Add model card

README.md

@@ -0,0 +1,100 @@
+---
+license: other
+license_name: pending-eth-biie
+license_link: LICENSE
+tags:
+  - biology
+  - bioinformatics
+  - immunology
+  - mhc
+  - peptide
+  - antigen-presentation
+  - vc-sfm
+pipeline_tag: feature-extraction
+---
+
+# mhcSFM — Peptide ↔ MHC Binding Specificity Foundation Model
+
+**Paper:** Vibe Coding Specificity Foundation Models · doi: TBD · [bioRxiv — link added at submission]
+**All VC-SFM models:** [huggingface.co/Reddy-BIIE-ETHZ](https://huggingface.co/Reddy-BIIE-ETHZ?search_models=VC-SFM)
+**Code:** [github.com/Reddy-BIIE-ETHZ/CALM](https://github.com/Reddy-BIIE-ETHZ/CALM)
+
+
+---
+
+## What it does
+
+mhcSFM learns a joint embedding space for **peptides** and **MHC alleles** via contrastive
+learning on binding affinity data. Given a peptide, retrieve the MHC alleles most likely
+to present it — enabling pan-allele binding prediction without per-allele fine-tuning.
+
+| Component | Model |
+|-----------|-------|
+| Peptide encoder | ESM-2 |
+| MHC encoder | ESM-2 (34-aa pseudo-sequence) |
+| Training data | IEDB binding affinity (BA) data |
+| Split | Sequence-identity 100% holdout · fold 0 |
+
+---
+
+## Performance (test set, identity-100 split)
+
+| Metric | Value |
+|--------|-------|
+| Precision @ 1 · R@1 | **0.153** |
+| MRR | **0.278** |
+| Pred. accuracy | 0.153 |
+
+*Retrieval over the full MHC allele library. Random baseline ≈ 0.5%. The identity-100 split enforces strict peptide novelty, making this a hard out-of-distribution task.*
+
+---
+
+## Quick start
+
+```python
+from huggingface_hub import hf_hub_download
+import torch, torch.nn.functional as F
+
+ckpt_path = hf_hub_download("Reddy-BIIE-ETHZ/mhcSFM_VC-SFM", "model.pth")
+
+from calm.encoder.model import CALMEncoder
+model = CALMEncoder.from_pretrained(ckpt_path)
+model.eval()
+
+peptide_emb = model.encode_query("GILGFVFTL")
+# MHC pseudo-sequence (34 aa from NetMHCpan convention)
+mhc_emb     = model.encode_target("YFAMYQENVAQTDVDTLYIIYRDAQTFQV")
+
+score = F.cosine_similarity(peptide_emb, mhc_emb, dim=-1)
+```
+
+---
+
+## Files in this repo
+
+| File | Description |
+|------|-------------|
+| `model.pth` | Final checkpoint · fold 0 · identity-100 split · epoch 4 |
+| `results_train_val_test.csv` | Full per-epoch train / val / test metrics |
+
+---
+
+## Citation
+
+```bibtex
+@article{vcsfm2025,
+  title   = {Vibe Coding Specificity Foundation Models},
+  author  = {Reddy, Sai T. and colleagues},
+  journal = {bioRxiv},
+  year    = {2025},
+  doi     = {TBD},
+  note    = {Preprint — doi and link added at submission}
+}
+```
+
+
+## License
+
+License pending finalisation with ETH Zurich / BIIE.
+For enquiries contact: sai.reddy@ethz.ch
+