A University of Queensland research team has developed a framework to assess whether artificial intelligence can provide reliable explanations for its recommendations during antibiotic discovery, aiming to improve confidence in AI-assisted drug development.
The study, published in the Journal of Cheminformatics, addresses one of the biggest barriers to using AI in antimicrobial research – the inability of many models to explain how they reach their conclusions.
Lead researcher Dr Abdulmujeeb Onawole, from UQ’s Centre for Superbug Solutions, said AI had the potential to accelerate the search for new antibiotics but scientists needed to be able to trust its reasoning.
Antimicrobial resistance, including resistance to antibiotics, is threatening healthcare globally by limiting effective treatment options against multidrug-resistant pathogens known as ‘superbugs.’
“This is a high-stakes field and while AI can help us to save lives faster, we want to ensure the humans involved can make an informed decision,” Dr Onawole said.
“AI is revolutionising drug development, but scientists struggle to trust its recommendations because the technology often can’t explain its reasoning,” he said.
“We call this the ‘black box’ of AI – where AI provides an answer but there’s no explanation of how it got there – and this is preventing scientists understanding the chemical reasoning behind its predictions.
“This lack of transparency is dangerous during antibiotic development as misleading AI explanations can lead to incorrect decisions and wasted resources in the lab.”
Researchers developed three AI models using datasets of chemical compounds previously tested against the superbug bacterium Staphylococcus aureus. They then applied the new framework to evaluate whether each model could correctly identify important antibiotic structures and explain how small chemical changes affected a compound’s activity.
The study found all three models successfully identified known antibiotic structures but varied significantly in their ability to provide reliable chemical explanations.
Co-author Dr Johannes Zuegg said the framework enabled researchers to assess whether AI systems could deliver trustworthy explanations alongside their predictions.
“We have shown our framework can successfully assess if AI systems can provide trustworthy chemical explanations, which is critical to medicinal chemists in drug development,” Dr Zuegg said.
The researchers say improving the transparency of AI models could help accelerate the discovery of new antibiotics to combat antimicrobial resistance while reducing the risk of misleading recommendations during drug development.
Add Comment