Learning from Evolutionary Trajectories to Predict Variant Effects and Generate Realistic Biomolecules.
Event Details
- Type
- Center for Studies in Physics and Biology Seminars
- Speaker(s)
-
Yun S. Song, Ph.D., professor of electrical engineering and computer sciences, statistics, director, center for computational biology, University of California, Berkeley
- Speaker bio(s)
-
Accurately predicting the functional effects of genetic variants is central to modern biology, with applications ranging from disease diagnosis to understanding complex trait architecture and engineering novel biomolecules. In this talk, I will present my lab's recent work on developing evolution-informed language models for biological sequences that capture the complex constraints shaping natural variation. I will showcase two frameworks: one that learns from genomic evolutionary patterns to predict non-coding variant effects across the human genome, and another that explicitly models evolutionary trajectories to enable realistic sequence generation. Through applications in human genetics and protein engineering, I will demonstrate how grounding AI models in evolutionary principles yields both improved predictive accuracy and biologically meaningful sequence generation, offering new tools for variant interpretation and biomolecular design.
- Open to
- Public
- Reception
- Refreshments, 3:30 p.m. - 4:00 p.m., Lower Level Greenberg Building (CRC)
- Phone
- (212) 327-8636
- Sponsor
-
Melanie Lee
(212) 327-8636
leem@rockefeller.edu