Add model card
#2
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
library_name: transformers
|
| 4 |
+
pipeline_tag: other
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# Proteo-R1: Reasoning Foundation Models for De Novo Protein Design
|
| 8 |
+
|
| 9 |
+
This repository contains the **generate** checkpoint of Proteo-R1, a reasoning-guided protein design framework that decouples molecular understanding from geometric generation. It uses a dual-expert architecture where a multimodal large language model (Qwen3 + Protenix) serves as the understanding expert, and a diffusion-based generation expert (Boltz1) performs conditional co-design under hard constraints.
|
| 10 |
+
|
| 11 |
+
- Paper: [Proteo-R1: Reasoning Foundation Models for De Novo Protein Design](https://huggingface.co/papers/2605.02937)
|
| 12 |
+
- Project page: [https://smiles724.github.io/r1/](https://smiles724.github.io/r1/)
|
| 13 |
+
- Code: [https://github.com/smiles724/Proteo-R1](https://github.com/smiles724/Proteo-R1)
|
| 14 |
+
|
| 15 |
+
## Quickstart
|
| 16 |
+
|
| 17 |
+
The model is used via two command-line tools: `proteor1-prepare-cdr` and `proteor1-design`. Both checkpoints are pulled automatically from the Hugging Face Hub.
|
| 18 |
+
|
| 19 |
+
```bash
|
| 20 |
+
# Step 1: emit per-CDR hidden states
|
| 21 |
+
proteor1-prepare-cdr \
|
| 22 |
+
--cif data/demo/8q7o_A__C/8q7o_A__C.cif \
|
| 23 |
+
--design-points "$(cat data/demo/8q7o_A__C/design_points.txt)" \
|
| 24 |
+
--out work/smoke/prepare_cdr \
|
| 25 |
+
--record-id 8q7o_A__C \
|
| 26 |
+
--emit-cdr-hidden \
|
| 27 |
+
--understand-ckpt thinking-bio-lab/proteor1-understand \
|
| 28 |
+
--override
|
| 29 |
+
|
| 30 |
+
# Step 2: structure design with framework inpainting
|
| 31 |
+
proteor1-design \
|
| 32 |
+
--input_dir data/demo_canonical_yamls/8q7o_A__C.yaml \
|
| 33 |
+
--output work/smoke \
|
| 34 |
+
--checkpoint thinking-bio-lab/proteor1-generate \
|
| 35 |
+
--precomputed_cdr_dir work/smoke/prepare_cdr/8q7o_A__C/cdr_hidden \
|
| 36 |
+
--processed_msa_dir data/demo_msa \
|
| 37 |
+
--structure_inpainting \
|
| 38 |
+
--ground_truth_structure_dir data/demo_gt_structures \
|
| 39 |
+
--no_cdr_json \
|
| 40 |
+
--override
|
| 41 |
+
```
|
| 42 |
+
|
| 43 |
+
See the [GitHub README](https://github.com/smiles724/Proteo-R1) for full installation instructions, troubleshooting, and additional examples.
|
| 44 |
+
|
| 45 |
+
## Citation
|
| 46 |
+
|
| 47 |
+
```bibtex
|
| 48 |
+
@article{wu2026proteo,
|
| 49 |
+
title={Proteo-R1: Reasoning Foundation Models for De Novo Protein Design},
|
| 50 |
+
author={Wu, Fang and Xuan, Weihao and Qi, Heli and Cao, Hanqun and Chang, Heng-Jui and Zhou, Zeqi and Zhao, Haokai and Jian, Ma and Ma, Carl and Cheng, Yu-Chi and others},
|
| 51 |
+
journal={arXiv preprint arXiv:2605.02937},
|
| 52 |
+
year={2026}
|
| 53 |
+
}
|
| 54 |
+
```
|