Researchers at MIT screened more than 12 million compounds with a graph neural network, but the key step was different from past AI drug searches: the model also had to show which chemical substructures it was using to make each prediction.
That explainability step led to a compound class active against MRSA and vancomycin-resistant enterococci. In mouse models of skin and systemic MRSA infection it reduced bacterial counts substantially, and resistance did not emerge in early testing the way it does against older antibiotics.
Knowing why the model flagged a compound, not just that it did, let chemists understand what made the structure work and refine it faster than trial and error would allow. The compound is still in early development, years from any human trial, but the explainable-AI method itself has since been applied to search for antibiotics against other resistant pathogens.
Read the original study
Wong et al., Nature, 2023 · doi.org/10.1038/s41586-023-06887-8
A machine that could say why, not just what.
Source
- Wong, F., Zheng, E. J., Valeri, J. A., Donghia, N. M., Anahtar, M. N., Omori, S., Li, A., Cubillos-Ruiz, A., Krishnan, A., Jin, W., Manson, A. L., Friedrichs, J., Helbig, R., Hajian, B., Fiejtek, D. K., Wagner, F. F., Soutter, H. H., Earl, A. M., Stokes, J. M., . . . Collins, J. J. (2024). Discovery of a structural class of antibiotics with explainable deep learning. Nature, 626(7997), 177–185. https://doi.org/10.1038/s41586-023-06887-8