DPLM: Dynamics-aware Protein Language Model via Contrastive Learning between Sequence and Molecular Dynamics Simulation Trajectory
Authors:
Preprint
Abstract
Protein dynamics captures the interactions between residues and the conformational changes they undergo. Encoding such information could enhance a deep learning model’s ability to predict protein properties. Here we introduce a dynamics-aware protein language model (DPLM) that encodes protein dynamics into a sequence-derived representation. We represent each molecular dynamics (MD) trajectory as a sequence of backbone Cα contact maps and encode it with a Video Vision Transformer (ViViT). We then align the ViViT-derived dynamics embedding with the ESM2-derived sequence embedding using dual-view contrastive learning. The root-mean-square fluctuation correlation and dynamic cross-correlation matrix reconstruction confirm that contrastive learning incorporates dynamics information into the sequence embedding. We applied DPLM as a protein representation backbone for predicting diverse dynamics-related downstream tasks, including liquid-liquid phase separation, cryptic-site detection, and mutation effect on protein thermal stability. DPLM shows strong performance, especially where the evolutionary or structural signal is weak. These results show that learning the geometry of molecular motion yields transferable dynamic information that complements the evolutionary and structural features of existing PLMs. Significance Statement: Proteins perform their functions in constant motion. Most protein language models (PLMs) miss this dynamic dimension, learning only from sequence or fixed structure. A few emerging PLMs model dynamics through conformational snapshots or derived dynamic properties rather than the full time-resolved motion. As a result, the temporal ordering inherent in molecular motion is not explicitly captured. We introduce DPLM, a dynamics-aware protein language model that uses self-supervised learning to align protein sequences with their molecular dynamics simulation trajectories. The resulting sequence-based representation encodes general dynamic information and benefits diverse downstream tasks, especially where evolution and structure alone fall short. DPLM offers a general route to physics-informed protein representation.Keywords
This publication has 80 references indexed in Scilit:
- Critical Assessment of Protein Intrinsic Disorder Round 3 ‐ Predicting Disorder in the Era of Protein Language ModelsProteins: Structure, Function, and Bioinformatics, 2025
- Critical assessment of protein intrinsic disorder prediction (CAID) ‐ Results of round 2Proteins: Structure, Function, and Bioinformatics, 2023
- Discovery of deaminase functions by structure-based protein clusteringCell, 2023
- Intrinsic protein disorder and conditional folding in AlphaFoldDBProtein Science, 2022
- GPS 5.0: An Update on the Prediction of Kinase-Specific Phosphorylation Sites in ProteinsGenomics, Proteomics & Bioinformatics, 2020
- SPOT-Disorder2: Improved Protein Intrinsic Disorder Prediction by Ensembled Deep LearningGenomics, Proteomics & Bioinformatics, 2019
- Molecular Dynamics Simulation for AllNeuron, 2018
- MDTraj: A Modern Open Library for the Analysis of Molecular Dynamics TrajectoriesBiophysical Journal, 2015
- Stability effects of mutations and protein evolvabilityCurrent Opinion in Structural Biology, 2009
- Binding of Small-Molecule Ligands to Proteins: “What You See” Is Not Always “What You Get”Structure, 2009