Construction and analysis of heart failure diagnosis model based on random forest and artificial neural network.

Heart failure is a global health problem and the number of sufferers is increasing as the population grows and ages. Existing diagnostic techniques for heart failure have various limitations in the clinical setting and there is a need to develop a new diagnostic model to complement the existing diagnostic methods. In recent years, with the development and improvement of gene sequencing technology, more genes associated with heart failure have been identified. We screened for differentially expressed genes in heart failure using available gene expression data from the Gene Expression Omnibus database and identified 6 important genes by a random forest classifier (ASPN, MXRA5, LUM, GLUL, CNN1, and SERPINA3). And we have successfully constructed a new heart failure diagnostic model using an artificial neural network and validated its diagnostic efficacy in a public dataset. We calculated heart failure-related differentially expressed genes and obtained 24 candidate genes by random forest classification, and selected the top 6 genes as important genes for subsequent analysis. The prediction weights of the genes of interest were determined by the neural network model and the model scores were evaluated in 2 independent sample datasets (GSE16499 and GSE57338 datasets). Since the weights of RNA-seq predictions for constructing neural network models were theoretically more suitable for disease classification of RNA-seq data, the GSE57338 dataset had the best performance in the validation results. The diagnostic model derived from our study can be of clinical value in determining the likelihood of HF occurring through cardiac biopsy. In the meantime, we need to further investigate the accuracy of the diagnostic model based on the results of our study.