mhcSFM

mhcSFM is a Specificity Foundation Model (SFM) for predicting peptide–MHC binding specificity directly from sequence. Which peptides are presented by which major histocompatibility complex (MHC) alleles governs T-cell recognition and is central to vaccine design, immunotherapy, and neoantigen discovery. Conventional predictors are trained allele-by-allele on mass-spectrometry and binding-affinity data; mhcSFM instead frames peptide–MHC matching as a cross-modal retrieval problem, learning to align cognate peptide–MHC pairs in a shared representation space so that likely presentation events can be scored from sequence alone.

Developed by the Reddy lab at ETH Zurich and posted as a bioRxiv preprint in June 2026, mhcSFM is one of six models in the SFM family, all built on a single, physics-derived dual-encoder architecture. It is the sequel to CALM-1, the antibody–antigen specificity model from the same group, generalizing that contrastive molecular-recognition recipe from antibody binding to MHC presentation.

The model encodes peptide and MHC sequences with separate encoders and aligns them using a symmetric contrastive objective, pulling true presentation pairs together and pushing non-binders apart. This formulation lets mhcSFM transfer knowledge across alleles into zero-shot predictions for held-out peptides and MHC variants.

#Key Features

• Sequence-to-specificity prediction: Predicts peptide–MHC binding from peptide and MHC sequence alone, without requiring allele-specific structural models.
• Physics-derived dual-encoder: Encodes peptide and MHC separately, with an architecture motivated by the physics of molecular recognition rather than a generic backbone.
• Symmetric contrastive learning: Aligns cognate peptide–MHC pairs in a shared embedding space, enabling retrieval in either direction (peptide-to-MHC or MHC-to-peptide).
• Learned Boltzmann temperature: A learned temperature parameter calibrates the contrastive similarity scores in a thermodynamically motivated way.
• Zero-shot cross-modal retrieval: Generalizes to unseen peptides and MHC alleles without task-specific fine-tuning.

#Technical Details

mhcSFM uses the shared SFM architecture: a physics-derived dual-encoder trained with a symmetric contrastive objective and a learned Boltzmann temperature that calibrates similarity scores. The two encoders embed peptide and MHC sequences independently, and the contrastive loss aligns cognate pairs while separating mismatches. The model is pretrained on public peptide–MHC specificity data and evaluated by zero-shot cross-modal retrieval on held-out pairs, where it reports strong top-k retrieval performance—mirroring the benchmarks used across the SFM family for measuring how reliably a model recovers true presentation partners.

#Applications

mhcSFM is aimed at computational immunology, where predicting which peptides an MHC allele presents from sequence can accelerate epitope prediction, neoantigen prioritization, and vaccine design. By scoring and retrieving likely peptide–MHC pairs, it can help triage candidate epitopes across patient-specific HLA backgrounds, support T-cell target discovery, and complement mass-spectrometry immunopeptidomics where coverage is incomplete.

#Impact

mhcSFM extends the contrastive specificity-prediction paradigm established by CALM-1 from antibody–antigen recognition to peptide–MHC recognition, demonstrating that a single physics-derived dual-encoder recipe transfers across molecular domains. As one of six SFMs released together, it contributes evidence that cross-modal contrastive learning is a general tool for biological specificity prediction. Its main current limitations are those of a recent preprint: results await peer review and independent benchmarking, and at the time of release no public code or weights repository was available, so reproduction depends on forthcoming artifact releases.