Artificial intelligence can help accelerate the identification of medicines capable of combating bacteria resistant to antibiotics, according to an international study published in the journal Advanced Science. The researchers used three different AI models to search for already approved drugs that could be used against Streptococcus pneumoniae, a bacterium responsible for potentially serious infections such as pneumonia and meningitis.
Antibiotic resistance has been making the treatment of these infections more difficult as the bacterium acquires alterations that reduce the efficacy of existing medicines. Facing this problem, the researchers turned to drug repurposing, which involves testing already developed drugs with known safety profiles for new uses.
The team trained three types of machine learning algorithms: ensembles of decision trees, graph neural networks, and sequence-based transformer models. The models were then used to analyze nearly 7,000 drugs and select those with the greatest potential to inhibit the growth of S. pneumoniae.
Among the thousands of candidates analyzed, 11 were chosen for experimental validation. Nine demonstrated the ability to inhibit the growth of the bacterium. One of the two repurposed drugs that showed the greatest efficacy also managed to act against strains of S. pneumoniae that are resistant to medicines.
Three models to select the candidates
According to the researchers, using different AI models allowed combining the advantages of each approach and improving the selection of candidates. The results suggest that the combined use of all three models was more effective than using any one of them individually.
The strategy could be particularly relevant in the context of antimicrobial resistance, as it allows laboratory testing to be concentrated on a smaller number of drugs with greater potential, rather than testing thousands of compounds individually.
“AI-driven drug repurposing has become a powerful strategy to combat antimicrobial resistance, particularly for pathogens for which some active molecules are already known and can be used as datasets for training or tuning the models,” says Pedro J. Ballester, senior author of the study, associate professor at Imperial College London and Wolfson Fellow of the Royal Society, in the United Kingdom.
The authors consider that AI-supported drug repurposing can reduce the time and costs associated with the search for new treatments. However, the candidates identified by AI require experimental validation and, if they prove promising, further investigation before they can be used clinically.
