Generative artificial intelligence (AI) models often create hallucinations or fabricate information that is not factual or cannot be cited from the source material. This behavior is usually a weakness, especially given the rise of AI-generated misinformation. But in the world of bacteria, hallucinations are helping researchers discover new life-saving drugs.
Researchers at Stanford Medicine and McMaster University have developed an AI model that uncovers possible solutions to deadly antibiotic-resistant bacteria.
Related article: Google and MIT launch free generative AI course for teachers
The model, called SyntheMol (short for Synthetic Molecule), is a combination of six novel drugs aimed at killing resistant strains of Acinetobacter baumannii, one of the leading pathogens responsible for deaths related to antimicrobial resistance. The research report states that they have created “the structure and chemical recipe for this.”
Studies estimate that approximately 5 million deaths are associated with antimicrobial resistance (AMR) worldwide each year. “There is a huge public health need to rapidly develop new antibiotics,” said James Zou, associate professor of biomedical data science and co-author of the study.
“Our hypothesis was that there are many molecules out there that have the potential to be effective drugs, but none have been manufactured or tested yet. “We wanted to design a completely new molecule that had never been seen in nature.” ”
Related article: How a new law protects your ideas from tech companies – and why it matters
The number of possible compounds is exponential. Before models like SyntheMol, researchers still used algorithms to classify drug libraries to find potential solutions, but they could only do so at a fraction of the speed and scale needed. was. SyntheMol's computing power and the fact that it causes hallucinations has enabled researchers to explore his solutions to AMR with new efficiency.
“This AI is actually designing and teaching us a whole new part of chemical space that humans have never explored before,” Zou said.
The researchers trained SyntheMol on a library of “molecular building blocks” and chemical reactions. It included guidelines on which chemicals are currently effective against Acinetobacter baumannii. According to Stanford University, the model “generated approximately 25,000 antibiotic candidates and recipes for creating them within nine hours.”
Also: Agile development can unlock the power of generative AI – here's how
Initially, SyntheMol was a little too imaginative, creating compounds that couldn't realistically exist, so researchers added guardrails. The results are more realistic. To ensure that bacteria did not become resistant to these new recipes, the researchers filtered out compounds similar to currently effective antibiotics.
“We now have not only completely new molecules, but also clear instructions on how to make those molecules,” Zou said.
The researchers narrowed down SyntheMol's proposed compounds based on feasibility. In the laboratory of the chemical company Enamine he was able to create 58 different compounds. Six were able to kill resistant bacteria in the test, and two progressed to the testing stage in mice.
The new compound has also shown promise in the fight against other infectious bacteria that can exhibit antibiotic resistance, such as E. coli, MRSA, and bacteria that can cause meningitis and pneumonia. Ta. The researchers are now fine-tuning SyntheMol and collaborating with other teams to see if the model can also be used to discover potential heart disease drugs.