Background: It is difficult to distinguish between arrhythmogenic cardiomyopathy (ACM) and dilated cardiomyopathy (DCM) because of their similar clinical manifestations. This study aimed to develop a novel diagnostic algorithm for distinguishing ACM from DCM.
Methods: Two public datasets containing human ACM and DCM myocardial samples were used. Consensus clustering, non-negative matrix factorization and principal component analysis were applied. Weighted gene co-expression network analysis and machine learning methods, including random forest and the least absolute shrinkage and selection operator, were used to identify candidate genes. Receiver operating characteristic curves and nomograms were performed to estimate diagnostic efficacy, and Spearman's correlation analysis was used to assess the correlation between candidate genes and cardiac function indices.
Results: Both ACM and DCM showed highly similar gene expression patterns in the clustering analyses. Hub gene modules associated with cardiomyopathy were obtained using weighted gene co-expression network analysis. Thirteen candidate genes were selected using machine learning algorithms, and their combination showed a high diagnostic value (area under the ROC curve = 0.86) for distinguishing ACM from DCM. In addition, TATA-box binding protein associated factor 15 showed a negative correlation with cardiac index (R = -0.54, p = 0.0054) and left ventricular ejection fraction (R = -0.48, p = 0.0015).
Conclusions: Our study revealed an effective diagnostic model with key gene signatures, which indicates a potential tool to differentiate between ACM and DCM in clinical practice. In addition, we identified several genes that are highly related to cardiac function, which may contribute to our understanding of ACM and DCM.
Keywords: WGCNA; arrhythmogenic cardiomyopathy; dilated cardiomyopathy; gene expression profiling; machine learning.
© 2022 John Wiley & Sons Ltd.