scAlign: a tool for alignment, integration, and rare cell identification from scRNA-seq data

Johansen, Nelson; Quon, Gerald

doi:10.1186/s13059-019-1766-4

Cited by 86 publications

(89 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To reduce the dimensionality of the dataset, we performed latent semantic indexing followed by singular value decomposition (Methods). Batch correction was performed using the deep neural network-based scAlign 29 to correct for technical sources of variance, including individual variation and processing method (Extended Data Fig. 1e-f, Methods).…”

Section: Chromatin States Define the Major Cell Types In The Developimentioning

confidence: 99%

Single cell epigenomic atlas of the developing human brain and organoids

Rs¹,

Wilfert

et al. 2019

Preprint

View full text Add to dashboard Cite

31Dynamic changes in chromatin accessibility coincide with important aspects of neuronal differentiation, such as 32 fate specification and arealization and confer cell type-specific associations to neurodevelopmental disorders. 33However, studies of the epigenomic landscape of the developing human brain have yet to be performed at single-34 cell resolution. Here, we profiled chromatin accessibility of >75,000 cells from eight distinct areas of developing 35human forebrain using single cell ATAC-seq (scATACseq). We identified thousands of loci that undergo 36 extensive cell type-specific changes in accessibility during corticogenesis. Chromatin state profiling also reveals 37 novel distinctions between neural progenitor cells from different cortical areas not seen in transcriptomic profiles 38 and suggests a role for retinoic acid signaling in cortical arealization. Comparison of the cell type-specific 39 chromatin landscape of cerebral organoids to primary developing cortex found that organoids establish broad 40 cell type-specific enhancer accessibility patterns similar to the developing cortex, but lack many putative 41 regulatory elements identified in homologous primary cell types. Together, our results reveal the important 42 contribution of chromatin state to the emerging patterns of cell type diversity and cell fate specification and 43 provide a blueprint for evaluating the fidelity and robustness of cerebral organoids as a model for cortical 44 development. 45 46Main text 47The diverse cell types of the human cerebral cortex (Fig. 1a) have been mostly classified based on a handful of 48 morphological, anatomical, and physiological features. Recent innovations in single cell genomics, such as single 49 cell mRNA sequencing (scRNA-seq), have enabled massively parallel profiling of thousands of molecular 50 features in every cell, uncovering the remarkable molecular diversity of cell types previously considered 51 homologous, such as excitatory neurons located in different areas of the cerebral cortex 1-6 . However, the 52 developmental mechanisms underlying the emergence of distinct cellular identities are largely unknown, as most 53 cortical neurons are generated at stages that are inaccessible to experimentation 5 . 54 55Over 60 years ago, Conrad Waddington introduced the concept of an epigenomic landscape to account for the 56 emergence of distinct cell fates 7 . In particular, chromatin state defines the functional architecture of the genome 57

show abstract

Section: Chromatin States Define the Major Cell Types In The Developimentioning

confidence: 99%

Single cell epigenomic atlas of the developing human brain and organoids

Rs¹,

Wilfert

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…clustering), followed by examination of marker gene expression in the resulting clusters. To remove the effect of disease progression on clustering, we performed, prior to clustering, cross-sample alignment [27][28][29] of the data from each brain region using scAlign (see Methods), which learns a low-dimensional manifold (i.e. the alignment space) in which cells tend to cluster in a manner consistent with their biological function independent of technical and experimental factors 29 .…”

mentioning

confidence: 99%

“…To remove the effect of disease progression on clustering, we performed, prior to clustering, cross-sample alignment [27][28][29] of the data from each brain region using scAlign (see Methods), which learns a low-dimensional manifold (i.e. the alignment space) in which cells tend to cluster in a manner consistent with their biological function independent of technical and experimental factors 29 . Importantly, after identifying clusters in the alignment space, we used the original data for subsequent analyses involving examination of gene expression, such as identifying differentially expressed genes between clusters.…”

mentioning

confidence: 99%

Molecular characterization of selectively vulnerable neurons in Alzheimer’s Disease

Leng

Eser

et al. 2020

Preprint

View full text Add to dashboard Cite

Alzheimer's disease (AD) is characterized by the selective vulnerability of specific neuronal populations, the molecular signatures of which are largely unknown. To identify and characterize selectively vulnerable neuronal populations, we used single-nucleus RNA sequencing to profile the caudal entorhinal cortex and the superior frontal gyrus -brain regions where neurofibrillary inclusions and neuronal loss occur early and late in AD, respectively -from individuals spanning the neuropathological progression of AD. We identified RORB as a marker of selectively vulnerable excitatory neurons in the entorhinal cortex, and subsequently validated their depletion and selective susceptibility to neurofibrillary inclusions during disease progression using quantitative neuropathological methods. We also discovered an astrocyte subpopulation, likely representing reactive astrocytes, characterized by decreased expression of genes involved in homeostatic functions. Our characterization of selectively vulnerable neurons in AD paves the way for future mechanistic studies of selective vulnerability and potential therapeutic strategies for enhancing neuronal resilience. MAIN TEXTSelective vulnerability is a fundamental feature of neurodegenerative diseases, in which different neuronal populations show a gradient of susceptibility to degeneration 1, 2 . Selective vulnerability at the network level has been extensively explored in Alzheimer's disease (AD) 3-5 , currently the leading cause of dementia and lacking in effective therapies. However, little is known about the mechanisms underlying selective vulnerability at the cellular level in AD, which could provide insight into disease mechanisms and lead to therapeutic strategies.The entorhinal cortex (EC), an allocortex, is one of the first cortical brain regions to exhibit neuronal loss in AD 6 . Neurons in the external EC layers, especially in layer II (also known as alpha clusters of the lamina principalis externa, abbreviated "Pre-alpha") 7 , accumulate taupositive neurofibrillary changes and die early on in the course of AD 8-13 . However, these selectively vulnerable neurons have yet to be characterized extensively at the molecular level. Furthermore, it is unknown whether there are differences in vulnerability among subpopulations of these neurons. Although rodent models of AD have offered some insights [14][15][16] , the human brain has unique features with regard to cellular physiology, composition and aging [17][18][19] , limiting the extrapoloation of findings from animal models to address selective vulnerability.Previous studies have combined laser capture microdissection with microarray analysis of gene expression 20, 21 to characterize EC neurons in AD, but focused on disease-related changes in gene expression, rather than selective vulnerability. More recently, single-nucleus RNA-sequencing (snRNA-seq) has enabled large-scale characterization of transcriptomic profiles of individual cells from post-mortem human brain tissue 22, 23 . However, snRNA-seq studies of AD p...

