In the relentless pursuit of novel solutions to combat the escalating antibiotic resistance crisis, a groundbreaking paper, published in the Nature Journal, reveals a pioneering approach to the discovery of new structural classes of antibiotics.
The paper challenges the current status quo in antibiotic discovery by leveraging the power of deep learning. Acknowledging the urgency of finding innovative antibiotics, the research team attempt to go beyond conventional methods and explore chemical spaces with a focus on explainability.
The prevailing issue with many deep learning models lies in their ‘black box’ nature, often lacking the ability to provide meaningful chemical insights. However, the researchers hypothesized that by identifying the chemical substructures associated with antibiotic activity, as learned by neural network models, it would be possible to predict and discover new structural classes of antibiotics.
To test this hypothesis, the team developed an explainable, substructure-based approach for the efficient, deep learning-guided exploration of chemical spaces. The study meticulously evaluated the antibiotic activities and human cell cytotoxicity profiles of an extensive set of 39,312 compounds. Employing ensembles of graph neural networks, the researchers successfully predicted antibiotic activity and cytotoxicity for an impressive 12,076,365 compounds.
What sets this approach apart is the use of explainable graph algorithms. Through these algorithms, the researchers identified substructure-based rationales for compounds exhibiting high predicted antibiotic activity and low predicted cytotoxicity. The real-world efficacy of this methodology was further validated through the empirical testing of 283 compounds.
Significantly, the study revealed that compounds demonstrating antibiotic activity against Staphylococcus aureus were enriched in putative structural classes arising from identified rationales. Of particular note, one of these structural classes exhibited selectivity against methicillin-resistant S. aureus (MRSA) and vancomycin-resistant enterococci, demonstrated resistance evasion, and notably reduced bacterial titers in mouse models of MRSA skin and systemic thigh infection.
This breakthrough not only opens new avenues for the discovery of structural classes of antibiotics but also emphasizes the feasibility of creating explainable machine learning models in drug discovery. By providing insights into the chemical substructures underlying selective antibiotic activity, this research marks a significant step towards combating antibiotic resistance.
Contributors of this research: MIT, Harvard, Broad Institute, Integrated Biosciences, Inc., the Wyss Institute for Biologically Inspired Engineering, and the Leibniz Institute of Polymer Research in Dresden, Germany.
More information: Wong, F., Zheng, E.J., Valeri, J.A. et al. Discovery of a structural class of antibiotics with explainable deep learning. Nature (2023). https://doi.org/10.1038/s41586-023-06887-8.
Stay tuned as Targeting Phage Therapy 2024 will cover all the most recent discoveries in the field.