Eric Pederson - From Contact Maps to Coordinates: Deep Learning of Crosslinking–Distance Relationships for RNA Structure Prediction

Title: From Contact Maps to Coordinates: Deep Learning of Crosslinking–Distance Relationships for RNA Structure Prediction

Abstract: Computational RNA structure prediction has attracted increasing attention following the success of AlphaFold for protein structure prediction. Many current methods rely on databases of RNA structures determined using established biophysical techniques, including cryogenic electron microscopy, X-ray crystallography, and nuclear magnetic resonance spectroscopy. However, these techniques face method-specific limitations related to RNA size, conformational dynamics, and achievable resolution, constraining the structural diversity represented in existing databases. Proximity-based RNA crosslinking coupled with high-throughput sequencing offers a complementary source of structural information by producing an N×N matrix of pairwise contact counts for an RNA of length N. Translating these 2D contact counts into a 3D spatial representation, however, remains a challenge. Here, we use neural networks to learn the nonlinear relationship between crosslinking contact counts and physical distances while quantifying uncertainty in the resulting predictions. The inferred distance distributions approximate the conformational states sampled by the crosslinking experiments and can be incorporated as pairwise restraints during three-dimensional structure prediction. This framework provides a route for integrating sequencing-derived contact information with computational modeling to improve sampling of experimentally supported RNA conformations.

Short Bio: Eric Pederson joined Zhengqing Ouyang’s Lab in the Biostatistics and Epidemiology Department at UMass Amherst as a Postdoctoral Research Associate in 2020. He received his B.S. in Biological Sciences from Carnegie Mellon University in 2015 and earned a PhD in Bioengineering from the University of Washington in 2019 for his work studying inflammation-related blood clot damage with molecular dynamics simulations. His current research integrates transcriptomic and biophysical approaches to study RNA structure and dynamics, ribosome biogenesis, and their connections to diseases such as cancer.