UMAP

Scatter Gene:

Scatter Point Size:

Filter Gene Expression (log2(counts + 1)):

Scatter Filter Category:

Download Scatter Plot

Meta Color:

Meta Point Size:

Meta Label:

Meta Numeric Transform

None

log2

Meta Filter Category:

Download Meta Plot

Tables

Gene Table Grouping(s)

Metadata Table Groupings

Expression Plot

Gene(s):

Plot Height:

Value:

Facet on:

Color on:

Number of columns:

Filter Category:

Filter On:

Minimum # Cells in Group:

Download Scatter Plot

In Situ Projection

Genes:

Plot Height:

Filter category:

Filter on:

Show data table?

Yes No

Facet UMAP

Facet On:

Color On:

Filter Category:

Filter On:

Point Size:

Plot Height:

Download Scatter Plot

Dotplot

Genes:

Plot Height:

Group by (two max):

Filter category:

Filter on:

Download Dot Plot

Gene Differential Expression Tests

As of 2021-01-06 we have swapped out the "pseudo-Bulk" differential testing for a simpler testing system: the scran findMarkers tool. Why? Well frankly as we used the "pseudo-Bulk" results more and more we regularly found odd results with low p values that were being driven by very few cells with high expression. The findMarkers tool, in comparison, is more robust and reliable. It uses the wilcox test. Genes are required to 1. be expressed in >20% of the cells in the group (e.g. Cluster or Cell Type) and 2. be expressed at least twice as high against the other cells (e.g. GeneA has a mean expression of "100" in cells in ClusterX and "40" in cells-not-in-ClusterX

What is that "AUC" column? Area Under the Curve. Instead of reporting differential expression, which can be skewed by a low number of cells with very high expression, we report the power of the gene of interest to distinguish between the base group (e.g. Rods) versus the comparison group (e.g. Cones). An AUC of 1 mean thats the marker can perfectly (100%) distinguish cells between the two groups with the marker. AUC of 0 (or missing) means that the gene has no power.

Why are all the PValues and FDR the same for each gene? The p value is calculated at the gene level, while the AUC is calculated for each pair wise test. What that means is that if you test Rods vs not-Rods, the FDR for GeneA could be 0.01. Which means that GeneA "significantly" (well actually means you can reject the null hypothesis that....) distinguishes Rods against all other cells. If you want to dive into Rods vs some-specific-Cell-Type then use the AUC, which are calculated for each tissue - tissue combination.

Search by:

Genes:

Base:

Against:

Download all results as csv

Data

The codebase for the creation of the scEiaD dataset is on github

If size not given, it is less than 1 GB

Run plae Locally

If you have 200GB of free hard drive space, you can run plae on your own computer. Installation instructions are available in our Github repository (this is the codebase for the app you are using now).

Seurat Objects

All (~20 GB)

AnnData Objects

All (~4 GB)

Diff Testing Results

All Diff Results (~1.1GB)

CellType (Predict) against CellType testing

CellType against CellType (Predict)

Cluster against Cluster

Metadata

Cell Metadata

Counts

Kallisto counts, R sparse matrix (~1.4 GB)

Kallisto counts, no filtering, R sparse matrix (~1.4 GB)

plae v0.74

PLatform for Analysis of scEiad

What is scEiaD?

single cell Eye in a Disk

The light-sensitive portion of the eye is the retina. The retina itself is not a monolithic tissue - there are over 10 major cell types. The cones and rods which convert light into signal are supported by a wide variety of neural cell types with distinct roles in interpretting and transmitting the visual signal to the brain. Behind the retina is the RPE and vasculature, which supports the high energetic needs of the rods and cones. scEiaD is a meta-atlas that compiles 1.2 million single-cell back of the eye transcriptomes across 28 studies, 18 publications, and 3 species. Deep metadata mining, rigorous quality control analysis, differential gene expression testing, and deep learning based batch effect correction in a unified bioinformatic framework allow the universe of retina single cell expression information to be analyzed in one location.

tldr

You can look up gene expression by retina cell type across loads of different studies, three organisms, and multiple developmental stages.

