Prism

Prism provides pretrained checkpoints for gene expression prediction by integrating genomic sequence and multimodal signals.

This repository is the model release for:

Extending Sequence Length is Not All You Need: Effective Integration of Multimodal Signals for Gene Expression Prediction (ICLR 2026)

Paper

• Hugging Face Paper Page
• arXiv: 2602.21550

Model Contents

• Pretrained checkpoints for K562 and GM12878
• Five random seeds for each cell type: 2, 22, 222, 2222, 22222

Dataset

Prism follows the same dataset setting as Seq2Exp (xingyusu/GeneExp).

Quick Start

Download checkpoints:

pip install huggingface_hub
python -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='yangyz1230/Prism', repo_type='model', local_dir='./ckpt')"

Run inference with the official code:

git clone https://github.com/yangzhao1230/Prism
cd Prism
pip install -r requirements.txt
DATA_ROOT=/path/to/data
bash test.sh $DATA_ROOT ./ckpt

Limitations

• Research use only
• Performance may vary across preprocessing settings and seeds
• Not intended for clinical or diagnostic use

Citation

@inproceedings{
yang2026extending,
title={Extending Sequence Length is Not All You Need: Effective Integration of Multimodal Signals for Gene Expression Prediction},
author={Zhao Yang and Yi Duan and Jiwei Zhu and Ying Ba and Chuan Cao and Bing Su},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026}
}

Links

• Code: https://github.com/yangzhao1230/Prism
• Model: https://huggingface.co/yangyz1230/Prism