Improving AlphaFold 3 structural modeling by incorporating explicit crosslinks
Kosinski, J.
Show abstract
AlphaFold 3 has significantly advanced the modeling of macromolecular structures, including proteins, DNA, RNA, and their interactions with small molecules or post-translational modifications. However, challenges remain when modeling specific structural conformations or complexes with limited evolutionary data, such as protein-antibody complexes. Previous studies with AlphaFold2 demonstrated that adding distance restraints from crosslinking mass spectrometry (XL-MS) can improve predictions for such cases. In this study, we investigate whether XL-MS restraints can be incorporated into AlphaFold 3 by explicitly modeling crosslinks as covalently-bound ligands. Our results show that this approach is able to increase the accuracy of AlphaFold 3 models. We explore the opportunities and limitations of this method, which has been implemented as a proof-of-concept pipeline named AF3x, available at https://github.com/KosinskiLab/af3x.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Local Disordered Region Sampling (LDRS) for Ensemble Modeling of Proteins with Experimentally Undetermined or Low Confidence Prediction Segments 96%
- Evaluation of AlphaFold-Multimer prediction on multi-chain protein complexes 95%
- A3D Database: Structure-based Protein Aggregation Predictions for the Human Proteome 95%
Similar papers in this journal
- Combining Information from Crosslinks and Monolinks in the Modelling of Protein Structures 97%
- Assessing PDB Macromolecular Crystal Structure Confidence at the Individual Amino Acid Residue Level 96%
- Protein language model embeddings for fast, accurate, alignment-free protein structure prediction 95%
Similar papers in this journal
- Assessing scoring metrics for AlphaFold2 and AlphaFold3 protein complex predictions 96%
- CaBLAM for CryoEM Chiropraxis, UnDowser to Rethink \"Waters\", and NGL Viewer to Recapture Online 3D Graphics in MolProbity Validation 96%
- De novo protein design by inversion of the AlphaFold structure prediction network 95%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.