Deep soft K-means clustering with self-training for single-cell RNA sequence data

Abstract

Single-cell RNA sequencing (scRNA-seq) allows researchers to study cell heterogeneity at the cellular level. A crucial step in analyzing scRNA-seq data is to cluster cells into subpopulations to facilitate subsequent downstream analysis. However, frequent dropout events and increasing size of scRNA-seq data make clustering such high-dimensional, sparse and massive transcriptional expression profiles challenging. Although some existing deep learning-based clustering algorithms for single cells combine dimensionality reduction with clustering, they either ignore the distance and affinity constraints between similar cells or make some additional latent space assumptions like mixture Gaussian distribution, failing to learn cluster-friendly low-dimensional space. Therefore, in this paper, we combine the deep learning technique with the use of a denoising autoencoder to characterize scRNA-seq data while propose a soft self-training K-means algorithm to cluster the cell population in the learned latent space. The self-training procedure can effectively aggregate the similar cells and pursue more cluster-friendly latent space. Our method, called ‘scziDesk’, alternately performs data compression, data reconstruction and soft clustering iteratively, and the results exhibit excellent compatibility and robustness in both simulated and real data. Moreover, our proposed method has perfect scalability in line with cell size on large-scale datasets.

Integrating Deep Supervised, Self-Supervised and Unsupervised Learning for Single-Cell RNA-seq Clustering and Annotation

“…Furthermore, they are actually located on an inenarrable low-dimensional manifold. Therefore, we use the deep autoencoder representation to approximate this parameter space and estimate three groups of parameters by three output layers in a manner similar to that of the DCA and scziDesk model [ 7 , 28 ]. To take the batch effects into account, we merge the expression data matrix x with batch matrix b as the input of the encoder network.…”

“…For example, exploring cell-to-cell interactions and gene-to-gene interactions are often based on specific cell types [4,5]. In the past few years, a large volume of single-cell transcriptome unsupervised clustering algorithms have emerged based on the similarity of gene expression patterns [6][7][8][9]. However, in the absence of a unified standard, the clustering results of different algorithms usually show a little degree of overlap and even vary widely [10].…”

“…When the target dataset holds the cell types that do not exist in the source dataset, it is also difficult for ItClust to peform accurate discovery. Furthermore, neither scANVI nor ItClust considers the affinity constraint of pairwise similar cells, which can often improve the effect of cell clustering and annotation [7].…”

“…In this paper, we combine deep learning technique with statistical modeling and propose an end-to-end single-cell RNA-seq data supervised clustering and annotation framework, named scAnCluster, based on our previous unsupervised clustering works scDMFK [6] and scziDesk [7]. The three methods adhere to similar modeling concepts, namely simultaneous denoising, dimensionality reduction and clustering.…”

An Effective Tumor Classification With Deep Forest and Self-Training

“…Semi-supervised learning, a data incremental learning technology, is invented to obtain more labelled data by labeled unlabelled samples . In addition, self-training, a famous semi-supervised method, can implement high-confidance and low loss learning mechanism to improve performance of the model [6]. Moreover, common classifiers met tumor data do not achieve satisfac-tory performance since characteristics of the high-dimension and small sample size lead to over-fiiting [7].…”