File size: 2,390 Bytes
ca54bba ef965ba 585d625 ca54bba 585d625 ef965ba 585d625 ef965ba 585d625 ef965ba 585d625 ef965ba 585d625 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 | ---
license: apache-2.0
library_name: peft
base_model:
- biohub/ESMC-6B
- westlake-repl/SaProt_1.3B_AFDB_OMG_NCBI
tags:
- protein
- protein-language-model
- lora
- expression-yield
- nanobody
---
# monoyield-predictor: trained weights
LoRA adapters and regression heads that predict the **monomeric expression yield in *E. coli*** (log10 mg/L) of a protein sequence.
Companion weights of the paper *[title - to be filled]* and the code repository **https://github.com/tsubo-lab/monoyield-predictor**.
The files contain only the trained LoRA matrices and the regression head (7-12 MB per fold). The frozen base models are downloaded from Hugging Face when the code runs.
| folder | model in the paper | base model | input | target |
|---|---|---|---|---|
| `esmc_monoyield/` | ESMC-MonoYield | `biohub/ESMC-6B` | amino-acid sequence | log10 monomer yield (mg/L), de novo + DMS |
| `saprot_lora/` | SaProt LoRA | `westlake-repl/SaProt_1.3B_AFDB_OMG_NCBI` | amino acid + Foldseek 3Di | same |
| `esmc_nb_monoyield/` | ESMC-Nb-MonoYield | `biohub/ESMC-6B` | nanobody sequence | log10 monomer yield (mg/L) |
| `esmc_nb_mdratio/` | ESMC-Nb-MDRatio | `biohub/ESMC-6B` | nanobody sequence | log10 monomer/dimer ratio |
Each folder holds five fold checkpoints, `<model>_1.pt ... <model>_5.pt`; predictions are the mean of the five.
## Usage
```bash
git clone https://github.com/tsubo-lab/monoyield-predictor
cd monoyield-predictor
pip install -e .
HF_REPO=Tsuboyama-Lab/monoyield-predictor bash download_weights.sh weights
python scripts/predict_esmc_mlp_head.py --external_csv examples/example_main_library.csv \
--model_dir weights/esmc_monoyield --model_name biohub/ESMC-6B \
--model_revision 45b0fa5d7fb06faefbd5e3b89bdcef35d564e79a --dtype float32 --batch_size 1 --out_csv predictions.csv
```
See the code repository for all models, input formats and training commands.
ESMC-6B in fp32 needs a GPU with about 48 GB of memory (`--dtype bfloat16` fits 24 GB but should be validated).
## License and terms
The adapters are released under the Apache License 2.0 (Copyright 2026 Kotaro Tsuboyama and Selena Setsu). They are derivatives of the base models and must be used under the terms of those models:
ESMC-6B (MIT with third-party notices and the Biohub Acceptable Use Policy, https://biohub.org/acceptable-use-policy/) and SaProt (MIT).
## Citation
[to be filled]
|