Poster abstracts
Poster number 30 submitted by Elijah Day
Protein structure prediction with hydroxyl radical protein footprinting and deep learning
Elijah H. Day (Ohio State Biochemistry Program), Steffen Lindert (Department of Chemistry and Biochemistry, The Ohio State University)
Abstract:
While deep learning networks such as AlphaFold2 (AF2) have revolutionized protein structure prediction, challenges remain for proteins that undergo conformational changes or lack homologous structures. Here, we introduce HRFold, a modified version of AF2 that directly integrates hydroxyl radical protein footprinting (HRPF) data into its network architecture. HRPF utilizes hydroxyl radicals to selectively oxidize solvent-exposed residues, providing experimental insight into protein dynamics, binding interfaces, and regions of structural flexibility. By incorporating solvent accessibility constraints derived from HRPF data, HRFold explicitly guides residue placement during structure inference. We validate HRFold using both synthetic and experimental HRPF datasets, demonstrating its ability to outperform AF2 on challenging targets. Our results highlight the power of integrating sparse experimental data into deep learning frameworks, paving the way for more accurate and biologically relevant protein structure predictions.
Keywords: Structure Prediction, Machine Learning, Mass Spectrometry