Single-cell RNA sequencing data clustering using graph convolutional networks

Single-cell RNA sequencing (scRNA-seq) provides high-throughput information about the genome-wide gene expression levels at the single-cell resolution, bringing a precise understanding on the transcriptome of individual cells. Unfortunately, the rapidly growing scRNA-seq data and the prevalence of dropout events pose substantial challenges for cell type annotation. Here, we propose a single-cell model-based deep graph embedding clustering (scTAG) method, which simultaneously learns cell–cell topology representations and identifies cell clusters based on deep graph convolutional network. scTAG integrates the zero-inflated negative binomial (ZINB) model into a topology adaptive graph convolutional autoencoder to learn the low-dimensional latent representation and adopts Kullback–Leibler (KL) divergence for the clustering tasks. By simultaneously optimizing the clustering loss, ZINB loss, and the cell graph reconstruction loss, scTAG jointly optimizes cluster label assignment and feature learning with the topological structures preserved in an end-to-end manner. Extensive experiments on 16 single-cell RNA-seq datasets from diverse yet representative single-cell sequencing platforms demonstrate the superiority of scTAG over various state-of-the-art clustering methods.

Background: Cancer stem cells (CSCs) play a key role in tumor initiation, progression, and resistance to conventional therapies. These cells possess the ability to self-renew and differentiate, contributing to tumor heterogeneity and recurrence. Preclinical data indicate that targeting cancer stem cell-specific pathways could lead to more effective treatments and prevent relapse, thereby improving outcomes. However, there is limited clinical data to support this, in part due to the challenges of measuring CSCs in the clinical setting. Single-cell RNA sequencing (scRNAseq) data in combination with stemness gene expression signatures were investigated as a novel approach to detect and assess changes in CSC numbers in response to treatment. Methods: Single-cell RNAseq datasets from two clinical trials were interrogated – OPPORTUNE, a window-of-opportunity trial evaluating anastrozole vs anastrozole plus pictilisib (PI3K inhibitor) in 75 patients with ER+ HER2- early breast cancer, and FELINE, a neoadjuvant trial which compared letrozole plus placebo with letrozole plus ribociclib (CDK4/6 inhibitor) in 120 patients with ER+ HER2- early breast cancer. The primary endpoint for OPPORTUNE was the inhibition of tumor cell proliferation as measured by Ki67. The primary endpoint for FELINE was the rate of preoperative endocrine prognostic index (PEPI) score 0 after neoadjuvant endocrine therapy. Thirty-four patients from the FELINE study had available scRNAseq data for analysis; this number was 62 patients for the OPPORTUNE study. Stemness gene expression signatures with published evidence of selectivity for breast CSCs were identified from the literature and their utility in detecting CSCs compared in the datasets. Subsequently, changes in CSC fraction with treatment was assessed. Results: Eight different stemness gene signatures were identified from the published literature. Among these, four (Kim_myc, benporath_es2, Bhattacharya_hESC, and Shats_consensus) exhibited selective expression in a minority of tumor cells, with the Shats consensus signature demonstrating the highest specificity to tumor cells over non-malignant cells. Utilizing a Gaussian mixture model, we estimated that approximately 4.5% of cells within a tumor are likely stem cells. In the OPPORTUNE study, higher stemness scores were observed in luminal B compared to luminal A tumors, but there was no association with PIK3CA mutation status. Changes to Ki67 were inversely associated with the CSC fraction – tumors with higher stemness scores were less likely to achieve a complete cell cycle arrest versus endocrine-sensitive tumors. Similarly, in the FELINE trial, non-responders in the letrozole arm showed a trend towards elevated stemness scores. Conclusion: The use of stemness gene expression signatures in scRNAseq data is a feasible method to assess changes in putative CSC fraction with treatment in breast cancer. Increased stemness scores were associated with more aggressive subtypes and resistance to treatment. Citation Format: Peter Hall,Alejandro Chibly, Peter Schmid, Sarah E Pinder, Arnie Purushotham, Alastair Thompson, Steven Gendreau. Evaluation of breast cancer stem cell gene expression signatures in single-cell RNA sequencing (scRNAseq) data from the OPPORTUNE and FELINE trials, and the association with treatment resistance [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr PS13-01.

