Relevance of Rotationally Equivariant Convolutions for Predicting Molecular Properties
This work addresses the problem of improving parameter efficiency in molecular property prediction for computational chemistry, though it is incremental as it builds on existing equivariant neural network methods.
The paper studied the impact of including angular features in rotationally equivariant convolutions for molecular property prediction, finding that adding angular features reduced test error by 23% on average, while increasing network depth only reduced error by 4%.
Equivariant neural networks (ENNs) are graph neural networks embedded in $\mathbb{R}^3$ and are well suited for predicting molecular properties. The ENN library e3nn has customizable convolutions, which can be designed to depend only on distances between points, or also on angular features, making them rotationally invariant, or equivariant, respectively. This paper studies the practical value of including angular dependencies for molecular property prediction directly via an ablation study with \texttt{e3nn} and the QM9 data set. We find that, for fixed network depth and parameter count, adding angular features decreased test error by an average of 23%. Meanwhile, increasing network depth decreased test error by only 4% on average, implying that rotationally equivariant layers are comparatively parameter efficient. We present an explanation of the accuracy improvement on the dipole moment, the target which benefited most from the introduction of angular features.