Researchers at the American Chemical Society Fall Meeting have presented their findings on the application of Explainable AI (XAI) in the field of chemistry. XAI, a subset of artificial intelligence, aims to provide justification for AI models’ decisions, offering transparency and insight into their decision-making process. The researchers, led by chemistry professor Rebecca Davis from the University of Manitoba, focused on applying XAI to AI models used in drug discovery.
The use of AI has become widespread across various industries, but the lack of transparency in AI models has raised concerns among scientists and the public. XAI offers a solution by allowing researchers to delve into the inner workings of AI decision-making. By understanding how AI models arrive at their predictions, scientists can gain more confidence in these methodologies.
Davis and her team fed databases of known drug molecules into an AI model to predict their biological effects. They then employed an XAI model developed by collaborator Pascal Friederich from Germany’s Karlsruhe Institute of Technology to analyze the specific components of the drug molecules that influenced the AI model’s predictions. This process shed light on why certain molecules exhibited activity or lacked it, providing insights into the AI model’s categorization process.
The researchers discovered that XAI can identify crucial molecular structures that may be overlooked by human chemists. For instance, when examining a set of penicillin molecules, XAI identified structures attached to the core as the critical factor in classification, rather than the core itself. This finding challenges the conventional belief among chemists and may explain why some penicillin derivatives with the core structure exhibit poor biological activity.
In addition to uncovering important molecular structures, the researchers aim to leverage XAI to enhance predictive AI models. By understanding what the AI algorithms consider important for antibiotic activity, they can train the models to focus on specific criteria. The team plans to collaborate with a microbiology lab to synthesize and test compounds predicted by the improved AI models as potential antibiotics.
The ultimate goal of this research is to develop better or entirely new antibiotic compounds, which could combat the growing threat of antibiotic-resistant pathogens. Davis emphasizes that XAI plays a crucial role in building trust and acceptance of AI technology by allowing it to explain its decision-making process.
Hunter Sturm, a graduate student in Davis’ lab, believes that AI applications in chemistry and drug discovery represent the future of the field. He sees his work as laying the foundation for advancements in these areas.