[IJCAI 2024] Genomic-Selection

Introduction

This repository contains the code and data for our IJCAI 2024 paper An Embarrassingly Simple Approach to Enhance Transformer Performance in Genomic Selection for Crop Breeding. [Paper]

Dataset Preparation

We release the employed dataset Rice3k at https://drive.google.com/drive/folders/1H6XL9IHDvXR8Suq64bd1NxH_YghGdUYC?usp=sharing. Note that the genotypic results are placed in the Genotypic folder and six different phenotypic results are placed in the Phenotypic folder in a 5-fold format. folds, 3K_list_sra_ids.txt, and 3kRG_PhenotypeData_v20170411.xlsx should be downloaded for the dataset Rice3k preparation. Please feel free to use it.

However, we are very sorry to inform you that the Wheat dataset is limited by the need for our partners to use it in another article, so we currently do not have the permission to open source it. But we will open source in the future.

Environment

We run our code with PyTorch 1.13.1 with CUDA 11.7. It is better to install a higher version for flash attention.

Then install: flash-atten >= 2.4.2, apex

Usage

You can simply follow the instruction to train and evaluate:

bash distributed_train_wheat.sh for Wheat dataset.

bash distributed_train_rice3k.sh for Rice3k dataset.

Note that our model is a simple end-to-end training.

Contact

If you have any questions, please contact at [chenrenqi@pjlab.org.cn,hanwenwei@pjlab.org.cn].

Acknowledgement

This work is supported by Shanghai Artificial Intelligence Laboratory.

BibTeX & Citation

If you find this code useful, please consider citing our work:

@article{chen2024embarrassingly,
  title={An Embarrassingly Simple Approach to Enhance Transformer Performance in Genomic Selection for Crop Breeding},
  author={Chen, Renqi and Han, Wenwei and Zhang, Haohao and Su, Haoyang and Wang, Zhefan and Liu, Xiaolei and Jiang, Hao and Ouyang, Wanli and Dong, Nanqing},
  journal={arXiv preprint arXiv:2405.09585},
  year={2024}
}