Multiple-model machine learning identifies potential functional genes in dilated cardiomyopathy.

Area of Science:

• Biomedical Informatics
• Genomics
• Cardiology

Background:

• Machine learning (ML) is increasingly used for disease diagnosis, but single algorithms struggle with the complexity of dilated cardiomyopathy (DCM).
• Assessing the diagnostic value of ML in DCM requires sophisticated approaches beyond single-model analysis.

Purpose of the Study:

• To develop a novel multi-model ML approach using normalized sum weights for DCM diagnosis.
• To identify reliable gene biomarkers for DCM using an ensemble ML strategy.

Main Methods:

• Utilized gene expression data from six microarray datasets (386 samples), with a 5:1 ratio for training and testing.
• Developed six classification ML methods and identified differentially expressed genes (DEGs) between DCM and control groups.
• Selected candidate genes based on overall weights and receiver operating characteristic (ROC) analysis.

Main Results:

• Identified 20 DEGs (7 upregulated, 13 downregulated) in DCM patients.
• Three genes—SERPINA3, FRZB, and FCN3—demonstrated significant diagnostic value with areas under the curve (AUCs) > 0.88 in both training and testing sets.
• SERPINA3 showed particularly high diagnostic accuracy (AUC 0.940 in training, 0.918 in testing) and correlated with plasma cells; plasma levels of all three genes were significantly different in DCM patients.

Conclusions:

• SERPINA3, FRZB, and FCN3 are promising diagnostic biomarkers for dilated cardiomyopathy.
• The multi-model ML approach effectively identified potential diagnostic targets for DCM.
• Further clinical validation of these identified genes is warranted.