been implemented and explored as an explainable AI technique, as well as an investigation on the STONED algorithm and how augmentation of this can impact the quality of produced counterfactuals.