Related Experiment Video
Updated: Jun 4, 2026
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
Explainable machine learning-driven identification of heart failure biomarkers: a multi-model feature selection
Yuhe Zhao1, Ruoyu Zhang1, Kelan Zha2
1Department of Cardiology, The Seventh People's Hospital of Chongqing/The Central Hospital Affiliated to Chongqing University of Technology, Chongqing, China.
Scientific Reports
|May 25, 2026
Summary
Researchers developed a machine learning framework to find new heart failure (HF) biomarkers. FNDC1, LPCAT3, and TIMP2 were identified as potential transcriptomic signatures for HF diagnosis and understanding its mechanisms.
Area of Science:
- Genomics
- Biomarker Discovery
- Computational Biology
Background:
- Heart failure (HF) presents complex challenges in pathophysiology and biomarker identification.
- Existing biomarkers for HF have limitations in accuracy and scope.
- Novel transcriptomic signatures are needed for improved HF diagnosis and management.
Purpose of the Study:
- To develop a machine learning (ML) framework for identifying novel transcriptomic signatures of heart failure (HF).
- To prioritize and validate potential HF biomarkers using integrated transcriptomic datasets.
- To explore the functional implications of identified biomarkers in HF pathogenesis.
Main Methods:
- Integration and harmonization of three Gene Expression Omnibus (GEO) RNA-seq datasets (GSE141910, GSE198945, GSE263297).
- A triphasic feature selection process involving LASSO, Random Forest (RF), and SVM-Recursive Feature Elimination (SVM-RFE).
- A 10-model ensemble system evaluated using Leave-One-Study-Out Cross-Validation (LOSO-CV) and validated with an external cohort (GSE135055) and RT-qPCR.
Main Results:
- Three candidate biomarkers—FNDC1, LPCAT3, and TIMP2—were identified and prioritized.
- FNDC1 and TIMP2 showed significant upregulation, while LPCAT3 was suppressed in HF tissues (p < 0.001), confirmed by RT-qPCR.
- The ML models achieved high diagnostic performance (peak LOSO-CV AUC 0.973) and external validation (AUC up to 0.876).
- SHAP analysis highlighted FNDC1 as the most influential predictor, with functional enrichment linking signatures to extracellular matrix remodeling and lipid metabolism.
Conclusions:
- FNDC1, LPCAT3, and TIMP2 represent promising novel transcriptomic biomarkers for heart failure.
- These biomarkers are associated with key pathological mechanisms in HF, including extracellular matrix remodeling and lipid metabolism.
- The developed ML framework offers a robust approach for discovering transcriptomic signatures in complex diseases like HF.