Preprint of the data creation and benchmarking now on bioRxiv!

Licensing

This work is released under the CC0 license

Data Sources

Citation	PMID	SRA Accession	organism	Platform	Count	Post QC Count	Labels
Voigt AP, Whitmore SS, Flamme- ...	31075224	SRP194595	Homo sapiens	10xv3	146353	242	Yes
Cowan CS, Renner M, De Gennaro ...	32946783	EGAD00001006350	Homo sapiens	10xv2	99878	44679	Yes
Lu Y, Shiau F, Yi W, Lu S et a ...	32386599	SRP223254	Homo sapiens	10xv2	88202	74624	Yes
Yan W, Peng YR, van Zyl T, Reg ...	32555229	SRP255195	Homo sapiens	10xv3	78706	16	Yes
Lu Y, Shiau F, Yi W, Lu S et a ...	32386599	SRP151023	Homo sapiens	10xv2	47217	43742	Yes
Voigt AP, Whitmore SS, Mulfaul ...	32531351	SRP257883	Homo sapiens	10xv3	34044	3882	Yes
Yan W, Peng YR, van Zyl T, Reg ...	32555229	SRP255195	Homo sapiens	10xv2	32195	15262	Yes
Sridhar A, Hoshino A, Finkbein ...	32023475	SRP238587	Homo sapiens	10xv2	23682	16616	No
Voigt AP, Mulfaul K, Mullin NK ...	31712411	SRP218652	Homo sapiens	10xv3	16633	1539	Yes
Lukowski SW, Lo CY, Sharov AA ...	31436334	E-MTAB-7316	Homo sapiens	10xv2	13803	10158	Yes
Menon M, Mohammadi S, Davila-V ...	31653841	SRP222001	Homo sapiens	10xv2	8737	680	Yes
Menon M, Mohammadi S, Davila-V ...	31653841	SRP222958	Homo sapiens	DropSeq	8016	1012	Yes
Lu Y, Shiau F, Yi W, Lu S et a ...	32386599	SRP170761	Homo sapiens	10xv2	7452	5372	No
Voigt AP, Binkley E, Flamme-Wi ...	32069977	SRP238409	Homo sapiens	10xv3	1696	28	No
		OGVFB_Hufnagel_iPSC_RPE	Homo sapiens	10xv2	697	694	Yes
Hu Y, Wang X, Hu B, Mao Y et a ...	31269016	SRP125998	Homo sapiens	SMARTSeq_v2	16	11	No
Peng YR, Shekhar K, Yan W, Her ...	30712875	SRP158528	Macaca fascicularis	10xv2	112058	109275	Yes
Clark BS, Stein-O'Brien GL, Sh ...	31128945	SRP158081	Mus musculus	10xv2	185546	176668	Yes
Tabula Muris Consortium., Over ...	30283141	SRP131661	Mus musculus	10xv2	78567	64314	Yes
Yan W, Laboulaye MA, Tran NM, ...	32457074	SRP259930	Mus musculus	10xv2	39596	38254	Yes
		SRP257758	Mus musculus	10xv2	37006	36010	No
Shekhar K, Lapan SW, Whitney I ...	27565351	SRP075719	Mus musculus	DropSeq	36295	24772	Yes
Tran NM, Shekhar K, Whitney IE ...	31784286	SRP212151	Mus musculus	10xv2	32526	30395	Yes
Heng JS, Hackett SF, Stein-O'B ...	31843893	SRP200499	Mus musculus	10xv2	22344	21244	Yes
Macosko EZ, Basu A, Satija R, ...	26000488	SRP050054	Mus musculus	DropSeq	12831	11093	Yes
Lehmann GL, Hanke-Gogokhia C, ...	32196081	SRP216903	Mus musculus	10xv2	10543	10275	No
Fadl BR, Brodie SA, Malasky M, ...	33088174	SRP269635	Mus musculus	10xv2	9019	7832	No
Buenaventura DF, Corseri A, Em ...	31260032	SRP200599	Mus musculus	10xv2	7415	7400	No
Lo Giudice Q, Leleu M, La Mann ...	31399471	SRP168426	Mus musculus	10xv2	5575	5355	No
O'Koren EG, Yu C, Klingeborn M ...	30850344	SRP186407	Mus musculus	10xv2	3793	2596	No
Clark BS, Stein-O'Brien GL, Sh ...	31128945	SRP158081	Mus musculus	SMARTSeq_v2	864	790	No
Lo Giudice Q, Leleu M, La Mann ...	31399471	SRP186396	Mus musculus	SMARTSeq_v2	800	797	No
Fadl BR, Brodie SA, Malasky M, ...	33088174	SRP269634	Mus musculus	10xv2	399	386	No
Shekhar K, Lapan SW, Whitney I ...	27565351	SRP075720	Mus musculus	SMARTSeq_v2	384	363	No
Shekhar K, Lapan SW, Whitney I ...	27565351	SRP073242	Mus musculus	SMARTSeq_v2	291	236	No
Dharmat R, Kim S, Liu H, Fu S ...	bioRxiv 774950	SRP220355	Mus musculus	SCRBSeq	3	3	No

