Pretraining Strategy for Neural Potentials

arXiv:2402.15921v1 Announce Type: new
Abstract: We propose a mask pretraining method for Graph Neural Networks (GNNs) to improve their performance on fitting potential energy surfaces, particularly in water systems. GNNs are pretrained by recovering spatial information related to masked-out atoms from molecules, then transferred and finetuned on atomic forcefields. Through such pretraining, GNNs learn meaningful prior about structural and underlying physical information of molecule systems that are useful for downstream tasks. From comprehensive experiments and ablation studies, we show that …

abstract arxiv cs.lg energy gnns graph graph neural networks information learn molecules networks neural networks performance physics.chem-ph pretraining spatial strategy systems through type water