Single-cell transcriptome sequencing, often referred to as single-cell RNA sequencing (scRNA-seq), is used to measure gene expression at the single-cell level and provides a higher resolution of cellular differences than bulk RNA-seq. With more detailed and accurate information, scRNA-seq will greatly promote the understanding of cell functions, disease progression, and treatment response. Although the scRNA-seq experimental protocols have been improved very quickly, many challenges in the scRNA-seq data analysis still need to be overcome. In this chapter, we focus on the introduction and discussion of the research status in the field of scRNA-seq data normalization and cluster analysis, which are the two most important challenges in the scRNA-seq data analysis. Particularly, we present a protocol to discover and validate cancer stem cells (CSCs) using scRNA-seq. Suggestions have also been made to help researchers rationally design their scRNA-seq experiments and data analysis in their future studies.

Single-cell RNA sequencing (scRNA-seq) data are widely used to study cancer cell states and their heterogeneity. However, the tumour microenvironment is usually a mixture of healthy and cancerous cells and it can be difficult to fully separate these two populations based on transcriptomics alone. If available, somatic single-nucleotide variants (SNVs) observed in the scRNA-seq data could be used to identify the cancer population and match that information with the single cells' expression profile. However, calling somatic SNVs in scRNA-seq data is a challenging task, as most variants seen in the short-read data are not somatic, but can instead be germline variants, RNA edits or transcription, sequencing, or processing errors. In addition, only variants present in actively transcribed regions for each individual cell will be seen in the data. To address these challenges, we develop CCLONE (Cancer Cell Labelling On Noisy Expression), an interpretable tool adapted to handle the uncertainty and sparsity of SNVs called from scRNA-seq data. CCLONE jointly identifies cancer clonal populations, and their associated variants. We apply CCLONE on two acute myeloid leukaemia datasets and one lung adenocarcinoma dataset and show that CCLONE captures both genetic clones and somatic events for multiple patients. These results show how CCLONE can be used to gather insight into the course of the disease and the origin of cancer cells in scRNA-seq data. Source code is available at github.com/HaghverdiLab/CCLONE.

Most sequencing-based spatial transcriptomics (ST) technologies do not achieve single-cell resolution where each captured location (spot) may contain a mixture of cells from heterogeneous cell types, and several cell-type decomposition methods have been proposed to estimate cell type proportions of each spot by integrating with single-cell RNA sequencing (scRNA-seq) data. However, these existing methods did not fully consider the effect of distribution difference between scRNA-seq and ST data for decomposition, leading to biased cell-type-specific genes derived from scRNA-seq for ST data. To address this issue, we develop an instance-based transfer learning framework to adjust scRNA-seq data by ST data to correctly match cell-type-specific gene expression. We evaluate the effect of raw and adjusted scRNA-seq data on cell-type decomposition by eight leading decomposition methods using both simulated and real datasets. Experimental results show that data adjustment can effectively reduce distribution difference and improve decomposition, thus enabling for a more precise depiction on spatial organization of cell types. We highlight the importance of data adjustment in integrative analysis of scRNA-seq with ST data and provide guidance for improved cell-type decomposition.

Cell-surface proteins play a critical role in cell function and are primary targets for therapeutics. CITE-seq is a single-cell technique that enables simultaneous measurement of gene and surface protein expression. It is powerful but costly and technically challenging. Computational methods have been developed to predict surface protein expression using gene expression information such as from single-cell RNA sequencing (scRNA-seq) data. Existing methods however are computationally demanding and lack the interpretability to reveal underlying biological processes. We propose CrossmodalNet, an interpretable machine learning model, to predict surface protein expression from scRNA-seq data. Our model with a customized adaptive loss accurately predicts surface protein abundances. When samples from multiple time points are given, our model encodes temporal information into an easy-to-interpret time embedding to make prediction in a time-point-specific manner, and is able to uncover noise-free causal gene-protein relationships. Using three publicly available time-resolved CITE-seq data sets, we validate the performance of our model by comparing it with benchmarking methods and evaluate its interpretability. Together, we show that our method accurately and interpretably profiles surface protein expression using scRNA-seq data, thereby expanding the capacity of CITE-seq experiments for investigating molecular mechanisms involving surface proteins.

