Professor Roland Dunbrack

Talk: Structural bioinformatics and AlphaFold modeling of the human kinome and its interactions

Protein kinases sit at the heart of cellular signaling, and their dysregulation drives many human diseases. The human kinome comprises 480 genes containing one or more “typical” protein kinase domains, including both catalytically active and pseudokinases. In this keynote, I will describe an integrated framework that combines structural bioinformatics of experimental structures with large-scale AlphaFold modeling to define, model, and interpret the conformational and interaction space of human protein kinases.

First, leveraging a set of 54 unique substrate-bound kinase complexes, we derived structural criteria that define the active form of a protein kinase. Using these criteria, we modeled all 437 catalytically active human kinase domains in their active, substrate-competent conformations with AlphaFold2. Second, by clustering activation-loop conformations from experimentally determined inactive structures, we identified recurrent, biologically relevant inactive states for dozens of kinases and used these to model kinases in their most informative inactive forms—providing mechanistic hypotheses for regulation and inhibitor susceptibility.

Third, extending beyond isolated kinase domains, we applied AlphaFold modeling to all 480 full-length human proteins containing kinase domains, identifying and naming folded domains across these multidomain proteins. This domain-resolved view enables hypothesis generation for intramolecular autoinhibition, allosteric coupling, and partner recognition. Finally, I will present a new interaction-scoring approach based on AlphaFold’s predicted aligned error (PAE) matrix (github.com/DunbrackLab/IPSAE) and show how it supports modeling of kinase protein–protein interactions spanning activating, inhibiting, and substrate-binding assemblies. Together, these advances provide a structurally grounded, proteome-scale platform for understanding kinase regulation and for guiding experiments and therapeutic discovery.

About this speaker

Roland L. Dunbrack, Jr. is a Professor at Fox Chase Cancer Center/Temple University, where he directs the Molecular Modeling Facility and co-leads the Cancer Signaling and Microenvironment Program. Trained as a computational structural biologist (A.B., Chemistry, Harvard College; Ph.D., Biophysics, Harvard University; postdoctoral work at UCSF), he has been on the Fox Chase faculty since 1997.

Dunbrack’s research develops data-driven methods to understand and predict the three-dimensional structures of proteins—work that underpins modern protein engineering. He created the backbone-dependent rotamer library, a widely used resource for building and designing accurate protein models, and his group develops algorithms and databases that classify protein conformations and support modeling of key signaling proteins, including protein kinases and antibodies. By combining statistical analysis of structural data with structure prediction and modeling, his lab helps connect molecular structure to function in cancer biology and beyond.

He has twice served as an assessor in the CASP protein structure prediction experiments and has received honors including the ASBMB DeLano Award for Computational Biosciences.