Dice Question Streamline Icon: https://streamlinehq.com

Practical re-scoring performance of models trained on ensemble-averaged MMPBSA labels without explicit MD modeling

Ascertain whether models trained using ensemble-averaged MMPBSA labels derived from crystal structures, without explicitly modeling molecular dynamics trajectories, can accurately re-score and generalize to novel protein-ligand co-complex structures in practical applications.

Information Square Streamline Icon: https://streamlinehq.com

Background

PLAS20k provides ensemble-averaged MMPBSA energies for PDB co-complexes and trains models using crystal structures, but does not explicitly model the MD trajectories themselves.

The authors point out that, due to this omission, it is unclear how such models would perform when tasked with re-scoring novel structures in practice, emphasizing a need for empirical evaluation of generalization.

References

While the study provides insight into the general problem of learning the MMPBSA energetics, it also doesn't explicitly model the MD simulation data itself, making it unclear how a model trained using a re-scoring method for label generation, would perform when tasked with re-scoring novel structures in practice.

— SurGBSA: Learning Representations From Molecular Dynamics Simulations (2509.03084 - Jones et al., 3 Sep 2025) in Section 1 (Introduction)