Cell type identification plays a vital role in single-cell RNA sequencing (scRNA-seq) data analysis. Although many deep embedded methods to cluster scRNA-seq data have been proposed, they still fail in elucidating the intrinsic properties of cells and genes. Here, we present a novel end-to-end deep graph clustering model for single-cell transcriptomics data based on unsupervised Gene-Cell Collective representation learning and Optimal Transport (scGCOT) which integrates both cell and gene correlations. Specifically, scGCOT learns the latent embedding of cells and genes simultaneously and reconstructs the cell graph, the gene graph, and the gene expression count matrix. A zero-inflated negative binomial (ZINB) model is estimated via the reconstructed count matrix to capture the essential properties of scRNA-seq data. By leveraging the optimal transport-based joint representation alignment, scGCOT learns the clustering process and the latent representations through a mutually supervised self optimization strategy. Extensive experiments with 14 competing methods on 15 real scRNA-seq datasets demonstrate the competitive edges of scGCOT.

Annotation of cells in single-cell clustering requires a homogeneous grouping of cell populations. There are various issues in single cell sequencing that effect homogeneous grouping (clustering) of cells, such as small amount of starting RNA, limited per-cell sequenced reads, cell-to-cell variability due to cell-cycle, cellular morphology, and variable reagent concentrations. Moreover, single cell data is susceptible to technical noise, which affects the quality of genes (or features) selected/extracted prior to clustering. Here we introduce sc-CGconv (copula based graph convolution network for single clustering), a stepwise robust unsupervised feature extraction and clustering approach that formulates and aggregates cell-cell relationships using copula correlation (Ccor), followed by a graph convolution network based clustering approach. sc-CGconv formulates a cell-cell graph using Ccor that is learned by a graph-based artificial intelligence model, graph convolution network. The learned representation (low dimensional embedding) is utilized for cell clustering. sc-CGconv features the following advantages. a. sc-CGconv works with substantially smaller sample sizes to identify homogeneous clusters. b. sc-CGconv can model the expression co-variability of a large number of genes, thereby outperforming state-of-the-art gene selection/extraction methods for clustering. c. sc-CGconv preserves the cell-to-cell variability within the selected gene set by constructing a cell-cell graph through copula correlation measure. d. sc-CGconv provides a topology-preserving embedding of cells in low dimensional space.

Based on benchmarking various methods and analyzing multiple real scRNA-seq datasets, a computational platform/workflow and a set of tips for best practices are developed for analyzing causal interactions or relationships in single cells.

