Nugkta/bgcbench-kawano

Kawano et al. Model I (BGC-Bench mirror)

This repository redistributes the Kawano et al. Model I weights released by the original authors, so that BGC-Bench can download them at a fixed tag. It is not an official release of the original authors, and the weights are not retrained. Licence: CC BY-NC 4.0, as upstream; non-commercial use only. Please cite the original paper (below). Intended use: frozen feature extraction in BGC-Bench; see the original paper for the model's training data and limitations.

Upstream

• Paper: Kawano, Shiraishi, Kuzuyama & Umemura (2026). A novel transformer-based platform for the prediction and design of biosynthetic gene clusters for (un)natural products. PLOS Computational Biology 22(2): e1013181. https://doi.org/10.1371/journal.pcbi.1013181
• Code: https://github.com/umemura-m/bgc-transformer
• Weights: Zenodo record 17577731 (CC BY-NC 4.0), https://doi.org/10.5281/zenodo.17577731: Kawano_etal_modelI.v2.tar.gz

Format

Files copied byte for byte from the archive: models/model_I_roberta_bgc/out.bgc/{config.json, model.safetensors}, and models/vocab_pfam_product_hmmer33.json saved as tokenizer.json. The config keeps pad_token_id = 1 although the tokenizer pads with id 3; RoBERTa derives position ids from it, so it is left as released. LICENSE is the upstream file: a summary of CC BY-NC 4.0 that links the legal code (https://creativecommons.org/licenses/by-nc/4.0/legalcode).

Model

RoBERTa masked language model (8 layers, 1024-d) with one word-level token per Pfam domain name. BGC-Bench uses the last hidden layer: its masked mean over a 512-token, max-length-padded input, and the state at an A-domain's AMP-binding token.

Files

file	bytes
LICENSE	740
config.json	637
model.safetensors	423,142,660
tokenizer.json	481,170

Use

git clone https://github.com/Nugkta/BGC-Bench && cd BGC-Bench
uv sync
uv run bgcbench run --encoder kawano

Citation

@article{kawano2026roberta,
  title   = {A Novel Transformer-Based Platform for the Prediction and Design of Biosynthetic Gene Clusters for (Un)Natural Products},
  author  = {Kawano, Tomoki and Shiraishi, Taro and Kuzuyama, Tomohisa and Umemura, Maiko},
  journal = {PLOS Computational Biology}, volume = {22}, number = {2}, pages = {e1013181}, year = {2026},
  doi     = {10.1371/journal.pcbi.1013181}
}