Tufts University researchers are using AI and machine learning to more quickly identify narrow-spectrum antibiotics that may be used to prevent and treat Lyme disease. This was thanks to generous gifts from alumni, which accelerated the research schedule by two to three years.
This research has already identified hundreds of unique compounds that can kill people. Borrelia burgdorferieliminates the bacteria that causes Lyme disease and leaves other bacteria unaffected. The development of such narrow-spectrum antibiotics is essential to prevent drugs that can kill Lyme bacteria from also killing beneficial bacteria in the body’s microbiome or causing drug resistance.
This gift enables researchers to expand their research with significant additional funding from the National Institutes of Health (NIH) and private foundations.
Without anonymous donations, these efforts might not have gotten off the ground or might have moved more slowly. The NIH does not tend to fund pilot projects. The NIH wants ideas for which there is already data that suggests greater investment will lead to success. This funding allowed us to obtain a proof of concept. ”
Linden Hu, Paul and Elaine Cherbinski Professor of Immunology at Tufts University School of Medicine and co-director of the Tufts Lyme Disease Initiative
Lyme disease, caused by deer tick bites, affects approximately 475,000 people each year in the United States, primarily in the Northeast and Mid-Atlantic states, the Upper Midwest, Northern California, Oregon, and the West Coast of Washington. Most cases can be successfully treated with antibiotics, but about 10 to 20 percent of infected people develop symptoms such as fatigue, “brain fog,” and joint or muscle pain that persist for months or years after antibiotic treatment ends.
Researchers are exploring ways to prevent Lyme disease and post-treatment Lyme disease syndrome (PTLDS). One such avenue is identifying new types of antibiotics that people living in areas where the disease is endemic can take prophylactically, in much the same way that anti-malarial drugs are taken by people traveling to and living in areas where malaria is endemic.
The goal is to find narrow-spectrum antibiotics that are lethal. B. Burgdorferi However, it does not kill other common harmful bacteria. Escherichia coli or Staphylococcus aureus, It does not affect the “good” bacteria that are part of our normal flora. Broad-spectrum antibiotics that kill multiple microorganisms run the risk of developing drug resistance if large numbers of people take them for long periods of time.
One effort led by Fu and Maha Farhat, an associate professor of bioinformatics at Harvard University, screened 60,000 existing compounds that can kill Lyme disease bacteria. Those that looked promising were then back-tested to ensure only what killed them. B. Burgdorferi.
This traditional compound screening process is expensive and requires putting each molecule into a test tube or well with a bacterial sample to see if anything kills the bacteria. Screening of 60,000 compounds resulted in hundreds of compounds effective against B. burgdorferi, but only a handful of these prove worth pursuing.
Development of new drugs
Based on these initial screens, the researchers developed an AI model that can probe an estimated 1 X 10 wide area.60 Extract compounds in the drug-like chemical space (1 followed by 60 zeros) more quickly, efficiently, and at a much lower cost to identify additional potential compounds likely to be effective against Lyme bacteria.
Additionally, Farhat’s team is using that information to design compounds that are supposed to be active against Lyme bacteria (and only Lyme bacteria), based on the initial screen. They do this by building a generative model that can “imagine” all kinds of molecules that can kill Lyme bacteria.
This allowed the team to develop an artificial compound with even more advantageous qualities. For example, it seems to kill Lyme bacteria more effectively, is less toxic, and can be taken orally.
Although the AI process is predictive and may not always be correct, it can help narrow down the field of compounds worth further testing more quickly and at lower cost, Hu says.
From proof of concept to further funding
Hu and Farhat, along with Bree Aldridge, professor of medicine and engineering, and Trevor Smith II, assistant professor of medicine, recently received funding from an NIH grant to further refine their research, screen more compounds, and design more compounds that could potentially be used to prevent and treat Lyme disease. They will also see if they can figure out how and why drugs that kill the Lyme disease bacteria attack Lyme disease, and Lyme disease alone.
Aldridge’s research has previously focused on Mycobacterium tuberculosis and has included developing an AI-powered tool called DECIPHAER to help pinpoint how antibiotics kill bacteria. DECIPHAER plays an important role in new Lyme research.
“Compounds can kill bacterial cells by targeting a variety of key cellular functions, including the cell wall, proteins, and the ability to produce energy,” Professor Aldridge explains. “We want to understand how each compound actually kills bacteria, and we can do that using something called morphological profiling, which is taking pictures of cells after treatment with a compound and looking at how they break apart.”
“It’s a ‘guilt by association’ algorithm,” she says. “Therefore, if a new compound causes cells to collapse or die in the same physical way that cells die with cell wall agents, we assume that the new compound is also a cell wall agent. Although we are not the first team to use this idea, we are excited to be able to introduce a multi-omics approach with DECIPHAER to learn more specific mechanistic details.”
DECIPHAER connects these details to reveal more precise insights into how compounds are affecting cells and why bacteria are dying, predicting the molecular effects of compounds from images alone, and revealing how compounds work in different conditions or in different combinations.
This information could be used to identify and design more effective drugs to prevent and treat Lyme disease.
Weaknesses of the genome
Donor support will also help Peter Gwin’s research pinpoint weaknesses in the Lyme disease genome, which could be exploited by identifying which parts control critical functions of the organism and can be disrupted by existing compounds.
“If you think of the genome like a subway map, most bacteria have relatively large genomes and use multiple pathways to manage important biological functions,” explains Gwin, an assistant professor in the School of Medicine’s Department of Molecular Biology and Microbiology. “If a drug attacks and disrupts one pathway, these bacteria have backup pathways that they use to perform their critical functions, much like a large subway system has multiple ways to get from one destination to another if one route fails.”
In contrast, Lyme bacteria have small genomes and appear to have no or at best a limited number of backup pathways for important functions. “This means that if we can identify a compound that disrupts a critical pathway, the bacteria has no backup and we can kill the Lyme bacteria,” Gwinn added.
Donors’ gifts paid for the screening of thousands of compounds to identify a small number of promising drug candidates that could attack some of these challenges. New funding from the Bay Area Lyme Foundation (BALF) is furthering Gwinn and his colleagues’ efforts, which will also use AI computational modeling to speed up the review process.
Significant Donors for Pilot Study
“Significant funding from individuals is very important for early-stage research like this,” Hu said. “This allows us to prove that our ideas are worthy of funding from the federal government and other foundations. This allows us to attract talent like Bree Aldridge and Maha Farhat. They may not be working in fields like Lyme disease, but the money is there and will be raised to help us.”
Because of this gift in particular, Hu estimates that the team was able to do work in six months that would otherwise take two to three years. Because researchers were able to act quickly in an era of rapid change in AI and machine learning, proposals for additional funding remained cutting-edge at the time of submission, increasing the likelihood of receiving additional funding.
“As researchers, if we cannot move as quickly as the artificial intelligence field is moving, we risk falling behind in harnessing the potential of AI,” Hu added. “Thanks to our donors, this research team has already identified several potential drug candidates, which the university hopes to patent and someday move into human trials. And we believe that the additional funding we were able to obtain will one day help us design even more effective drugs that can be used to prevent and treat Lyme disease.”
