Scientists say they have discovered one of the first new classes of antibiotics identified in the past 60 years, and the first discovered leveraging an AI-powered platform built around explainable deep learning.
Published in Nature, the peer-reviewed paper, “Discovery of a structural class of antibiotics with explainable deep learning,” was co-authored by a team of 21 researchers, led by Felix Wong, PhD, co-founder of Integrated Biosciences, and James J. Collins, PhD, the Termeer professor of medical engineering and science at MIT and founding chair of the Integrated Biosciences Scientific Advisory Board.
“The discovery of novel structural classes of antibiotics is urgently needed to address the ongoing antibiotic resistance crisis. Deep learning approaches have aided in exploring chemical spaces; these typically use black box models and do not provide chemical insights. Here we reasoned that the chemical substructures associated with antibiotic activity learned by neural network models can be identified and used to predict structural classes of antibiotics,” the investigators wrote.
“We tested this hypothesis by developing an explainable, substructure-based approach for the efficient, deep learning-guided exploration of chemical spaces. We determined the antibiotic activities and human cell cytotoxicity profiles of 39,312 compounds and applied ensembles of graph neural networks to predict antibiotic activity and cytotoxicity for 12,076,365 compounds.
“Using explainable graph algorithms, we identified substructure-based rationales for compounds with high predicted antibiotic activity and low predicted cytotoxicity. We empirically tested 283 compounds and found that compounds exhibiting antibiotic activity against Staphylococcus aureus were enriched in putative structural classes arising from rationales. Of these structural classes of compounds, one is selective against methicillin-resistant S. aureus (MRSA) and vancomycin-resistant enterococci, evades substantial resistance, and reduces bacterial titers in mouse models of MRSA skin and systemic thigh infection.
“Our approach enables the deep learning-guided discovery of structural classes of antibiotics and demonstrates that machine learning models in drug discovery can be explainable, providing insights into the chemical substructures that underlie selective antibiotic activity.”
Experimentally generated data
In this approach, the scientific team trained deep learning models on experimentally generated data to predict the antibiotic activity and toxicity of any compound. Drawing ideas from AI used in other contexts, including DeepMind’s AlphaGo gaming technology, the authors designed new models to explain which parts of a molecule were important for antibiotic activity.
The result was the identification of a new class of antibiotics with potent activity against multidrug-resistant pathogens. In one series of experiments, the researchers tested a candidate antibiotic in mouse models of MRSA infection and found that it was efficacious both topically and systemically, indicating that the compound could be suitable for further development as a treatment for severe and sepsis-related bacterial infections.
“This discovery of a new class of antibiotics is a breakthrough result showing that artificial intelligence and explainable deep learning are uniquely capable of catalyzing drug discovery,” said Wong. “Our work makes publicly available several high-powered models to accurately predict both antibiotic activity and toxicity. Importantly, this is one of the first demonstrations that deep learning models can explain what they are predicting, with immediate and far-reaching implications to how drug discovery is done and how efficiently we can find new drugs using AI.”
“This is an important validation of how important the integration of AI and explainable deep learning will be to overcoming some of the most vexing challenges in medicine, in this case antibiotic resistance,” added Collins. “Building on these validating studies and similar approaches, the Integrated Biosciences team is poised to further accelerate their integration of synthetic biology and a deep understanding of cellular stress to address a significant unmet need for new treatments targeting age-related diseases.”
“An important implication of this study is that deep learning models in drug discovery can, and in many cases should, be made explainable,” noted Satotaka Omori, PhD, founding member and head of aging biology at Integrated Biosciences. “While AI continues to make an impact, it is also limited by the many black box models that are commonly used and obfuscate the underlying decision-making process. By opening up these black boxes, we aim to create more generalizable insights that may be more useful in accelerating the use and development of next-generation approaches to drug discovery.”
Additional collaborators included researchers at MIT, the Broad Institute of MIT and Harvard, the Wyss Institute for Biologically Inspired Engineering, and the Leibniz Institute of Polymer Research in Dresden, Germany. In their study, the researchers virtually screened more than 12 million candidate compounds to identify this new class of antibiotics, which show potential to address antibiotic resistance.
