What can attribution methods show us about chemical language models?

Language models trained on molecular string representations have shown strong performance in predictive and generative tasks. However, practical applications require not only making accurate predictions, but also explainability - the ability to explain the reasons and rationale behind the predictions. In this work, we explore explainability for a chemical language model by adapting a transformer-specific and a model-agnostic input attribution technique.

Self-Organized Resonance during Search of a Diverse Chemical Space.

Recent studies of active matter have stimulated interest in the driven self-assembly of complex structures. Phenomenological modeling of particular examples has yielded insight, but general thermodynamic principles unifying the rich diversity of behaviors observed have been elusive. Here, we study the stochastic search of a toy chemical space by a collection of reacting Brownian particles subject to periodic forcing.