Artificial intelligence (AI) has become increasingly popular and is used in various applications such as driving vehicles, proofreading emails, and designing new molecules for medications. However, understanding how AI makes decisions can be challenging, as many AI models are considered black boxes. Explainable AI (XAI) is a subset of AI technology that aims to justify a model’s decisions, providing insight into the reasoning behind the outcomes. Researchers are now using XAI to delve deeper into the field of chemistry, with the goal of scrutinizing predictive AI models and improving understanding of how AI makes decisions in drug discovery.
At the fall meeting of the American Chemical Society (ACS), researchers will present their work on applying XAI to AI models for drug discovery. The use of XAI can help scientists see behind the scenes of AI decision making, offering valuable insights into the factors that contribute to the predictions made by AI models. With thousands of candidate molecules being screened to approve just one new drug, accurate prediction models are crucial for identifying potential antibiotic candidates. By using XAI to better understand the information needed to teach computers chemistry, researchers aim to improve prediction models and enhance drug discovery processes.
The researchers utilized an AI model to predict the biological effects of drug molecules by feeding databases of known drug molecules. They also employed an XAI model to examine the specific parts of the drug molecules that influenced the model’s predictions. Through this process, they gained insights into the critical factors that the AI model deemed important in predicting antibiotic activity, highlighting structures beyond the traditional core of penicillin as key elements in antibiotic classification. This discovery demonstrates the ability of XAI to identify vital molecular structures that may have been overlooked by human researchers.
In addition to identifying crucial molecular structures, the researchers aim to use XAI to enhance predictive AI models for drug discovery. By leveraging the insights provided by XAI on what algorithms define as important for antibiotic activity, they can train AI models more effectively on identifying relevant features in drug molecules. The team plans to collaborate with a microbiology lab to synthesize and test compounds predicted by improved AI models, with the goal of developing better antibiotic compounds to combat antibiotic-resistant pathogens.
The researchers hope that the use of XAI in drug discovery will not only lead to the creation of better antibiotic compounds but also increase acceptance and trust in AI technology. By asking AI to explain its decisions, there is a greater likelihood that people will embrace and understand the technology. Furthermore, the team believes that AI applications in chemistry and drug discovery represent the future of the field, with the potential to revolutionize the way new drugs are developed. The research was supported by various funding sources, including the University of Manitoba, the Canadian Institutes of Health Research, and the Digital Research Alliance of Canada.