scEiaD Curated Published Cell Type Labels

CellType	Species	Studies	Count
Retinal Ganglion Cells	HS, MF, MM	11	88253
Amacrine Cells	HS, MF, MM	14	67036
Rods	HS, MF, MM	12	54622
Muller Glia	HS, MF, MM	13	45484
Early RPCs	MM	1	28643
Bipolar Cells	HS, MF, MM	13	27535
Late RPCs	MM	1	21643
RPCs	HS	2	17927
Cones	HS, MF, MM	12	15738
Neurogenic Cells	HS, MM	3	10604
Rod Bipolar Cells	HS, MM	2	9164
B-Cell	HS, MM	4	8656
Horizontal Cells	HS, MF, MM	11	8323
Photoreceptor Precursors	HS, MM	3	7123
T-Cell	HS, MM	4	5443
Endothelial	HS, MF, MM	10	4565
Fibroblasts	HS, MM	5	1879
Red Blood Cells	HS, MM	3	1873
Pericytes	HS, MF, MM	6	1608
Amacrine/Horizontal Precursors	HS	2	1450
Astrocytes	HS, MM	4	1197
Microglia	HS, MF, MM	9	602
Macrophage	HS	2	561
Vein	HS	1	490
Monocyte	HS	1	317
Schwann	HS	2	288
Melanocytes	HS	3	259
Choriocapillaris	HS	1	225
Natural Killer	HS	1	168
Artery	HS	1	167
Mast	HS	2	101
Smooth Muscle Cell	HS	1	45

Labelled cell types from published papers were pulled, where possible, from a combination of the Sequence Read Archive (SRA), lab web sites, and personal correspondence, then adjusted to be consistent (e.g. MG to Muller Glia) between all studies.

scEiaD Machine Learned Cell Type Labels

CellType	Species	Studies	Count
Rods	HS, MF, MM	29	141472
Amacrine Cells	HS, MF, MM	27	111445
Retinal Ganglion Cells	HS, MF, MM	28	63915
Muller Glia	HS, MF, MM	28	60858
Bipolar Cells	HS, MF, MM	26	56770
RPCs	HS, MF, MM	21	41290
Early RPCs	HS, MF, MM	13	35984
Late RPCs	HS, MM	9	34762
Neurogenic Cells	HS, MF, MM	13	34078
Cones	HS, MF, MM	28	26460
Photoreceptor Precursors	HS, MF, MM	9	22648
Horizontal Cells	HS, MF, MM	22	17793
Rod Bipolar Cells	HS, MM	16	12438
Fibroblasts	HS, MF, MM	21	5530
Pericytes	HS, MF, MM	17	5514
Endothelial	HS, MF, MM	22	5344
Amacrine/Horizontal Precursors	HS, MF, MM	8	3072
Red Blood Cells	HS, MF, MM	24	2554
RPE	HS, MF, MM	15	2101
Astrocytes	HS, MF, MM	18	2089
Microglia	HS, MF, MM	24	2013
T-Cell	HS, MF, MM	15	1311
Vein	HS, MM	8	1058
Monocyte	HS, MM	17	823
Schwann	HS, MF, MM	12	790
Macrophage	HS, MM	12	677
B-Cell	HS, MF, MM	17	439
Mast	HS, MM	12	393
Melanocytes	HS, MF, MM	10	327
Natural Killer	HS, MM	7	225

