Automated Discovery of Allosteric Networks from Cryo-EM-Derived Conformational Ensembles

Current cryo-EM heterogeneity analysis (e.g., CryoFormer, Liu et al., 2023) focuses on reconstructing continuous conformational landscapes. However, the identification of functional allosteric networks—groups of residues whose motions are correlated and may mediate long-range communication—is still mostly done by manual inspection or MD-based correlation analysis. This project proposes a systematic, data-driven approach: learning residue-residue correlation matrices and dynamic networks from large cryo-EM-derived ensembles, using graph neural networks or diffusion models. By mapping these networks onto structural models, and validating with mutagenesis or functional assays, this could automate the discovery of allosteric sites and pathways, opening new avenues for drug design and biomolecular engineering. Unlike previous studies, this approach would explicitly link cryo-EM conformational heterogeneity to functional communication within the molecule.

References:

1. CryoFormer: Continuous Heterogeneous Cryo-EM Reconstruction using Transformer-based Neural Representations. Xinhang Liu, Yan Zeng, Yifan Qin, Hao Li, Jiakai Zhang, Lan Xu, Jingyi Yu (2023).