Artificial intelligence (AI) models with generative capabilities have a tendency to produce fictional or imaginary information, which is often referred to as hallucinations. While this is generally considered a flaw, it has been discovered that these hallucinations can be useful for scientists in the field of drug discovery. Similar to how the artificial diamond was unintentionally invented while attempting to recreate the extreme conditions found in the earth’s mantle, hallucinations can lead to the discovery of new drugs.
According to experts, antibiotic resistance is responsible for approximately 5 million deaths worldwide. Therefore, finding new ways to combat antibiotic-resistant bacteria is crucial and urgent. Researchers from McMaster University and Stanford Medicine School have developed a new model called SyntheMol, which aims to identify potential solutions for antibiotic-resistant bacteria.
James Zou, associate professor of biomedical data science and co-author of the study, emphasizes the need for rapid development of new antibiotics for public health. The researchers have experimentally validated the new compounds generated by the SyntheMol model.
Their hypothesis is that there are numerous potential molecules that could be transformed into effective drugs, but these molecules have not been tested or developed yet. This is why they are utilizing AI to generate molecules that do not naturally exist.
Prior to the use of generative AI, researchers employed computational approaches to develop antibiotics. They used algorithms to search through databases of known drugs and identify compounds that had the potential to combat specific pathogens. While this method yielded results, it was not an exhaustive process for identifying all possible chemical compounds that could be effective against bacteria.
Kyle Swanson, the co-lead author of the study and a doctoral student at Stanford, explains that the chemical space is vast. The hallucinating tendency of AI can be harnessed for the discovery of new drugs by developing compounds that were previously unimaginable. The researchers had to establish boundaries for the model to artificially generate molecules that it envisioned.
Zou states that this model has provided insights into a previously unexplored area of the chemical field by designing new molecules that were unknown to humans. Together with Swanson, Zou is refining the model to explore its applications in heart drugs and the creation of fluorescent molecules with novel properties for laboratory research.
For more information on Stanford University’s research in this area, refer to their official statement.
