Cluster Analysis of Scrna-Seq Data Combining Bioinformatics with Graph Attention Autoencoders and Ensemble Clustering

As single-cell RNA sequencing (scRNA-seq) technology has rapidly become a powerful technique for revealing gene expression information at the cellular level. In scRNA-seq data analysis, cell clustering is a key step in downstream analysis as it can identify cell types and discover new cell subtypes. However, the high dimensionality, sparsity, and high noise characteristics of scRNA-seq datasets present significant challenges for clustering analysis.In this study, a model based on bipartite graph integration clustering and graph attention autoencoder is proposed. Firstly, the scRNA-seq dataset is preprocessed using network enhancement (NE) and principal component analysis (PCA) for denoising and feature selection. Next, a graph attention autoencoder is employed for dimension reduction to obtain low-dimensional embeddings. Finally, bipartite graph integration clustering is utilized to derive the final clustering results based on the relationship between cells and low-dimensional embeddings. Based on various clustering metrics, a comparison was made between different scRNA-seq datasets, and the experimental results showed that scBAGA outperformed other advanced methods. This indicates that our model can serve as a reliable classification tool.