---
tags:
- biology
- protein
- binding-site
---

# BindSite weights

Weights loaded by the BindSite AI server ([projectxddi/be_protein_binding](https://huggingface.co/spaces/projectxddi/be_protein_binding)),
which predicts the residues of a protein that bind metal ions, small molecules, DNA and RNA.

| Path | Contents | Origin and license |
|---|---|---|
| `final_model/` | AdaBind: 140 checkpoints for 5 ligand classes, `manifest.json` with members and thresholds | this project; license not chosen yet |
| `esm_if1/esm_if1_gvp4_t16_142M_UR50.pt` | ESM-IF1 inverse-folding model, unchanged copy | [facebookresearch/esm](https://github.com/facebookresearch/esm), MIT |
| `baselines/bindEmbed21DL/` | 5 checkpoints (comparison model) | [Rostlab/bindPredict](https://github.com/Rostlab/bindPredict), MIT, `LICENSE` included |
| `baselines/bindNode24/` | 5 state dicts (comparison model) | [erckert/bindNode24](https://github.com/erckert/bindNode24), GPL-3.0 |
| `esmfold_files.tar.gz` | 4,090 ESMFold v1 structures: the training and test proteins, and the example proteins of the Predict page | predicted with [facebook/esmfold_v1](https://huggingface.co/facebook/esmfold_v1) (MIT) |

ESM-2 650M and ProtT5-XL half are not copied here. The server loads them from their own repositories at fixed
revisions, byte-identical to the files the models were trained with (see `MANIFEST.json`):
[facebook/esm2_t33_650M_UR50D](https://huggingface.co/facebook/esm2_t33_650M_UR50D) @ `08e4846e`,
[Rostlab/prot_t5_xl_half_uniref50-enc](https://huggingface.co/Rostlab/prot_t5_xl_half_uniref50-enc) @ `94a6abc0`.

`SHA256SUMS` lists every file of this repository.