Molecular dynamics simulation prediction method and apparatus, device, and storage medium

Abstract

A molecular dynamics simulation prediction method, executed by a pre-trained dynamics simulation prediction model. The model comprises a graph network structure and a gating structure which are deployed on a GPU device, and at least one parallel expert model deployed on a CPU device. The specific method comprises: acquiring a protein sequence to be tested corresponding to a protein; inputting said protein sequence into the graph network structure for feature extraction to obtain graph feature data, and inputting the graph feature data into the gating structure for processing to obtain an allocation result; and inputting the graph feature data and the allocation result into at least one expert model for data prediction to obtain a predicted potential energy and a predicted force field that correspond to each peptide unit in said protein sequence. In the method, the inference speed of a mixture-of-experts model is accelerated by using the processing advantages of a heterogeneous cluster, the scale of the mixture-of-experts model in a memory is increased, and inference accuracy is improved.

Images