Analysis of Atomistic Representations Using Weighted Skip-Connections

TL;DR

This paper extends the SchNet architecture with weighted skip connections to analyze the importance of interaction blocks in molecular property prediction, revealing dependence on chemical composition and molecular configuration.

Contribution

It introduces weighted skip connections into SchNet, allowing analysis of interaction block importance based on molecular features, advancing understanding of ML models for molecules.

Findings

Interaction block importance varies with chemical composition.

Weighted skip connections reveal model interpretability.

Dependence on molecular configuration affects prediction accuracy.

Abstract

In this work, we extend the SchNet architecture by using weighted skip connections to assemble the final representation. This enables us to study the relative importance of each interaction block for property prediction. We demonstrate on both the QM9 and MD17 dataset that their relative weighting depends strongly on the chemical composition and configurational degrees of freedom of the molecules which opens the path towards a more detailed understanding of machine learning models for molecules.