Improving Molecular Modeling with Geometric GNNs: an Empirical Study
It provides guidance for researchers in materials science to choose optimal modeling components, but it is incremental as it focuses on empirical comparisons rather than introducing new methods.
This paper tackles the challenge of selecting effective machine learning techniques for molecular modeling by empirically studying Geometric Graph Neural Networks on 3D atomic systems, evaluating the impact of canonicalization methods, graph creation strategies, and auxiliary tasks on performance, scalability, and symmetry enforcement.
Rapid advancements in machine learning (ML) are transforming materials science by significantly speeding up material property calculations. However, the proliferation of ML approaches has made it challenging for scientists to keep up with the most promising techniques. This paper presents an empirical study on Geometric Graph Neural Networks for 3D atomic systems, focusing on the impact of different (1) canonicalization methods, (2) graph creation strategies, and (3) auxiliary tasks, on performance, scalability and symmetry enforcement. Our findings and insights aim to guide researchers in selecting optimal modeling components for molecular modeling tasks.