KnotFold: Improving Peptide Structure Predictions with Simulated Coevolution

We present a novel approach for embedding experimental contact information in the protein structure prediction tool AlphaFold2 that harnesses the principles of protein sequence coevolution to enforce disulfide connectivity patterns in the prediction of disulfide-rich peptide (DRP) structures. While AlphaFold2 generates accurate DRP structure prediction in most cases, it sometimes fails at predicting the specific connectivity pattern of the multiple disulfide bonds. Here, we take advantage of the principles of sequence coevolution to engineer the multiple sequence alignment (MSA) input with specific connectivity patterns by mutating highly conserved cysteines in subsets of the MSA—which we term “simulated coevolution.” Engineering the MSA prompt with simulated coevolution provides sufficient context in most cases for AlphaFold2 to obey the provided connectivity. Our approach—KnotFold—can be used to incorporate experimental disulfide connectivity patterns from mass spectrometry into DRP structure prediction. Lastly, we find that peptide properties and AF2Rank Composite Confidence scores can be used to differentiate native from non-native connectivity patterns in some cases, improving the ability to determine native connectivity of DRPs from sequence alone. Published by the American Physical Society 2024