In an encouraging development for the search for safe and effective antibiotics for human use, researchers have turned to artificial intelligence to create a new drug that is showing promise in animal testing.
In a promising development for the need for effective and safe antibiotics for humans, researchers at The University of Texas at Austin have used artificial intelligence to create a new drug that has already demonstrated potential in animal trials.
In their findings published in Nature Biomedical Engineering, the scientists discuss how they utilized a large language model—similar to the technology behind ChatGPT—to modify a previously harmful bacteria-fighting drug, making it safe for human use.
The outlook for patients facing severe bacterial infections has deteriorated recently due to the rise of antibiotic-resistant bacterial strains and a slowdown in the emergence of new treatment options. Nonetheless, the researchers from UT believe that AI tools could be groundbreaking.
“We have discovered that large language models offer a significant advancement for machine learning in protein and peptide engineering,” stated Claus Wilke, a professor specializing in integrative biology as well as statistics and data sciences, and co-senior author of the study. “Many applications that were previously unworkable with older methods are now beginning to show results. I anticipate that these techniques, along with similar innovations, will be widely adopted for creating therapeutics and medications in the future.”
Originally intended for generating and exploring text sequences, large language models (LLMs) are being creatively repurposed by scientists. Just as phrases consist of sequences of words, proteins are made up of sequences of amino acids. LLMs group words that display similar characteristics (e.g., cat, dog, hamster) into what is known as an “embedding space,” which comprises thousands of dimensions. Likewise, proteins that have comparable functions—like fighting harmful bacteria while sparing human cells—may cluster in their own AI-generated embedding space.
“The chemical landscape is vast,” explained Davies, co-senior author of the research. “Machine learning enables us to swiftly and comprehensively discover regions of this landscape that possess the desired characteristics, much faster and more thoroughly than traditional laboratory methods that analyze one molecule at a time.”
For this initiative, the researchers used AI to explore potential modifications to an existing antibiotic known as Protegrin-1, which is effective at killing bacteria but poses toxicity risks to humans. Protegrin-1, which is naturally produced by pigs to fend off infections, belongs to a category of antibiotics called antimicrobial peptides (AMPs). Typically, AMPs directly kill bacteria by disrupting the integrity of their membranes, but many also adversely affect human cell membranes.
Initially, the researchers implemented a high-throughput technique they had previously developed to produce over 7,000 variations of Protegrin-1, enabling them to swiftly identify which parts of the AMP could be altered without compromising its antibacterial properties.
They then trained a protein LLM on these findings, allowing the model to assess millions of potential variations based on three criteria: selectively targeting bacterial membranes, effectively killing bacteria, and not harming human red blood cells. This process led them to a safer and more effective variant of Protegrin-1, named bacterially selective Protegrin-1.2 (bsPG-1.2).
Mice that were infected with multidrug-resistant bacteria and treated with bsPG-1.2 showed significantly lower bacteria levels in their organs six hours post-infection compared to untreated mice. Should further research yield positive outcomes, the team hopes to advance this AI-assisted antibiotic into human trials.
“Machine learning contributes in two significant ways,” said Davies. “It will identify new molecules that could potentially benefit people and will help us enhance existing antibiotic compounds, streamlining our efforts to bring these innovations into clinical use more rapidly.”
This research underscores the commitment of academic scientists to advance artificial intelligence in ways that address societal needs—an important focus for UT Austin, which has proclaimed 2024 as the Year of AI.
Other contributors to the study include research associate Justin Randall and graduate student Luiz Vieira, both affiliated with UT Austin.
Research funding was supported by the National Institutes of Health, The Welch Foundation, the Defense Threat Reduction Agency, and Tito’s Handmade Vodka.
