Prediction rigidities for data-driven chemistry

TL;DR

This paper introduces prediction rigidities, metrics derived from the loss function, to evaluate and improve the robustness, interpretability, and transferability of machine learning models in chemical property prediction.

Contribution

It demonstrates the utility of prediction rigidities for assessing model robustness at multiple levels and guides dataset construction for more efficient training in chemical ML applications.

Findings

Prediction rigidities effectively assess model robustness globally and locally.

These metrics guide the construction of training datasets for improved model performance.

Application to a coarse-grained system shows broad applicability of the approach.

Abstract

The widespread application of machine learning (ML) to the chemical sciences is making it very important to understand how the ML models learn to correlate chemical structures with their properties, and what can be done to improve the training efficiency whilst guaranteeing interpretability and transferability. In this work, we demonstrate the wide utility of prediction rigidities, a family of metrics derived from the loss function, in understanding the robustness of ML model predictions. We show that the prediction rigidities allow the assessment of the model not only at the global level, but also on the local or the component-wise level at which the intermediate (e.g. atomic, body-ordered, or range-separated) predictions are made. We leverage these metrics to understand the learning behavior of different ML models, and to guide efficient dataset construction for model training. We finally implement the formalism for a ML model targeting a coarse-grained system to demonstrate the applicability of the prediction rigidities to an even broader class of atomistic modeling problems.