graph neural network

Graph neural network (GNN) is a computational drug discovery framework that learns from molecular graphs to model compounds and their interactions, often by extracting both local substructures and global molecular topology. In practice, it can be multiscale-aware and densely connected, which helped improve screening of about 1.3 million compounds for novel antileishmanial hits and accelerated discovery of lc 61. It is also used in compound-protein interaction prediction, where the GNN component in deephybridcpi captures compound features for downstream modeling. Recent work shows interpretable GNNs can predict and optimize drug membrane permeability by targeting lipid membrane behavior, and the framework has been reported to improve AUC by 2.2–29.2% and 3.4–22.5% versus default GNNs.

Antiparasitic drug discovery

  • A multiscale-aware GNN framework improved identification of novel antileishmanial compounds and enabled screening of approximately 1.3 million compounds. (PMID:41841344)
  • The approach directly contributed to discovery of lc 61, a potent anti-Leishmania infantum compound. (PMID:41841344)
  • A 2026 Journal of Chemical Information and Modeling study highlighted the framework’s utility for large-scale virtual screening in neglected-disease chemistry. (PMID:41841344)

Compound-protein interaction prediction

  • deephybridcpi integrates a multiscale, densely connected GNN to extract compound features for compound-protein interaction prediction. (PMID:41564724)
  • The neural network component captures both local substructures and global molecular topology, improving representation of chemical structure. (PMID:41564724)
  • This hybrid framework was presented in a 2026 Journal of Molecular Graphics and Modelling paper as a general-purpose deep learning method for interaction prediction. (PMID:41564724)

Drug permeability and optimization

  • Interpretable GNN models were developed to predict and design drug membrane permeability, with explicit focus on lipid membrane interactions. (PMID:41941324)
  • The framework is thermodynamics-based and emphasizes interpretability for permeability optimization. (PMID:41941324)
  • Reported performance gains included AUC improvements of 2.2–29.2% and 3.4–22.5% over default GNN baselines. (PMID:41941324)
  • A 2026 Journal of Medicinal Chemistry study positioned GNNs as practical tools for permeability-aware medicinal chemistry. (PMID:41941324)