Colorectal cancer (CRC) stands as the most common gastrointestinal malignancy, with the Follistatin-like 1 (FSTL1) gene associated with unfavorable outcomes in diverse cancer forms. Nonetheless, the precise mechanisms through which FSTL1 influences CRC pathogenesis and its associated multi-omics characteristics remain unclear. This study leveraged RNA sequencing and single-cell RNA sequencing (scRNAseq) data from the Gene Expression Omnibus (GEO) database to scrutinize the expression profile, prognostic significance, and clinicopathological relevance of FSTL1 in colorectal cancer. Gene set enrichment analysis (GSEA) and protein-protein interaction (PPI) network analysis were conducted to delve into the potential biological functions of FSTL1 in colorectal cancer and its associated pathways. The ESTIM algorithm and Tumor Immune Estimation Resource (TIDE) database were utilized to investigate FSTL1 expression in immune cell infiltration and immune checkpoints. Mutation characteristics of FSTL1 were elucidated using the cBioPortal database. The pRRophetic package was employed to identify potential chemotherapeutic drugs targeting FSTL1. Expression levels of FSTL1 in colon cancer and adjacent normal tissues were evaluated using the Human Protein Atlas (HPA) database. Furthermore, single-cell sequencing technology and cell communication were employed to assess the immunogenomic features of FSTL1 in the tumor microenvironment. In vitro experiments were conducted to validate the efficacy of FSTL1. FSTL1 levels were markedly higher in colorectal cancer tissues compared to healthy tissues, correlating significantly with unfavorable patient outcomes. Elevated FSTL1 expression was closely linked to key biological processes including the PI3K pathway, cell adhesion, proliferation, migration, and extracellular matrix remodeling. Furthermore, FSTL1 expression exhibited strong associations with immune cell levels and the tumor immune microenvironment in colorectal cancer. Notably, FSTL1 mutations predominantly comprised missense mutations and displayed significant correlations with various immune checkpoints and low methylation levels. Axitinib emerged as a promising targeted therapeutic option for patients with high FSTL1 expression. Analysis from the HPA database confirmed elevated FSTL1 expression in colorectal tissues compared to adjacent normal tissues. Single-cell sequencing analysis identified stromal cells as the primary source of abundant FSTL1 expression, potentially influencing tumor microenvironment remodeling by myeloid and B cells via the APP-CD74 pathway. Downregulating FSTL1 expression in colon cancer cells suppresses the proliferation, motility, and invasiveness of SW480 and HCT116 colon cancer cells. In summary, the findings of the study indicate that elevated FSTL1 expression could serve as a prognostic biomarker for unfavorable outcomes in colorectal cancer (CRC) diagnosis and may also identify potential targets for immunotherapy in CRC.

MLSpatial: A machine-learning method to reconstruct the spatial distribution of cells from scRNA-seq by extracting spatial features

Gastric cancer (GC) is a common malignant tumor worldwide, characterized by complex biological processes. The distribution of various cell types and gene expression profiles in the GC microenvironment remains unclear. This study uses single-cell RNA sequencing to explore gene expression patterns and identify differentially expressed genes in GC samples, offering new insights into cellular diversity and potential molecular mechanisms. We conducted temporal and clustering analyses with single-cell sequencing, followed by Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses to clarify their functions. Using machine learning, we identified relevant genes to create highly accurate prediction models. Additionally, ssGSEA analysis provided detailed insights into the immunosuppressive tumor microenvironment, revealing complex gene expression interactions and diverse immune infiltrates in cancer. Correlation analysis highlighted TIMP1 as having significant prognostic value across different immune cell subtypes. Single-cell RNA sequencing revealed the cellular landscape and gene expression profiles of the GC microenvironment, offering crucial data on how cell heterogeneity is regulated in relation to the tumor microenvironment. Moreover, new insights into the expression levels of AGT, INHBA, and TIMP1 showed distinct sex-biased gene functions within the tumor microenvironment. These findings enhance our understanding of the molecular mechanisms associated with gastric cancer development and may lay the groundwork for identifying novel therapeutic targets and diagnostic strategies.

Analysis of single-cell RNA sequencing (scRNA-seq) data of immune cells from the tumor microenvironment (TME) may identify tumor progression biomarkers. This study was designed to investigate the prognostic value of differentially expressed genes (DEGs) in intrahepatic cholangiocarcinoma (ICC) using scRNA-seq. We downloaded the scRNA-seq data of 33,991 cell samples, including 17,090 ICC cell samples and 16,901 ICC adjacent tissue cell samples regarded as normal cells. scRNA-seq data were processed and classified into 20 clusters. The immune cell clusters were extracted and processed again in the same way, and each type of immune cells was divided into several subclusters. In total, 337 marker genes of macrophages and 427 marker genes of B cells were identified by comparing ICC subclusters with normal subclusters. Finally, 659 DEGs were obtained by merging B cell and macrophage marker genes. ICC sample clinical information and gene expression data were downloaded. A nine-prognosis-related-gene (PRG) signature was established by analyzing the correlation between DEGs and overall survival in ICC. The robustness and validity of the signature were verified. Functional enrichment analysis revealed that the nine PRGs were mainly involved in tumor immune mechanisms. In conclusion, we established a PRG signature based on scRNA-seq data from immune cells of patients with ICC. This PRG signature not only reflects the TME immune status but also provides new biomarkers for ICC prognosis.

