Abstract
An outstanding challenge in deep learning in chemistry is its lack of interpretability. The inability of explaining why a neural network makes a prediction is a major barrier to deployment of AI models. This not only dissuades chemists from using deep learning predictions, but also has led to neural networks learning spurious correlations that are difficult to notice. Counterfactuals are a category of explanations that provide a rationale behind a model prediction with satisfying properties like providing chemical structure insights. Yet, counterfactuals have been previously limited to specific model architectures or required reinforcement learning as a separate process. In this work, we show a universal model-agnostic approach that can explain any black-box model prediction. We demonstrate this method on random forest models, sequence models, and graph neural networks in both classification and regression.
Highlights
Deep learning has made signi cant impacts in chemistry because of its ability to regress non-linear relationships between structure and function.[1]
We demonstrate this method on random forest models, sequence models, and graph neural networks in both classification and regression
Deep neural networks that take in raw graph representations of molecules have proven to be successful when compared with counterparts based on xed descriptors in both regression and classi cation tasks.[8]
Summary
Deep learning has made signi cant impacts in chemistry because of its ability to regress non-linear relationships between structure and function.[1] Applications vary from computing quantum properties[2,3] to predicting chemical properties[4,5] to screening drug molecules.[6,7] More speci cally, deep neural networks that take in raw graph representations of molecules have proven to be successful when compared with counterparts based on xed descriptors in both regression and classi cation tasks.[8] Despite their empirical accuracy, neural networks are black-box models; they lack interpretability and predictions come without explanation. Explainable arti cial intelligence (XAI) is an emerging eld which aims to provide explanations, interpretation, and justi cation for model predictions. XAI should be a normal part of the AI model lifecycle. It is becoming a legal requirement in some jurisdictions for AI to provide an explanation when used commercially.[11,12] From a researcher's perspective, XAI can nd the so-called “Clever Hans” effects whereby a model has learned spurious correlations such as the existence of a watermark in images or an over representation of counterions
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
