UMAP: Uniform Manifold Approximation and Projection

McInnes, Leland; Healy, John J.; Saul, Nathaniel; Großberger, Lukas

doi:10.21105/joss.00861

Cited by 8,730 publications

(7,612 citation statements)

References 20 publications

Supporting

Mentioning

6,972

Contrasting

Unclassified

Order By: Relevance

“…Principal components showing significant overrepresentation of genes linked to the GO annotations “cell proliferation” and “cell cycle” in the top or bottom 1% quantile of loadings were removed in a fashion similar to the CCcorrect function in RaceID3. Cells were clustered using the Louvain algorithm (Blondel et al, 2008), and visualized using UMAP (McInnes and Healy, 2018). …”

Section: Star Methodsmentioning

confidence: 99%

Dermal Condensate Niche Fate Specification Occurs Prior to Formation and Is Placode Progenitor Dependent

et al. 2019

View full text Add to dashboard Cite

SUMMARY Cell fate transitions are essential for specification of stem cells and their niches, but the precise timing and sequence of molecular events during embryonic development are largely unknown. Here, we identify with 3D/4D microscopy unclustered precursors of dermal condensates (DC), signaling niches for epithelial progenitors in hair placodes. With population-based and single-cell transcriptomics, we define a molecular time-lapse from pre-DC fate specification through DC niche formation and establish the developmental trajectory as the DC lineage emerges from fibroblasts. Co-expression of downregulated fibroblast and upregulated DC genes in niche precursors reveals a transitory molecular state following a proliferation shutdown. Waves of transcription factor and signaling molecule expression then coincide with DC formation. Finally, ablation of epidermal Wnt signaling and placode-derived FGF20 demonstrates their requirement for pre-DC specification. These findings uncover a progenitor-dependent niche precursor fate and the transitory molecular events controlling niche formation and function.

show abstract

Section: Star Methodsmentioning

confidence: 99%

Dermal Condensate Niche Fate Specification Occurs Prior to Formation and Is Placode Progenitor Dependent

et al. 2019

View full text Add to dashboard Cite

show abstract

“…However, t-SNE often fails to accurately capture global structure in the data, such as distances between clusters, making interpreting higher order features of t- SNE plots difficult. While a recent method, UMAP, addresses the issue of capturing global structure in discrete datasets, it seems to still distort single cell gene expression trajectories (McInnes and Healy, 2018). …”

Section: Introductionmentioning

confidence: 99%

Visualizing and Interpreting Single-Cell Gene Expression Datasets with Similarity Weighted Nonnegative Embedding

Tamayo

Zhang

2018

Cell Systems

View full text Add to dashboard Cite

Summary High throughput single-cell gene expression profiling has enabled the definition of new cell types and developmental trajectories. Visualizing these datasets is crucial to biological interpretation, and a popular method is t-Stochastic Neighbor embedding (t-SNE), which visualizes local patterns well, but distorts global structure, such as distances between clusters. We developed Similarity Weighted Nonnegative Embedding (SWNE), which enhances interpretation of datasets by embedding the genes and factors that separate cell states on the visualization alongside the cells, and maintains fidelity when visualizing local and global structure for both developmental trajectories and discrete cell types. SWNE uses nonnegative matrix factorization to decompose the gene expression matrix into biologically relevant factors, embeds the cells, genes and factors in a 2D visualization, and uses a similarity matrix to smooth the embeddings. We demonstrate SWNE on single cell RNA-seq data from hematopoietic progenitors and human brain cells. SWNE is available as an R package at github.com/yanwu2014/swne.

show abstract

“…In this context, we consider matrices that have rows for every spot and columns for every feature (which could be genes or spatial factor activities). Utilizing t-distributed stochastic neighbor embedding (t-SNE) [14] or similar methods, such as principal component analysis (PCA) or uniform manifold approximation and projection (UMAP) [15], the number of columns of the matrices is reduced to three. The data are then rescaled into the unit cube and the rows used as coordinates in color space to colorize the spots in a spatial plot.…”

