Graduation thesis at the Laboratory of Chemoinformatics, Nankai University, advised by Prof. Wensheng Cai.

  • Adapted the SwAV contrastive framework to molecular dynamics and designed a GNN to learn representations directly from raw 3D coordinates.
  • Enforced translational invariance through zero-centering and empirical rotational invariance via SO(3) data augmentation.
  • Benchmarked on alanine dipeptide trajectories and showed that the learned representations capture distinct metastable states in the classical free-energy landscape.