Hypertrophic cardiomyopathy (HCM) is a common inherited heart disease characterized by left ventricular hypertrophy and excessive myocardial contraction. Endothelial cells (ECs) play a key role in the pathogenesis of HCM, and their secreted growth factors regulate the growth and function of cardiomyocytes. However, the specific molecular mechanisms and biomarkers associated with endothelial dysfunction in HCM remain unclear. Single-cell RNA sequencing (scRNA-seq) data were downloaded from the Gene Expression Omnibus (GEO) database (GSE255296), and bulk RNA sequencing data were obtained from GSE249925 and GSE180313. The scRNA-seq data were processed using the Seurat package for cell type annotation and batch effect removal. Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis was performed, and pathway activity was evaluated using the AUCell algorithm. SCENIC was applied to construct transcription factor-specific regulatory networks. Differential gene expression analysis was conducted using the limma package, and key genes were determined by intersecting differentially expressed genes (DEGs) with EC-specific highly expressed genes. The diagnostic efficacy of these key genes was assessed via receiver operating characteristic (ROC) curve analysis. The mRNA expression of genes in HUVECs treated with or without angiotensin II (Ang II) was assessed using qPCR. Cell viability was determined using the CCK-8 assay, and cell invasive ability was evaluated through Transwell assays to investigate the regulatory effect of TMTC1 overexpression on Ang II-induced injury in HUVECs. Seven major cell types were identified, with cardiomyocytes and ECs being the main cell populations in HCM. KEGG enrichment analysis revealed significant activation of pathways such as the transforming growth factor-beta (TGF-β) signaling pathway in ECs. High-activity transcription factors NFIB, MECOM, and FLI1 were identified in ECs, whose target genes were enriched in processes related to cell morphogenesis. Three key genes-TMTC1, RPS6KA2, and F8-were selected based on differential expression and EC specificity. These genes were lowly expressed in HCM and exhibited high diagnostic efficacy, with area under the curve (AUC) values > 0.7. Ultimately, the CCK-8 assay results indicated thatAng II inhibited the viability of HUVECs, and this inhibitory effect could be reversed by the overexpression of TMTC1 (oe-TMTC1). Furthermore, Transwell assays indicated that Ang II impaired cellular migration, and this impairment could be restored by oe-TMTC1. The identified key genes and pathways provide new targets for therapeutic intervention and contribute to a deeper understanding of the pathogenesis of HCM.

Research on cell differentiation facilitates a deeper understanding of the fundamental processes of life, elucidates the intrinsic mechanisms underlying diseases such as cancer, and advances the development of therapeutics and precision medicine. Existing methods for inferring cell differentiation trajectories from single-cell RNA sequencing (scRNA-seq) data primarily rely on static gene expression data to measure distances between cells and subsequently infer pseudotime trajectories. In this work, we introduce a novel method, scGRN-Entropy, for inferring cell differentiation trajectories and pseudotime from scRNA-seq data. Unlike existing approaches, scGRN-Entropy improves inference accuracy by incorporating dynamic changes in gene regulatory networks (GRN). In scGRN-Entropy, an undirected graph representing state transitions between cells is constructed by integrating both static relationships in gene expression space and dynamic relationships in the GRN space. The edges of the undirected graph are then refined using pseudotime inferred based on cell entropy in the GRN space. Finally, the Minimum Spanning Tree (MST) algorithm is applied to derive the cell differentiation trajectory. We validate the accuracy of scGRN-Entropy on eight different real scRNA-seq datasets, demonstrating its superior performance in inferring cell differentiation trajectories through comparative analysis with existing state-of-the-art methods.