Enhanced Protein Complex Prediction via Rosetta, AlphaFold, and Nondifferential Covalent Labeling Mass Spectrometry

Covalent labeling (CL) mass spectrometry is a versatile structural technique for elucidating structural information about proteins, including protein-protein complexes. When integrated with computational protein structure prediction, CL can generate accurate protein complex models. Structural insights from CL are more readily interpreted when differential data (e.g., monomer versus complex) are available, as changes in labeling can directly pinpoint interface residues. However, isolating and labeling monomeric subunits can be experimentally challenging, leading to the common scenario in which labeling data are available only in the complex form. However, importantly, nondifferential CL still encodes useful structural information that has yet to be utilized for automated protein complex prediction. In this work, we present a framework for using nondifferential CL in protein complex prediction. We introduced a new hydroxyl radical protein footprinting (HRPF)-derived scoring term that penalizes docked models based on their agreement with experimentally measured CL data. In a benchmark set, the best-scoring models showed an average root-mean-square deviation (RMSD) improvement of 4.6 Å when HRPF data was used. With the inclusion of CL data, four of the top-scoring complexes exhibited RMSDs below 5Å, whereas none did without it. This study demonstrates that CL can enhance protein complex prediction, even when only bound-state CL measurements are available.