Integrating various Experimental Information to Assist Protein Complex Structure Prediction by GRASP

Protein complex structures are essential for understanding of biological activities and drug development. Improving complex structure prediction accuracy of AI models for cases such as antigen-antibody complexes is expected to further enhance their applicability. Meanwhile, a large variety of experimental methods are used to provide structural insights for protein complexes, with only sparse or approximate knowledge obtained. A general tool is needed to integrate AI models with limited experimental information for high-throughput and accurate protein complex structure prediction. To efficiently and flexibly incorporate the different forms of experimental information, we introduce here GRASP. GRASP outperforms existing tools in handling both simulated and real-world experimental restraints including those obtained from XL, CL, CSP, and DMS. As an example, GRASP excels in predicting antigen-antibody complex structures, even surpassing AF3 when utilizing experimental DMS and CL restraints. In addition to accelerating the restrained modeling process, its ability to integrate multiple forms of restraints makes it capable of integrative modeling. We also showcase its potential in modeling protein structural interactome in the near-cellular condition based on large-scale in vivo XL data for mitchondria.