Section: Visual Summarization and Clusteringmentioning

confidence: 99%

“…However, they do not faithfully reflect the discrete count nature of RNA-Seq data. Alternatives are discrete count expression models, such as models based on the Poisson [13] or the negative binomial [14,15] distributions, with the latter being better suited for modeling over-dispersed gene expression data.ZINB-WaVE [14] offers a zero-inflated negative binomial (ZINB) regression framework for unknown covariate discovery, including gene-level covariates. Embedded in a hierarchical probabilistic ZINB model, scVI [15] utilizes deep neural networks to model gene-level responses based on both a latent space and known sample-level covariates.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Charting Tissue Expression Anatomy by Spatial Transcriptome Decomposition

Maaskola

Bergenstråhle

Jurek

et al. 2018

Preprint

View full text Add to dashboard Cite

We create data-driven maps of transcriptomic anatomy with a probabilistic framework for unsupervised pattern discovery in spatial gene expression data. Convolved negative binomial regression is used to find patterns which correspond to cell types, microenvironments, or tissue components, and that consist of gene expression profiles and spatial activity maps. Expression profiles quantify how strongly each gene is expressed in a given pattern, and spatial activity maps reflect where in space each pattern is active. Arbitrary covariates and prior hierarchies can be specified to support and leverage complex experimental designs.Transcriptomic patterns in mouse brain and olfactory bulb correspond to neuroanatomically distinct cell layers. Moreover, batch effects are successfully addressed, leading to consistent pattern inference for multi-sample analyses. On this basis, we identify known and uncharacterized genes that are spatially differentially expressed in the hippocampal field between Ammon's horn and the dentate gyrus.The analysis of spatially stratified, transcriptome-wide data [1][2][3] poses additional challenges compared to classical analysis of bulk RNA sequencing (RNA-Seq) samples. In the classical setting, annotated covariates dictate how samples are to be grouped and compared. These covariates are typically known, and often controlled for, as is the case when performing differential gene expression (DGE) analysis for bulk sequencing count data in DESeq2 [4]. In contrast, the covariates determining gene expression in space are often unknown and can change both gradually or abruptly. For example, the number of infiltrating immune cells per unit area often If the cell types underlying a sample are well characterized, it becomes analytically possible to determine mixing proportions of the cell types' known expression profiles in spatial gene expression data. But in the general case, there exists the dual discovery problems of expression profiles and spatial distributions.Two recently introduced statistical tests, Trendsceek and SpatialDE, quantify the extent of spatial variation of individual genes' expression [5,6]. Transcending individual genes, the "automatic expression histology" feature of SpatialDE implements a gene clustering approach to hidden pattern discovery. Such clustering formulations, however, appear challenged by genes that participate in multiple, spatially overlapping expression programs, and grade-of-membership formulations seem more appropriate for hidden pattern discovery.Joint analysis of multiple data sets substantially increases statistical power and sensitivity but is challenging due to the presence of batch effects. Both nonparametric [7,8] and model-based approaches are used to address such effects in the analysis of single cell RNA sequencing (scRNA-Seq) data. The model-based methods perform regression for the count data based on known sample-level covariates and allow for the discovery of unknown ones.Log-normal expression models are used for bulk [9,10] and single cell [11,12] R...

show abstract

UMAP: Uniform Manifold Approximation and Projection

Cited by 8,730 publications

References 20 publications

Dermal Condensate Niche Fate Specification Occurs Prior to Formation and Is Placode Progenitor Dependent

Dermal Condensate Niche Fate Specification Occurs Prior to Formation and Is Placode Progenitor Dependent

Visualizing and Interpreting Single-Cell Gene Expression Datasets with Similarity Weighted Nonnegative Embedding

Charting Tissue Expression Anatomy by Spatial Transcriptome Decomposition

Contact Info

Product

Resources

About