Progress report of Artificial intelligence (AI) in protein structure prediction and design

Abstract

Many cancer-driver or drug-resistant mutant proteins (e.g., KRASG12X or EGFRC797S) currently lack effective therapeutic options. If the 3D structures of these mutant proteins can be accurately determined, it may become possible to rationally design corresponding therapeutic antibodies or other structure-guided treatments.
The most widely used model for protein structure prediction is AlphaFold, which has contributed more than 214 million predicted protein sequences to the Protein Data Bank (PDB). However, the accuracy and biological relevance of many AlphaFold-predicted structures have not been experimentally validated. For de novo protein structure prediction, Rosetta remains one of the most powerful computational tools, yet the biological functions and in vivo efficacy of antibodies designed by Rosetta are still insufficiently verified. To address these gaps, our research group aims to improve the accuracy and functional reliability of AlphaFold and Rosetta predictions through several complementary strategies, including database comparison and refinement, cross-validation using cryo-EM–derived experimental structures, and biological functional assays to validate predicted effects.
In this group report, we present preliminary results on the predicted structures of key cancer-driver EGFR mutations (e.g., EGFRL858R, EGFRL858R/T790M, and EGFRL858R/T790M/C797S). Because cryo-EM studies require highly purified and concentrated protein samples, we are currently optimizing protein purification workflows across multiple cell lines expressing different EGFR mutation statuses. We anticipate that by integrating our curated cancer mutation database with a cross-validation framework combining cryo-EM structural data and biological experiments, we will be able to refine AlphaFold and Rosetta predictions. This integrated approach has the potential to create a novel, accurate, and high-impact drug-design platform, capable of guiding therapeutic development for mutant proteins that currently lack effective treatments.

線上視訊請點選連結