The labels above were used to create a machine learning modeled which was used to relabel all* cells in the scEiaD (*above a confidence threshold of 0.5).

Using and Extending plae and the scEiaD

All Links are External

Analyses	Colab	Bash	Web Guide
Using plae			Go
Using scEiaD Seurat object			Go
UMAP projection of your data on scEiaD	Go
Auto cell type label your data	Go	Go

Contact

If you have questions about scEiaD dataset or the plae application, please contact David McGaughey, Ph.D .

Otherwise the National Eye Institute's Office of Science Communications, Public Liaison and Education responds directly to requests for information on eye diseases and vision research in English and Spanish. We cannot provide personalized medical advice to individuals about their condition or treatment.

2020@nei.nih.gov

Phone: 301-496-5248 — English and Spanish

Mail: National Eye Institute

Information Office

31 Center Drive MSC 2510

Bethesda, MD 20892-2510

Change log

0.74 (2021-08-12): Remove broken link, fix bug in Expression Plot that was making average expression far too low.

0.73 (2021-06-10): Allow for UMAP plot to show all values (filter now starts at >=1)

0.72 (2021-04-29): Added more content and a table organzation to the Info -> Analysis... section

0.71 (2021-04-14): scEiaD preprint on bioRxiv! Added a filter in the "exp plot" section to remove data points with user-selected (default 50) minimum cells. I was finding that data points (e.g. cell type - study) with low N would often have "outlier" results. Added a new section to the web page - Analyses (under the "Info" tab)!

0.70 (2021-03-22): New scEiaD built with corrected fastq file sets (potential bug in 10x bamtofastq tool resulted ina few datasets getting scrambled barcodes). Removed a macaque dataset (SRR7733526) with odd behavior (clustering in 2D UMAP space largely alone). Tweaked the UMAP 2D gene view with a darker "background" cell color scheme to reduce "over-emphasis" on cells with low expression of a gene. CPM replaced with counts as some odd behaviour was detected in some genes in the UMAP view where there was high "background" expression. Counts have more consistent behavior. Removed hard filter that tossed cells with >2500 detected genes.

0.60 (2021-02-08): New scEiaD built with more studies. Removed several retinal organoid datasets that had snuck in. Added a filter option for the diff searching to search, for example, one cluster directly against another cluster.

0.52 (2021-01-13): Adding "missing" genes (we had only retained genes which were expressed in all three species, which naturally led to many genes (some important, like OPN1MW) to be dropped. That has been fixed. Tweak "Expression Plot" dot size to prevent crazy tiny point sizes.

0.51 (2021-01-06): Hello 2021! Adam Gayoso kindly pointed out that I was using scVI in a non-optimal manner, so I updated the scVI modeling to match their recommend "scArches" parameters. This (fortunately for my sanity) only subtly changes the downstream result. The more significant change is that we have totally changed the diff testing section to use the scran findMarkers test instead of our complicated and compute expensive pseudo-Bulk testing which continually gave odd results.

0.50 (2020-12-31): Goodbye 2020! Major update to the scVI-based UMAP projection which improves data quality. Removed non-tissue samples (e.g. organoid/cell lines). They will be added back later once I figure out a logical/simple way to do it. Fixed major bug in QC filtering which failed to remove high mitochondrial count (likely apoptosing cells) cells. Dot plot tweaked to improve relative dot sizes. Cowan et al. dataset added.

0.43 (2020-11-09): Downloadable diff results added to "Data." The diff results reactive data table now has a "Download all ..." button which replaces the "CSV" button that only downloaded the viewable data (100 max).

