AI models flag existing drugs that could fight drug-resistant pneumonia bacteria

Researchers have used artificial intelligence to identify existing, approved drugs that may be repurposed to fight Streptococcus pneumoniae, a bacterial pathogen responsible for life-threatening infections such as pneumonia and meningitis, according to a study published by Wiley in the journal Advanced Science.

S. pneumoniae is becoming increasingly difficult to treat as the pathogen develops resistance to antibiotics, a phenomenon known as antimicrobial resistance (AMR). The World Health Organization has named S. pneumoniae one of its priority pathogens for antimicrobial resistance, alongside other drug-resistant bacteria that have already been targeted by AI-guided drug repurposing efforts, according to the study.

The research team, led by senior author Pedro J. Ballester, associate professor at Imperial College London and a Wolfson Fellow of the Royal Society in the United Kingdom, used drug repurposing — applying drugs with established safety profiles to new disease targets — as a faster, lower-risk, lower-cost alternative to developing new antibiotics from scratch.

To identify candidates, the researchers trained three distinct machine learning approaches: ensembles of decision trees, ensembles of graph neural networks, and ensembles of sequence-based transformers, each pretrained on hundreds of millions of molecules. The models were trained using a dataset of molecules already known to be active against S. pneumoniae, along with a larger dataset of molecules inactive against a different drug-resistant bacterium.

Using these three models together, the researchers screened close to 7,000 candidate drugs for their potential to inhibit S. pneumoniae. From that pool, 11 compounds were selected for laboratory validation. Nine of the 11 successfully inhibited the growth of S. pneumoniae, and one of the two most potent compounds remained effective even against strains of the bacterium that had already developed drug resistance, according to the study.

The authors noted that combining three different types of AI models, rather than relying on a single approach, improved the selection process. Each model type appeared to complement the others, making the combined screening effort more effective than any single model would have been on its own.

"AI-guided drug repurposing has become a powerful strategy for combating antimicrobial resistance, particularly for pathogens for which some active molecules are already known and can be leveraged as a training or fine-tuning dataset," Ballester said in a statement.

The findings add to a growing body of evidence supporting computational drug repurposing as a viable strategy in the fight against antimicrobial resistance, a problem the researchers describe as an urgent global health threat with significant costs tied to health care spending and mortality. The study's authors say the approach offers a streamlined, affordable path to identifying effective compounds at a time when new treatment options are urgently needed.

AI's growing role in the fight against drug-resistant infections

The S. pneumoniae findings arrive amid a broader push across the pharmaceutical and research sectors to apply AI-driven screening methods to the antimicrobial resistance crisis. Traditional antibiotic development has slowed considerably in recent decades, in part because the economics of bringing a new antibiotic to market often don't favor drugmakers compared with therapies for chronic conditions.

Drug repurposing has emerged as one response to this gap, allowing researchers to bypass much of the early-stage safety testing required for entirely new compounds. AI and machine learning tools have accelerated this process by allowing scientists to screen thousands of candidate molecules computationally rather than relying solely on slower laboratory-based testing.

For physicians treating patients with drug-resistant infections, the practical impact of this research is still likely years away — repurposed candidates identified through AI screening still require clinical trials before they can be prescribed. But the growing use of AI to narrow the field of promising candidates could help shorten the overall timeline between laboratory discovery and bedside availability, particularly for pathogens on the WHO's priority list where treatment options are becoming increasingly limited. As antimicrobial resistance continues to strain hospital budgets and patient outcomes alike, physicians may see more of these AI-assisted repurposing studies move toward clinical testing in the coming years.