PreGP: A Crop Genomic Prediction Method based on a Pretraining-Finetuning Framework

PreGP is a crop genomic prediction method based on a pretraining-finetuning Framework. Through this tutorial, you can complete model pre-training and fine-tuning to achieve phenotype prediction. 🍀

📗 Tutorial

🔨 Create Environment

1. Install Anaconda.

It is recommended to use Anaconda to build the model training environment.

2. Enter the PreGP directory.

cd PreGP

3. Execute the following command to create the PreGP environment

conda env export -n PreGP > pregp_env.yml

🏃 Pretrain Phase

1. Activate the PreGP environment.

conda activate PreGP

2. Start pretraining.

bash pretrain.sh

Here are the meanings of some key parameters, which need to be modified as needed.

Parameter	Description
geno_path	Path to genotype data CSV file
pretrain_model_path	Directory to save pre-trained model checkpoints
run_log_path	Directory for training logs
vocab_path	Directory containing vocabulary files
checkpoint_save_path	Directory to save training checkpoints
checkpoint_load_file_path	Directory to load checkpoints from

For more parameter descriptions, please refer to PARAMETER.md.

🚶 Finetuning Phase

1. Activate the PreGP environment.

conda activate PreGP

2. Start finetuning.

bash finetuning.sh

Here are the meanings of some key parameters, which need to be modified as needed.

Parameter	Description
load_model_name	Filename of the pre-trained model checkpoint to load
cvf_path	Path to cross-validation folds CSV file
phe_path	Path to phenotype data CSV file
pretrain_model_path	Directory containing the pre-trained model
fine_tuning_model_path	Directory to save fine-tuned model checkpoints
geno_path	Path to genotype data CSV file
run_log_path	Directory for training logs
vocab_path	Directory containing vocabulary files
vocab_name	Name of the vocabulary file
pred_save_path	Directory to save prediction results
unfreeze_from_layer	Layer index from which to unfreeze parameters

For more parameter descriptions, please refer to PARAMETER.md.

🙋 Download Pretrained Models and Vocabularies

You can download our pretrained models trained on large-scale genotype data and vocabularies at Hugging Face. Access Hugging Face 🤗 through the following link.

https://huggingface.co/integer8/PreGP

• Method 1: Directly download the model and vocabularies through the browser.

Simply click the download button next to the model to download the model weights and vocabularies you need.

• Method 2: Use git clone to get the models and vocabularies.

1. Install git lfs. If you are using Ubuntu, you can install it with the following command.

sudo apt update
sudo apt install git  
sudo apt install git-lfs  
git lfs install

2. Execute the following command.

git clone https://huggingface.co/integer8/PreGP