0.42 (2020-10-16): Alt text added to each button, tweaked UMAP-Tables layout again. Slide logo added. Site went public at the version on 2020-11-02!

0.41 (2020-10-06): Download buttons added for each plot.

0.40 (2020-10-05): UI and text labels tweaked in UMAP-Tables to improve tab selection order. Dot plot given a bar plot to show category size. Error handling improved when user fails to provide a category value to filter on. Data Table help button added. Help buttons moved to bottom of page with consistent visual - tabbing order. Colors of UI elements tweaked to improve contrast.

0.39 (2020-09-18): Fixed calculation error in dotplot where expression not scaled by number of cells in grouping variable.

0.38 (2020-09-02): Contact section and footer added for compliance.

0.37 (2020-08-24): Help pop section populated with text. Put white halo back around text in UMAP - Meta section. Loading circles added to plots. Row names added to tables. Diff testing filtered to only return results with FDR < 0.05 and abs(logFC) > 0.5.

0.36 (2020-08-17): Data download section added. Change log moved to separate section. CSS tweaked to show links in blue. First overview table updated to improve contrast. UMAP plots axis fixed.

0.35 (2020-08-14): Moved Overview tables to html for improved rendering and switched over color-blind friendly palette. Temporarily removed Temporal Plotting section. Improved filtering for Facet Plot. DotPlot plotting fixed and improved. Back-end server.R code moved into separate functions. Colors fixed so they stay consistent when filtering/subsetting the plots. Site now starts from scratch in under 5 seconds with improved fst-based data loading and pre-calculating more operations.

0.34 (2020-08-03): Fixed issue with TabulaMuris labels not appearing. Scanned app with koa11y for 508 compliance - changed headers from h2 to h1 to comply.

0.33 (2020-07-30): Exp plot can now take space or comma separated Genes as input. User can selected number of columns in Exp Plot. Diff Table formatting improved with rounding and PB_Test can be selected as a drop down now in the data table search.

0.32 (2020-07-29): In situ Projection viz added courtesy of Zachary Batz! It's a simulated cross section of the retina with each cell type colored by intensity of scRNA expression! Move table draw button under filtering in UMAP - Tables. Sort Diff Exp results by FDR. Filtering on numeric column now returns slider UI. Remove super dangerous ability to create faceted plots on numeric values.

0.31 (2020-07-24): Re-created scEiaD with better internal (Hufnagel) transwell RPE labelling (there are roughly two groups - mature RPE with high TTR expression and less (?) mature RPE with lower TTR), removal of the SRP166660 study as it was *all* non-normal (injured retina) (confirmed with correspondence with Dr. Poche), removed the pan RGC CellType labelling for the SRP212151 as I see post-hoc that there are LOADS of non-RGC cells. Did the same for SRP186407, which has substantial non-microglia. Generally, FACS != 100% celltype purity. Added differential testing against all Tabula Muris cell types. Removing clusters/cells with high doublet scores. Added cell cycle phase (G1/G2M/S) assignment. More study level metadata.

0.30 (2020-07-20): Huge update. Hundreds of thousands of cells added. The Tabula Muris project data (pan mouse) has been added to faciliate non-eye comparison. Filtering options added to most of the plotting views to allow for quick slicing into this huge dataset. Differential expression testing totally reworked - now uses "pseudoBulk" approach to better utilize the large number of studies we have.

0.23 (2020-06-16): Remove low N cell type from diff expression tables, tweak Overview with spacing alterations and updated text.

0.22 (2020-06-15): Added expression plot by user selected groups plot view. Fixed bug in mean cpm expression calculation for Viz -> UMAP - Table gene tables

0.21 (2020-06-15): Added subcluster diff testing tables, temporal gene expression by celltype plot section.

0.20 (2020-06-06): New 2D UMAP projection that includes the full Yu - Clark Human scRNA dataset. Added tables to "Overview" section showing data stats. Added "filtering" functionality to UMAP plot section.