An Enhanced Random Forests Approach to Predict Heart Failure From Small Imbalanced Gene Expression Data.

Area of Science:

• Biomedical Informatics
• Cardiovascular Research
• Genomics

Background:

• Myocardial infarctions and heart failure cause over 17 million deaths globally each year.
• ST-segment elevation myocardial infarctions (STEMI) demand rapid treatment due to severe clinical consequences of delays.
• Machine learning offers potential for predicting heart failure and identifying associated genes.

Purpose of the Study:

• To apply machine learning for predicting heart failure from gene expression data.
• To identify and rank genes significantly associated with heart failure.
• To validate the identified genes through literature review and gene set enrichment analysis.

Main Methods:

• Utilized a Random Forests classifier with feature elimination on microarray gene expression data from 111 STEMI patients.
• Evaluated classifier performance using Matthews correlation coefficient (MCC) and ROC AUC metrics.
• Ranked genes by importance to identify those most strongly linked to heart failure.

Main Results:

• The machine learning classifier achieved high accuracy with MCC = +0.87 and ROC AUC = 0.918 in predicting heart failure.
• Identified KLHL22, WDR11, OR4Q3, GPATCH3, and FAH as the top five protein-coding genes associated with heart failure.
• Validated the gene ranking through literature review and gene set enrichment analysis.

Conclusions:

• Machine learning with feature elimination is effective for predicting heart failure from gene expression.
• The identified top genes provide valuable targets for further research in heart failure.
• These findings can assist biologists and cardiologists in advancing the understanding and treatment of heart failure.