A parameter-free deep embedded clustering method for single-cell RNA-seq data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for identifying cell heterogeneity and states.
• Existing clustering methods often require users to specify the number of clusters, which is difficult to determine accurately beforehand.

Purpose of the Study:

• To develop an automatic deep embedding clustering method (ADClust) for scRNA-seq data.
• To enable accurate cell clustering without a predefined number of clusters.

Main Methods:

• ADClust employs a pre-trained autoencoder for low-dimensional data representation.
• It clusters cells into initial micro-clusters, then merges similar clusters using statistical tests.
• Cell representations are updated iteratively by optimizing clustering and autoencoder loss functions.

Main Results:

• ADClust accurately clusters cells without requiring the number of clusters to be specified.
• The method was validated on 11 real scRNA-seq datasets, outperforming existing methods.
• ADClust demonstrates superior performance in both clustering accuracy and determining the correct number of clusters.

Conclusions:

• ADClust provides an effective and automated solution for scRNA-seq data clustering.
• The method is highly accurate, scalable, and efficient for large-scale single-cell analyses.