Spectral Clustering of Single-Cell RNA-Sequencing Data by Multiple Feature Sets Affinity

A critical stage in the study of single-cell RNA-sequencing (scRNA-seq) data is cell clustering. The quality of feature selection, which comes first in unsupervised clustering, directly affects the quality of the analysis that follows. It is difficult to choose high-quality characteristics since the gene expression data from scRNA-seq are high dimensional. Feature extraction is often used on gene expression data to choose highly expressed features, that is, subsets of original features. The typical ways for feature selection are to either reserve by percentage or to simply establish a specified threshold number based on experience. It is challenging to guarantee that the first-rank clustering results can be procured using these methods because they are so subjective. In this study, we propose a feature selection method scMFSA to overcome the one-dimensional shortcoming of the traditional PCA method by selecting multiple top-level feature sets. The similarity matrix constructed from each feature set is enhanced by affinity to optimize the feature learning. Lastly, studies are carried out on the actual scRNA-seq datasets using the features discovered in scMFSA. The findings indicate that when paired with clustering methods, the features chosen by scMFSA can increase the accuracy of clustering results. As a result, scMFSA can be an effective tool for researchers to employ when analyzing scRNA-seq data.