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

Abstract

Clustering analysis is widely used in single-cell ribonucleic acid (RNA)-sequencing (scRNA-seq) data to discover cell heterogeneity and cell states. While many clustering methods have been developed for scRNA-seq analysis, most of these methods require to provide the number of clusters. However, it is not easy to know the exact number of cell types in advance, and experienced determination is not always reliable. Here, we have developed ADClust, an automatic deep embedding clustering method for scRNA-seq data, which can accurately cluster cells without requiring a predefined number of clusters. Specifically, ADClust first obtains low-dimensional representation through pre-trained autoencoder and uses the representations to cluster cells into initial micro-clusters. The clusters are then compared in between by a statistical test, and similar micro-clusters are merged into larger clusters. According to the clustering, cell representations are updated so that each cell will be pulled toward centers of its assigned cluster and similar clusters, while cells are separated to keep distances between clusters. This is accomplished through jointly optimizing the carefully designed clustering and autoencoder loss functions. This merging process continues until convergence. ADClust was tested on 11 real scRNA-seq datasets and was shown to outperform existing methods in terms of both clustering performance and the accuracy on the number of the determined clusters. More importantly, our model provides high speed and scalability for large datasets.

A critical assessment of clustering algorithms to improve cell clustering and identification in single-cell transcriptome study

“…Diverse clustering algorithms were evaluated using 10 scRNA-seq datasets. Seurat was able to identify distinct clusters, but the clusters were found to be highly sensitive to the choice of resolution parameter, where higher values tend to result in a larger number of clusters [ 40 , 41 ]. Deep learning clustering algorithms have emerged as promising approaches to address challenges such as batch effects in scRNA-seq data, and were shown to be effective algorithm for cell clustering and identification [ 23 , 24 ].…”

scAce: an adaptive embedding and clustering method for single-cell gene expression data

“…From clustering methods based on VAEs, we chose the most widely used scVI method ( Lopez et al 2018 ) and the recently published scGMAAE method ( Wang et al 2023 ). In addition, we selected two clustering methods that also use cluster merging approaches, SCCAF ( Miao et al 2020 ) and ADClust ( Zeng et al 2022 ), and one clustering method based on the graph neural network, graph-sc ( Ciortan and Defrance 2022 ). For methods requiring a cluster number as input, the number of real cell types was provided to the algorithms.…”

“…In light of the above limitation, some clustering methods attempt to reduce the dependence on predetermined parameters (such as cluster number) by starting with a relatively large number of micro clusters and gradually merging similar ones into larger clusters. For example, both SCCAF ( Miao et al 2020 ) and ADClust ( Zeng et al 2022 ) obtain initial clusters via the Louvain algorithm. Then, SCCAF iteratively updates the cluster labels by training a classifier on the clusters and evaluating the similarities between the clusters based on a confusion matrix; ADClust uses a unimodality test to evaluate the similarity between clusters and identify those that could be merged.…”

DCRELM: dual correlation reduction network-based extreme learning machine for single-cell RNA-seq data clustering

“…By combining deep learning and clustering, deep learning methods can handle high-dimensional data. ADCluster 27 can simultaneously achieve anomaly detection and clustering analysis, aiming to identify outliers in the dataset and cluster normal samples. scCAEs 28 is a scRNA-seq clustering algorithm that utilizes convolutional autoencoder embedding and soft K-means deep embedding and can learn latent clustered cell populations in space.…”