AI paves the way for new, safe, and effective antibiotics to fight resistant bacteria

In a promising development for the demand for safer and more effective antibiotics, researchers at The University of Texas at Austin have utilized artificial intelligence to create a new drug that is already showing positive results in animal trials.

Published in Nature Biomedical Engineering, the scientists detail how they used a large language model—an AI tool similar to the one behind ChatGPT—to redesign a previously toxic bacteria-killing drug to make it safe for human use.

The outlook for patients with severe bacterial infections has worsened in recent years due to the spread of antibiotic-resistant strains and the stagnation in developing new treatments. However, UT researchers believe AI tools are transformative.

“We have found that large language models represent a significant advancement for machine learning applications in protein and peptide engineering,” said Claus Wilke, professor of integrative biology and statistics and data sciences, and co-senior author of the new paper. “Many applications that were not feasible with previous methods are now becoming possible. I anticipate that these and similar approaches will be widely used for developing therapeutics or drugs in the future.”

Large language models (LLMs) were initially designed to generate and analyze text sequences, but scientists are now creatively applying these models to other fields. For instance, just as sentences are composed of word sequences, proteins are composed of amino acid sequences. LLMs group words with common attributes (like cat, dog, and hamster) in an “embedding space” with thousands of dimensions. Similarly, proteins with similar functions, such as fighting off harmful bacteria without harming their human hosts, may cluster together in their own AI embedding space.

“The space containing all molecules is vast,” said Bryan Davies, co-senior author of the new paper. “Machine learning allows us to identify the areas of chemical space with the properties we are interested in, and it can do so much more quickly and thoroughly than traditional one-at-a-time lab methods.”

For this project, the researchers used AI to find ways to reengineer an existing antibiotic called Protegrin-1, which is excellent at killing bacteria but toxic to humans. Protegrin-1, naturally produced by pigs to combat infections, belongs to a subtype of antibiotics known as antimicrobial peptides (AMPs). AMPs typically kill bacteria by disrupting cell membranes, but many target both bacterial and human cell membranes.

Initially, the researchers employed a high-throughput method they had previously developed to create over 7,000 variations of Protegrin-1 and quickly identify regions of the AMP that could be modified without losing its antibiotic activity.

Next, they trained a protein LLM on these results, enabling the model to evaluate millions of possible variations for three features: selectively targeting bacterial membranes, effectively killing bacteria, and not harming human red blood cells. The model then guided the team to a safer, more effective version of Protegrin-1, which they named bacterially selective Protegrin-1.2 (bsPG-1.2).

Mice infected with multidrug-resistant bacteria and treated with bsPG-1.2 were significantly less likely to have detectable bacteria in their organs six hours after infection compared to untreated mice. If further testing yields similarly positive results, the researchers hope to eventually advance the AI-informed antibiotic drug to human trials.

“Machine learning’s impact is twofold,” Davies said. “It will identify new molecules with potential benefits for people and show us how to improve existing antibiotic molecules, allowing us to focus our efforts and expedite the development process.”

Original source: https://www.sciencedaily.com/releases/2024/07/240731141014.htm