Predict the effect of missense variants on protein–protein interactions

MutPred-PPI predicts whether a missense variant disrupts a specific protein–protein interaction from the three-dimensional structure of the complex.

Submit a prediction Help & format About

Stewart R, Laval F, Coppin G, Spirohn-Fitzgerald K, Tixhon M, Hao T, Calderwood MA, Mort M, Cooper DN, Vidal M, Radivojac P. Predicting interaction-specific protein-protein interaction perturbations by missense variants with MutPred-PPI. The 30th Annual International Conference on Research in Computational Molecular Biology, RECOMB 2026. Available on bioRxiv 2025.12.20.695738.

Submit a prediction job

Upload a protein complex structure and a list of missense variants to predict interaction disruption. Up to 100 variants per job. Input format guide →

Upload the mmCIF or PDB structure file for the protein complex. If it contains more than two protein chains you will be asked which two to score. AlphaFold3 structures can be generated at alphafoldserver.com (subject to Google DeepMind Output Terms of Use).

⇩

Drag & drop your .cif, .mmcif, or .pdb file here,
or click to browse

Optional labels used to name the two proteins in your results. Leave them blank and they are taken from your variant list if it has ID columns, otherwise from the selected chains.

Label for the protein your variants apply to.

Label for its interaction partner.

Either a plain list of variants, one per line or separated by spaces or commas, which is taken to be variants on the mutated protein; or three columns: MUTATED_PROTEIN_ID  VARIANT  PARTNER_ID (tab- or space-separated), which also lets you mutate the partner protein by listing it first.
Missense only, written as WT residue + position + MT residue (e.g. V123A). See examples

⇩ Drop a variants file here or browse  

v1.2 is the current model. v1.0 is the published model (RECOMB 2026 / bioRxiv v1–v2), kept so those results stay reproducible. Scores from different versions are not directly comparable. Use v1.2 unless you are reproducing the paper.

AlphaFold3 models are numbered from 1, so both conventions agree. Experimental PDB entries often start at another residue, or skip positions. Auto-detect resolves which convention your positions use, and results are reported back in that same convention.