Abstract
Recent advances in deep learning have revolutionized macromolecular structure prediction, as exemplified by the success of AlphaFold and related frameworks. Despite these breakthroughs, accurately modeling biomolecular assemblies—particularly protein–RNA complexes remains a major challenge, partly due to the scarcity of evolutionary information used as inputs in existing approaches.
In this talk, I will focus on our recent work, ProRNA3D-single, a novel deep learning framework for predicting protein–RNA complex structures. ProRNA3D-single employs geometric attention-enabled pairing of biological language models of proteins and RNAs to predict interatomic interaction maps, which are subsequently transformed into multi-scale geometric restraints for 3D structure modeling. Benchmark results demonstrate that ProRNA3D-single outperforms state-of-the-art methods, including AlphaFold 3, particularly when evolutionary information is limited. I will conclude with a brief overview of my ongoing and future research directions.
Speaker Bio
Rahmatullah Roche, PhD. is a tenure-track Assistant Professor in the Department of Computer Science at Columbus State University. His research focuses on computational biology, applied machine learning, data science, and human-computer interaction, with a particular emphasis on macromolecular predictive modeling using advanced artificial intelligence techniques. Dr. Roche earned his Ph.D. in Computer Science from Virginia Tech in 2024. He holds a Master of Science in Computer Science and Software Engineering from Auburn University (2021) and a Bachelor of Science in Computer Science and Engineering from Bangladesh University of Engineering and Technology (BUET) (2016). He is interested in interdisciplinary collaborations to advance scientific discovery and technological innovation.