Related Experiment Video
Updated: Oct 27, 2025
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
7.7K
PEDF, a pleiotropic WTC-LI biomarker: Machine learning biomarker identification and validation
George Crowley1, James Kim1, Sophia Kwon1
1Department of Medicine, Division of Pulmonary, Critical Care and Sleep Medicine, New York University School of Medicine, New York, New York, United States of America.
Plos Computational Biology
|July 21, 2021
Summary
New biomarkers for World Trade Center-Lung Injury (WTC-LI) were identified using machine learning, including pigment epithelium-derived factor (PEDF). These novel WTC-LI biomarkers showed high predictive performance in a firefighter cohort.
Area of Science:
- Environmental Health
- Biomarker Discovery
- Pulmonary Medicine
Background:
- World Trade Center-Lung Injury (WTC-LI) diagnosis is challenging due to multicollinearity in existing biomarker datasets.
- Serum cytokines, chemokines, and high-throughput data require advanced methods for accurate WTC-disease phenotyping.
Purpose of the Study:
- To identify and validate novel biomarkers for WTC-LI using automated, machine-learning-based data pruning.
- To address multicollinearity in high-dimensional datasets for improved WTC-LI prediction.
Main Methods:
- Employed machine learning and high-dimensional data pruning on serum cytokines, chemokines, metabolites, and clinical data from firefighters with and without WTC-LI.
- Utilized random forests for feature selection and binary logistic regression with 5-fold cross-validation for biomarker validation.
- Assessed a cohort of 100 male, never-smoking firefighters with WTC-LI and 127 controls, with a subset (n=15/group) undergoing metabolomics.
Main Results:
- Identified key biomarkers including pigment epithelium-derived factor (PEDF), macrophage derived chemokine (MDC), systolic blood pressure, macrophage inflammatory protein-4 (MIP-4), growth-regulated oncogene protein (GRO), monocyte chemoattractant protein-1 (MCP-1), and apolipoprotein-AII (Apo-AII).
- Validated models achieved an AUCROC of 0.90 (0.84-0.96) after adjusting for confounders.
- Specific biomarker associations found: decreased PEDF and MIP-4, increased Apo-AII linked to higher WTC-LI odds; increased GRO, MCP-1, and decreased MDC linked to lower WTC-LI odds.
Conclusions:
- Automated data pruning successfully identified novel and predictive biomarkers for WTC-LI, validated in an independent cohort.
- Pigment epithelium-derived factor (PEDF) is a novel predictive biomarker for particulate-matter-related lung disease.
- Biomarkers like GRO, MCP-1, MDC, and MIP-4 suggest immune cell involvement in WTC-LI pathogenesis, indicating potential pharmacotherapy targets.
