Interpreting Graph Neural Networks for Antibacterial Drug Repurposing against Acinetobacter baumannii

Piperazine is the most consistently supported recurring contributor to antibacterial activity predictions for Acinetobacter baumannii in a graph neural network ensemble, identified via atom-level perturbation and restoration.

• Why it matters: Carbapenem-resistant Acinetobacter baumannii is a critical priority, and model predictions need interpretable molecular features to translate screening outputs into chemical hypotheses.
• What they did: The study used an atom-feature and connectivity graph neural network ensemble, interpreting atom-level contributions through perturbation and restoration, and compared against equally sized perturbations within each molecule.
• The result: Prospective Drug Repurposing Hub screening produced a ranked shortlist, and experimental testing of CE3F4 reduced Acinetobacter baumannii growth by 57% at 0.25 mM, with a non-monotonic concentration–response profile.

AI summary of the abstract · may contain errors · not medical advice · read the paper.