A vexing bacterial infection
Over 450,000 Americans each year will contract Lyme disease from a tick bite. Ticks can carry the bacterium Borrelia burgdorferi. Known as a spirochaete (that is, a bacterium that is shaped and moves like a corkscrew), B. burgdorferi can live in the tick species Ixodes scapularis.
When an affected tick affixes itself to a human and feeds on their blood, the bacterium can enter the human’s bloodstream. Infection with B. burgdorferi causes initial flu-like symptoms and can lead to organ damage in its most severe form.
People most susceptible to tick bites tend to live in rural areas, often far from the medical care necessary to treat Lyme disease. While Lyme disease is the most common vector-borne illness in the United States, there is still much to learn about its mechanism of action: that is, how the bacterium actually disseminates through the body once it gains access to a human host.
While most Lyme disease patients respond well to traditional anti-bacterial treatments, some develop prolonged and severe symptoms, especially when diagnosis is delayed or inaccurate.
In previous work, Sinnott-Armstrong worked on a team that identified a human protein that appears to offer resistance to B. burgdorferi. Known as SCGB1D2 (for Secretoglobin family 1D member 2), it is believed to be a defensive protein present in sweat and other bodily secretions that protects against B. burgdorferi infection.
Combining AI and in vitro science to accelerate interventions
Sinnott-Armstrong, who holds degrees in archeology and computer science in addition to a PhD in genetics from Stanford University, devised an AI platform to evaluate potential therapeutic interventions for Lyme disease in concert with ongoing laboratory experiments.
Known as OLIVE (Online Learning of In Vitro Experiments), the program will use an active learning model to identify proteins similar to SCGB1D2 that inhibit the growth of B. burgdorferi, as well as test the effectiveness of combinations of antibiotics that could work more effectively than single drugs. Sinnott-Armstrong will also use CRISPR screens to identify genes in the bacterium that may signal its sensitivity to specific therapeutics.
Sinnott-Armstrong explained how the integration of AI into in vitro experiments (experiments performed on biological samples in the laboratory, as opposed to experiments in living organisms such as mouse models) will work.
“Instead of focusing on getting an experiment designed correctly the first time and hoping it’s right, you can use an AI system to help design the best experiment that we can, given the knowledge that we have, and then use that to inform a model which we can then plan and design and change,” Sinnott-Armstrong explained. That model can be improved over time as more data informs the active learning system.
“And so, you’re generating data and generating a model simultaneously, and by doing that cyclically, the idea is that it will be much more efficient for us to be able to test a large number of extremely complex hypotheses," they said.
Sinnott-Armstrong’s research could lead to this type of parallel active learning/laboratory experiment model being used to more effectively evaluate new drug combinations for other illnesses.
“This platform generalizes to any host-pathogen system or combination drug,” Sinnott-Armstrong wrote in their Pew application. “[This] novel computational system will enable academics and citizen scientists to develop their own therapeutics in ways previously only accessible to drug companies.”
The views expressed herein are those of the author and do not necessarily reflect the views of The Pew Charitable Trusts.
