Predicting depression using serum perfluoroalkyl and polyfluoroalkyl substances levels via interpretable machine
Hui Jin1, Yang Wen1, Shuai Luo2
1Mental Health Center, West China Hospital, Sichuan University, Chengdu, Sichuan 610041, China; Department of Social Psychiatry, West China Hospital, Sichuan University, Chengdu, Sichuan 610041, China.
Journal of Affective Disorders
|August 3, 2025
Summary
Machine learning models predict depression risk from per- and polyfluoroalkyl substances (PFAS) exposure. Higher levels of a specific PFAS, PFOS, showed a threshold effect linked to increased depression risk.
Area of Science:
- Environmental Health
- Toxicology
- Computational Biology
Background:
- Per- and polyfluoroalkyl substances (PFAS) are persistent synthetic chemicals with widespread human exposure.
- No prior studies have utilized machine learning (ML) to predict depression risk based on PFAS exposure.
- This research introduces an interpretable ML model for assessing depression risk linked to serum PFAS levels.
Purpose of the Study:
- To develop and validate an interpretable machine learning model for predicting depression risk.
- To identify key PFAS associated with depression.
- To provide a tool for assessing individual depression risk based on PFAS exposure.
Main Methods:
- Utilized National Health and Nutrition Examination Survey (NHANES) data (2005-2018) from 9074 participants.
- Developed and evaluated nine ML models, selecting CatBoost for superior predictive performance (73% accuracy, 0.75 AUC).
- Employed Partial Dependence Analysis (PDA) and SHapley Additive exPlanations (SHAP) for model interpretability and feature analysis. Developed a web calculator for real-time risk assessment.
Main Results:
- The CatBoost model achieved 73% accuracy and an AUC of 0.75.
- Perfluorooctanesulfonic acid (PFOS) was identified as the most influential PFAS.
- A threshold effect for PFOS was observed at 11.66 ng/ml, correlating with elevated depression risk.
Conclusions:
- An interpretable ML model successfully links PFAS exposure to depression risk.
- Findings support public health interventions and personalized care strategies.
- An interactive web calculator is available to translate research findings into clinical practice.
