AI Identifies Existing Drugs to Combat Resistant Bacteria

Artificial intelligence has identified a set of existing drugs that could be repurposed to fight a stubborn and common infection-causing bacterium. This approach offers a practical alternative to the slow and costly process of developing entirely new antibiotics, which is often outpaced by the rapid evolution of bacterial resistance.

The study, reported by GN technics/ai (en-US), highlights how computational tools can screen thousands of known compounds to find those with hidden antimicrobial properties. By leveraging drugs that are already approved for other medical conditions, researchers can bypass years of safety testing, allowing for quicker deployment against threats like pneumonia and meningitis.

Targeting a dangerous respiratory pathogen

The specific focus of this research is Streptococcus pneumoniae, a bacterium that the World Health Organization has flagged as a high-priority target for new antibiotic development. It is a leading cause of common respiratory infections and is responsible for severe, life-threatening conditions such as pneumonia and meningitis. As this pathogen develops resistance to standard treatments through genetic modifications, the need for new therapeutic options has become urgent.

The challenge with S. pneumoniae is its ability to acquire resistance genes and alter its own core genes to evade existing drugs. This creates a growing gap between the infections patients face and the effective treatments available to doctors, necessitating innovative strategies to expand the arsenal of available medicines.

Screening thousands of known compounds

Researchers from Imperial College London used a dataset of compounds with known activity against this bacterium and similar pathogens to train three distinct AI models. These models were then used to evaluate nearly seven thousand small-molecule drugs for their potential to inhibit the growth of S. pneumoniae. The system prioritized candidates that showed high consensus across the different models, ensuring that the most likely effective compounds were selected for further review.

From this large pool of predictions, the team narrowed the list to eleven compounds for laboratory testing. They chose these specific molecules not only because of high model scores but also because their chemical structures were distinct from the molecules used to train the AI. This diversity was crucial to ensure that the AI was identifying new mechanisms of action rather than simply recognizing patterns it had already learned.

Laboratory results show promising efficacy

In vitro testing revealed that nine of the eleven selected compounds successfully inhibited the growth of the bacterium. Notably, one of the two most potent candidates demonstrated effectiveness against drug-resistant strains of S. pneumoniae. This result is significant because it suggests that AI can identify treatments that work specifically against the harder-to-treat variants of the pathogen.

The primary trade-off in this approach is that while repurposing saves time and reduces financial risk, it relies on the existing safety profiles of the drugs for their original indications. However, the study concludes that this method can serve as a bridge, providing short-term treatment options while also guiding the longer-term development of entirely new classes of antimicrobials.