Related Experiment Video
Updated: May 7, 2025
05:40
The Motivation for Alcohol Reward: Predictors of Progressive-Ratio Intravenous Alcohol Self-Administration in Humans
Published on: April 28, 2022
2.9K
Machine Learning Approaches to Predict Alcohol Consumption from Biomarkers in the UK Biobank
Medrxiv : the Preprint Server for Health Sciences
|January 7, 2025
Summary
Machine learning models accurately predict alcohol consumption using biological markers, offering a complementary approach to self-reports for identifying heavy drinkers. Explainable AI highlights key biomarkers for risk assessment.
Area of Science:
- Biomarkers and Health Analytics
- Machine Learning in Public Health
- Genetics and Lifestyle Factors
Background:
- Accurate alcohol consumption (AC) measurement is vital for public health.
- Traditional methods like self-reports and interviews have limitations.
- Biological markers offer a complementary approach to AC estimation.
Purpose of the Study:
- To evaluate machine learning (ML) predictions of AC using blood and urine biomarkers.
- To compare the performance of five ML models in predicting alcohol intake.
- To assess the utility of biomarkers in enhancing AC prediction accuracy.
Main Methods:
- Utilized UK Biobank data for analysis.
- Employed five ML models: LASSO, Ridge, GBM, MBOOST, and XGBOOST.
- Predicted alcohol consumption (Drinks Per Week - DPW) and related phenotypes.
Main Results:
- All ML models achieved moderate prediction of DPW (r²=0.304-0.356).
- Biomarkers significantly improved prediction accuracy compared to covariates alone (r²=0.105).
- XGBOOST showed the best performance (r²=0.356) and accurately identified heavy drinkers.
Conclusions:
- ML-based biological measures can identify individuals at risk of heavy AC.
- This approach complements traditional methods, mitigating self-report biases.
- Explainable AI identified key biomarkers associated with alcohol consumption.