show abstract

“…In some cases, basic normalization 16,17 or batch correction 18,19 methods have been used to combine multiple scRNAseq datasets with limited success. Recently, several computational methods have been developed to address this challenge more comprehensively 20–25 . General steps in these methods include feature selection/dimensionality reduction and quantitative learning for matching.…”

Section: Introductionmentioning

confidence: 99%

“…Both query and reference cells are aligned in a search space projected by PCA-based dimensionality reduction and canonical correlation analysis, to transfer cluster labels through “anchors”. Among many others 23–25 , these methods have focused on individual cell level strategies when comparing a query dataset to a reference dataset, not relying on clustering results to guide supervised feature selection or cluster-level matching.…”

Section: Introductionmentioning

confidence: 99%

FR-Match: Robust matching of cell type clusters from single cell RNA sequencing data using the Friedman-Rafsky non-parametric test

Zhang

Aevermann

Bakken

et al. 2020

Preprint

View full text Add to dashboard Cite

Single cell/nucleus RNA sequencing (scRNAseq) provides an unbiased transcriptomic profiling of individual cells, and is emerging as an essential tool to unravel the phenotypic heterogeneity of cells in complex biological systems. While computational methods for scRNAseq cell type clustering have advanced, the ability to integrate datasets to identify common and novel cell types across experiments remains a challenge. Here, we introduce a cluster-to-cluster cell type matching algorithm -FR-Match -that incorporates shared information among cells to determine whether two cell type clusters are from the same underlying multivariate gene expression distribution. In validation studies using simulated data, FR-Match proved to be a stringent method that produces fewer erroneous matches of distinct cell subtypes compared to existing methods. FR-Match also has the unique ability to identify novel cell phenotypes in new datasets.When integrating two real datasets from specimens isolated from overlapping human brain regions (cortical layer one versus full thickness middle temporal gyrus), FR-Match precisely preserved the laminar characteristics of matched cell type clusters that reflect distinct neuroanatomical functions. Open-source software is available at https://github.com/JCVenterInstitute/FRmatch. A major challenge emerging from the broad application of these scRNAseq technologies is the 1 7 ability to compare transcriptional profiles across studies. In some cases, basic normalization 16,17 1 8 or batch correction 18,19 methods have been used to combine multiple scRNAseq datasets with 1 9 3 1 cell level strategies when comparing a query dataset to a reference dataset, not relying on 3 2 clustering results to guide supervised feature selection or cluster-level matching. 3Here, we present a supervised cell phenotype matching strategy, called FR-Match, for cluster-3 4 to-cluster cell transcriptome integration across scRNAseq experiments. Utilizing a priori learned 3 5 cluster labels and computationally-or experimentally-derived marker genes, FR-Match uses the 3 6Friedman-Rafsky statistical test 26,27 (FR test) to learn the multivariate distributional concordance 3 7 between query and reference data clusters in a graphical model. In this manuscript, we first 3 8 illustrate the matching properties of FR test in this scRNAseq adaptation using thorough 3 9 simulation and validation studies in comparison with other popular matching methods. We then 4 0 use FR-Match to match brain cell types defined in the full thickness of human middle temporal 4 1 gyrus (MTG) neocortex with cell types defined in a Layer 1 dissection of MTG using public 4 2 datasets from the Cell Types Database of the Allen Brain Map [https://portal.brain-4 3 map.org/atlases-and-data/rnaseq]. We also report the cell types that are consistently matched 4 4 between the two brain regions using multiple matching methods. An R-based implementation, 4 5 user guide, and ShinyApp for FR-Match are available in the open-source GitHub repository 4 6 [https://github.com/...

show abstract

scAlign: a tool for alignment, integration, and rare cell identification from scRNA-seq data

Cited by 86 publications

References 52 publications

Single cell epigenomic atlas of the developing human brain and organoids

Single cell epigenomic atlas of the developing human brain and organoids

Molecular characterization of selectively vulnerable neurons in Alzheimer’s Disease

FR-Match: Robust matching of cell type clusters from single cell RNA sequencing data using the Friedman-Rafsky non-parametric test

Contact Info

Product

Resources

About