EXGEP

A framework for predicting genotype-by-environment interactions using ensembles of explainable machine-learning models

Related Software and Tools

• DNNGP – Deep neural network for genomic prediction.
• AutoGS – A framework for predicting genotype-by-environment interactions using ensem)bles of explainable machine-learning models.
• GxEtoolkit – An automated and explainable machine learning framework for Genome Prediction.

Table of Contents

• Getting started
• Base Usage

Getting started

Requirements

• Python 3.9
• pip

Installation

Install packages:

1. Create a python environment.

conda create -n exgep python=3.9
conda activate exgep

2. Clone this repository and cd into it.

git clone https://github.com/AIBreeding/EXGEP.git
cd ./exgep
pip install -r requirements.txt

Base Usage

import os
import time
import argparse
import pandas as pd
from datetime import datetime
from exgep.data import datautils
from exgep.model import RegEXGEP
from exgep.data.reg_metrics import (mae_score as mae, 
                           mse_score as mse, 
                           rmse_score as rmse, 
                           r2_score as r2,
                           rmsle_score as rmsle, 
                           mape_score as mape, 
                           medae_score as medae, 
                           pcc_score as pcc)


geno = './data/genotype.csv'
phen = './data/pheno.csv'
soil = './data/soil.csv'
weather = './data/weather.csv'

data = datautils.merge_data(geno, phen, soil, weather)
X = pd.DataFrame(data.iloc[:, 3:])
y = data['Yield']
y = pd.core.series.Series(y)
    
regression = RegEXGEP(
    y=y,
    X=X,
    test_frac=0.1,
    n_splits=10,
    n_trial=5,
    reload_study=True,
    reload_trial_cap=True,
    write_folder=os.getcwd()+'/result/',
    metric_optimise=r2,
    metric_assess=[mae, mse, rmse, pcc, rmsle, mape, medae],
    optimisation_direction='maximize',
    models_to_optimize=['LightGBM'],
    models_to_assess=['LightGBM'],
    boosted_early_stopping_rounds=5,
    random_state=2024
)

start = time.time()
regression.train()
end = time.time()
print(end - start)

🌲 Optional Tree Models

• DTR – Decision Tree Regressor
• ETR – Extra Trees Regressor
• LightGBM – Light Gradient Boosting Machine
• XGBoost – Extreme Gradient Boosting
• CatBoost – Categorical Boosting
• AdaBoost – Adaptive Boosting
• GBDT – Gradient Boosting Decision Tree
• Bagging – Bagging Regressor
• RF – Random Forest Regressor
• HistGradientBoosting – Histogram-based Gradient Boosting

Training Example （Parameter Configuration）

python train_exgep.py \
--geno ./data/geno.csv \
--phen ./data/pheno.csv \
--soil ./data/soil.csv \
--weather ./data/weather.csv \
--target Yield \
--test_size 0.1 \
--n_splits 10 \
--n_trial 5 \
--models_optimize XGBoost LightGBM \
--models_assess XGBoost LightGBM

Training Parameters Details

Parameter	Description	Default	Type
--geno	Path to genotype CSV file	./data/genotype.csv	string
--phen	Path to phenotype CSV file	./data/pheno.csv	string
--soil	Path to soil CSV file	./data/soil.csv	string
--weather	Path to weather CSV file	./data/weather.csv	string
--target	Target column name in the dataset	Yield	string
--test_size	Fraction of data to be used for testing	0.1	float
--n_splits	Number of splits for cross-validation	10	integer
--n_trial	Number of optimization trials	5	integer
--models_optimize	Models to use for hyperparameter optimization	['XGBoost']	list
--models_assess	Models to use for performance assessment	['XGBoost']	list

Model Explainable Example

python test_explain.py \
--geno ./data/genotype.csv \
--phen ./data/pheno.csv \
--soil ./data/soil.csv \
--weather ./data/weather.csv \
--target Yield \
--model EXGEP \
--job_id 20240813103950 \
--sample 2 \
--feature_i pc2 \
--feature_j RH2M \
--top_features 10 \
--top_interactions 20 \
--test_size 0.1 \
--random_state 2024

Model Explanation Parameters Details

Parameter	Description	Default	Type
--model	Model to explain	-	string
--job_id	Job directory containing trained models	-	string
--geno	Path to genotype CSV file	./data/genotype.csv	string
--phen	Path to phenotype CSV file	./data/pheno.csv	string
--soil	Path to soil CSV file	./data/soil.csv	string
--weather	Path to weather CSV file	./data/weather.csv	string
--target	Target column name	Yield	string
--test_size	Test set fraction	0.2	float
--random_state	Random seed for reproducibility	2024	integer
--sample	Sample index for waterfall plot	2	integer
--feature_i	Primary feature for dependence plots	pc2	string
--feature_j	Secondary feature for interaction plots	RH2M	string
--top_features	Number of top features to display	20	integer
--top_interactions	Number of top interactions for network	20	integer
--cluster	Use KMeans clustering for background data	False	flag
--n_train_points	Training points for background	200	integer
--n_test_points	Test samples to explain	None	integer

📚 Citation

You can read our paper explaining EXGEP.

Yu T, Zhang H, Chen S, et al. EXGEP: a framework for predicting genotype-by-environment interactions using ensembles of explainable machine-learning models. Brief Bioinform, 2025. https://doi.org/10.1093/bib/bbaf414

📜Copyright and License

This project is free to use for non-commercial purposes - see the LICENSE file for details.

👥Contacts

For more information, please contact with Huihui Li (lihuihui@caas.cn).