Neural Network Models for Prediction of Biological Activity using Molecular Dynamics Data

Strategic Insights -

Researchers at the Karlsruhe Institute of Technology (KIT) have developed neural network models that can predict the biological activity of complex molecules using molecular dynamics data. According to the study, these models can accurately predict the cytotoxic activity of similar peptide analogs and reliably distinguish between the photoisomers of the same peptide, even if the type of its activity differs from one in the training dataset. The researchers' approach, which incorporates classical molecular dynamics trajectories, has successfully generalized the ligand-based activity prediction neural network models to cases of large and conformationally flexible molecules.

Key Takeaways:

  • The researchers developed two neural network models that predict activities of photoswitchable peptidomimetics, analogs of the natural peptidic antibiotic gramicidin S.
  • The first model precisely predicts the cytotoxic activity of similar peptide analogs.
  • The second model reliably predicts the differences in biological activities of DAE photoisomers of the same peptide.
  • The study shows that accounting for molecular dynamics-derived dynamic features allows for the generalization of ligand-based activity prediction neural network models to cases of large and conformationally flexible molecules.
  • The researchers successfully created two neural network models using molecular dynamics-derived dynamic features.

Statistics:

  • 2 neural network models were developed in the study.
  • 1 model precisely predicts the cytotoxic activity of similar peptide analogs.
  • 1 model reliably predicts the differences in biological activities of DAE photoisomers of the same peptide.
  • The study includes molecular dynamics data from classical molecular dynamics (MD) trajectories.
  • The researchers used a dataset of photoswitchable peptidomimetics, analogs of the natural peptidic antibiotic gramicidin S.

Sources:

  • Neural Network Models for Prediction of Biological Activity using Molecular Dynamics Data: A Case of Photoswitchable Peptides by Sergii Afonin et al., Molecular Informatics, 2025;44(7)
  • Karlsruhe Institute of Technology (KIT)
  • Molecular Informatics (Wiley-v C H Verlag Gmbh, Postfach 101161, 69451 Weinheim, Germany)
  • Wiley-Blackwell (www.wiley.com/)
  • Online Library (onlinelibrary.wiley.com/journal/10.1002/(ISSN)1868-1751)