With almost 5 million deaths attributed to antibiotic resistance worldwide each year, there’s an urgent need for innovative solutions to address resistant bacterial strains. In response, researchers at Stanford Medicine and McMaster University are harnessing the power of generative artificial intelligence (AI). Their novel model, named SyntheMol (short for synthesizing molecules), has devised structures and chemical formulas for six innovative drugs targeted at eliminating resistant strains of Acinetobacter baumannii, a prominent pathogen contributing to antibacterial resistance-related fatalities.
The study, outlined in a publication in the journal Nature Machine Intelligence on March 22, outlines the researchers’ model and the experimental validation of these new compounds. According to James Zou, Ph.D., an associate professor of biomedical data science and co-senior author of the study, there’s a pressing public health demand for the rapid development of new antibiotics. The hypothesis driving their research was the existence of numerous potential molecules with effective drug properties yet to be explored or tested, motivating the utilization of AI to design entirely novel molecules.
Before the emergence of generative AI, akin to the technology underlying large language models like ChatGPT, researchers employed various computational methods for antibiotic development. They utilized algorithms to sift through extensive drug libraries, identifying compounds with potential efficacy against specific pathogens. However, this approach, scrutinizing a mere fraction of the vast chemical space, only scratched the surface of potential antibacterial compounds.
Generative AI’s propensity for “hallucinating” or conjuring responses from scratch presents an opportunity in drug discovery. Yet, previous attempts at generating new drugs through this AI often produced compounds impractical to synthesize. To address this challenge, the researchers imposed constraints on SyntheMol’s activity to ensure the practical synthesis of any molecules it devised.
The model was trained using a library of over 130,000 molecular building blocks and a collection of validated chemical reactions. It not only generated the final compounds but also outlined the steps involved in their synthesis. Leveraging existing data on chemicals’ antibacterial activity against A. baumannii, SyntheMol generated approximately 25,000 potential antibiotics and corresponding synthesis instructions within nine hours.
To mitigate the risk of bacterial resistance, the researchers filtered the generated compounds to select those sufficiently distinct from existing ones. Subsequently, 70 compounds with the highest potential to combat the bacterium were chosen for synthesis, with 58 successfully synthesized by the Ukrainian chemical company Enamine.
Among these synthesized compounds, six demonstrated efficacy in killing a resistant strain of A. baumannii during laboratory testing. Additionally, these new compounds exhibited antibacterial activity against other antibiotic-resistant infectious bacteria, including E. coli, Klebsiella pneumoniae, and MRSA.
Further toxicity testing in mice confirmed the safety of two of the six compounds, while the next phase involves evaluating their efficacy in mice infected with A. baumannii. These compounds represent a significant departure from existing antibiotics, and though the exact mechanisms of their antibacterial properties remain unknown, their exploration could yield valuable insights applicable to broader antibiotic development.
According to Zou, SyntheMol is not only designing new molecules but also unveiling an entirely unexplored chemical space, offering potential breakthroughs in antibiotic discovery. Moving forward, the researchers aim to refine SyntheMol’s capabilities and extend its applications, collaborating with other research groups to explore drug discovery for heart disease and develop new fluorescent molecules for laboratory research.