Researcher wins $2.6M NIH grant for machine-learning tools aimed at molecular interactions
Cells are more than just bags of molecules. They are complex and dynamic environments with frequent molecular interactions orchestrating life’s most basic events. Many researchers investigate molecules by monitoring them outside their natural environments, sometimes in crystals, or simulating them on a computer. However, to better understand molecular events in their natural environment, for example how DNA is transcribed to RNA, we need to indirectly observe them in action.
Steve Pressé, a faculty member of Arizona State University's Center for Biological Physics and the School of Molecular Sciences, along with a team of collaborators around the world that includes Marcia Levitus and Douglas Shepherd at ASU, is unraveling life’s intracellular processes at single-molecule levels on the rapid time scales and small spatial scales at which reactions and interactions occur.
Pressé was recently awarded a five-year, $2.6 million grant from the National Institutes of Health to help develop the machine-learning tools needed to observe life’s events in their natural environment.
“Every disease and drug taken operates at the level of the molecular actors within the cell,” Pressé said. “Accurately monitoring molecular interactions in living systems is a key step toward evaluating the role of therapeutic agents and developing clear disease diagnostics, as well as refining our understanding of how living systems work. To do this, we have to adapt machine learning tools to capture molecules doing what they do in a noisy and crowded biological system uniquely sensitive to our probing tools. No biological system likes to be poked and prodded for too long with strong lasers. Therefore, we need to make the most of the very limited information we can gather quickly.”
Pressé, the principal investigator of the project, proposes means to unravel molecular interactions by collecting information carried by individual photons emitted by the molecular system and developing data-efficient Bayesian machine learning tools to make the most of the limited light budget available.
“It is analogous to studying light emitted by faraway stars using Earth-bound telescopes where light passes through our atmosphere, which scrambles the signal. Except imagine the stars being timid and moving in space!” Pressé said. “We are working at the absolute limit of what is achievable to learn from the limited data we gather before the system starts turning its figurative back to us.”
Indeed, studying the behavior of molecules deeply embedded in complex biological environments is critical in unraveling a molecular basis for disease.
School of Molecular Sciences Director Tijana Rajh commented on the significance of this grant and Pressé’s research.
“We still have to learn a lot about the way individual cells process information and make decisions in response to environmental perturbations. These flexible decisions are made with the highly stochastic tools cells have available in tuning their response," she said. "Following Bayesian approaches, Steve Pressé and his collaborators infer from data how cells ultimately decide their fates, for example by learning transcription kinetics and gene states from noisy single-cell RNA counts. This data-driven approach is unavoidable given the information cells are willing to reveal to us through experiments and holds the promise of teaching us the rules of integrated sensing and coordinated response to varying environments in which these cells exist.”