{"architecture":{"family":"position weight matrices; 3.0 is a hybrid where a network predicts the matrix","input":"peptide + 34-residue groove sequence (3.0)","notes_en":"A separate model per peptide length instead of alignment search. The independence assumption a PWM must make is close to true here, which is why such a simple model competes with neural networks.","output":"score and %Rank"},"article":"/blog/modele-prezentacji-antygenu","cancer_slugs":["pan-cancer"],"category":"immunopeptidomics","confidence":"medium","datasets":[{"name":"HLA Ligand Atlas","note":"","role":"training","slug":"hla-ligand-atlas"},{"name":"MHC Motif Atlas","note":"","role":"training","slug":"mhc-motif-atlas"},{"name":"Mono-allelic HLA class I peptidome (Sarkizova / Abelin)","note":"","role":"training","slug":"monoallelic-peptidome"}],"developer":"Gfeller lab, Ludwig Institute for Cancer Research / University of Lausanne","evaluation":[{"benchmark":"cross-allele and cross-species ligand prediction","external":true,"metric":"AUC / PPV","source":"https://pmc.ncbi.nlm.nih.gov/articles/PMC11927126/","value":"competitive with NetMHCpan-4.1; authors caution that AUC up to 0.95 is achievable even with fairly unspecific motifs on random decoys"}],"hf":null,"kind":"task-model","license":"free for academic use (see repository terms)","limitations":"- PWMs cannot express position interactions; where they matter (some class II settings), the model is structurally blind to them.\n- Motif-to-allele assignment via co-occurrence needs a diverse donor panel.","links":{"demo":null,"docs":null,"doi":"10.1186/s13073-025-01449-1","github":"https://github.com/GfellerLab/MixMHCpred","huggingface":null,"paper":"https://pmc.ncbi.nlm.nih.gov/articles/PMC11927126/","pmid":null},"modalities":["protein-sequence","immunopeptidomics"],"name":"MixMHCpred 3.0","notable_uses":"The MixMHCp trick \u2014 if two samples share exactly one allele, a motif present in both must come from it \u2014 is how many allele motifs were characterised without ever building a mono-allelic cell line.","openness":"open-weights","regulatory":{"intended_use_en":"Research software.","intended_use_pl":"Oprogramowanie badawcze.","source_url":"https://github.com/GfellerLab/MixMHCpred","status":"research-only"},"regulatory_status":"research-only","release_date":"2025-03-19","run_snippet":"","settings":["basic-research","vaccine-design","immunotherapy"],"slug":"mixmhcpred","sources":[{"label":"Tadros DM, Racle J, Gfeller D. Predicting MHC-I ligands across alleles and species. Genome Med 2025","url":"https://pmc.ncbi.nlm.nih.gov/articles/PMC11927126/"},{"label":"Bassani-Sternberg M et al. Deciphering HLA-I motifs across HLA peptidomes. PLoS Comput Biol 2017","url":"https://pmc.ncbi.nlm.nih.gov/articles/PMC5584980/"},{"label":"cancer3.ai \u2014 Trzydzie\u015bci cztery litery zamka","url":"https://cancer3.ai/blog/modele-prezentacji-antygenu"}],"summary":"Position-weight-matrix predictor built on the observation that class I ligands show almost no dependence between positions; version 3.0 closes the circle by having a neural network predict the matrix from the same 34 groove residues.","tasks":["antigen-presentation","peptide-mhc-binding","neoantigen-prioritisation"],"training":{"summary_en":"Mono- and multi-allelic immunopeptidomics, the latter deconvolved by MixMHCp before training.","summary_pl":"Immunopeptydomika jedno- i wieloalleliczna, ta druga rozpl\u0105tana przez MixMHCp przed treningiem."},"updated_at":"2026-09-05T22:26:00.871374","url":"/ai-oncology/models/mixmhcpred","usage":{"library":"standalone C/Python (repo)","notes_en":"Clone the repo and compile; MixMHCp is a separate tool for motif deconvolution of your own immunopeptidomics data."},"verified_at":"2026-09-05T22:26:00.871048","verified_by":"editorial","version":"3.0","what_it_does":"Scores a peptide by summing per-position preferences of the allele. Its companion MixMHCp solves multi-allelic data before training, by fitting several motifs to the mixture and using allele co-occurrence across donors to decide which motif belongs to which allele."}