Overview of RNA datasets for tissue and cell type enrichment analysis

[ME/CFS Science Blog]

I thought it would be useful to make an overview of RNA datasets that we can match to the DecodeME DNA results. This could give us clues to which tissues or cell types are potentially involved in ME/CFS. I've made an overview of datasets below and plan to update it as we find new ones.

The Chan Zuckerberg Initiative’s CellxGene application is a nice respository where researchers can upload their data.
https://cellxgene.cziscience.com/

In the posts below, I'll share the results ME/CFS analyses that have been done with these datasets with a link to the relevant thread for more info. The current thread is only meant to provide an overview of all these analyses and discuss RNA datasets in general (in-depth discussion for a particular dataset or results are best reserved for the thread with the analysis itself).

Name	Animal	location	Population	Source
Genotype-Tissue Expression (GTEx) The GTEx Consortium, Science 2020. NIH funded.	Human	Entire body	Bulk-RNA 54 tissues 17,382 samples 838 postmortem donors	FUMA: “Gene expression values are log2 transformed average RPKM per tissue type after winsorized at 50 based on GTEx RNA-seq data. Tissue expression analysis is performed for 30 general tissue types and 53 specific tissue types separately. MAGMA was performed using the result of gene analysis (gene-based P-value) and tested for one side (greater) with conditioning on average expression across all tissue types.” https://fuma.ctglab.nl/tutorial#snp2gene
Human Brain Cell Atlas v1.0 Siletti et al. 2023 Linnarsson lab Karolinska Institute	Human	Brain	3 adult donors ~3 million cells 31 superclusters, 461 clusters	https://cellxgene.cziscience.com/collections/283d65eb-dd53-496d-adb7-7570c7caa443
DESCARTES Cao et al. 2020 Shendure lab Allen Institute	Human Fetus	Entire body	>110 fetal samples 15 organs ~4 million cells	https://cellxgene.cziscience.com/collections/c114c20f-1ef4-49a5-9c2e-d965787fb90c
Braun et al. 2023 Linnarsson lab Karolinska Institute	Human Fetus	Brain	26 brain specimens 111 distinct biological samples First semester (5 – 14 weeks post-conception) ~ 1.6 million cells	https://cellxgene.cziscience.com/collections/4d8fed08-2d6d-4692-b5ea-464f1d072077
Seeker 2023 CZI Cell Science	Human	Brain	White matter ~ 0.048 million cells	https://cellxgene.cziscience.com/collections/9d63fcf1-5ca0-4006-8d8f-872f3327dbe9
Herring et al. 2022 (GSE168408) Lister lab	Human	Prefrontal Cortex	26 postmortem samples spanning 6 developmental stages: Fetal, Neonatal, Infancy, Childhood, Adolescence, and Adult. ~0.15 million cells	https://brain.listerlab.org/ https://console.cloud.google.com/storage/browser/neuro-dev/Processed_data;tab=objects?prefix=&forceOnObjectsSortingFiltering=false
Allen Brain Atlas Sunkin et al. 2012 Allen Institute	Human	Brain	Two brains (older dataset)	https://atlas.brain-map.org/ https://human.brain-map.org/static/download
Franke lab data Fehrmann et al. 2015	Human, mouse, rat	Entire body	Affymetrix methodology 33 tissues 77,840 samples GPL96: Homo sapiens: 17,309 GPL570: Homo sapiens: 37,427 GPL1261: Mus musculus: 17,081 GPL1355: Rattus norvegicus: 6,032	Used by the DEPICT tool and the Finucane et al. 2018 paper. Is only available on a Google Cloud bucket where you have to pay for download costs. https://console.cloud.google.com/storage/browser/broad-alkesgroup-public-requester-pays/LDSCORE/LDSC_SEG_ldscores;tab=objects
Saunders et al. 2018 Dropviz Harvard	Mouse	Brain	~0.69 million cells	http://dropviz.org/
Mouse Brain Atlas Zeisel et al. 2015-2018 Linnarsson lab Karolinska Institute	Mouse	Brain	9,970 cells 24 level 1 and 149 level 2 cell types Cortex, Hippocampus Hypothalamus Midbrain, Oligodendrocytes, Striatum	https://github.com/NathanSkene/MAGMA_Celltyping Used by Skene 2018, Olislagers 2018
ImmGen	Mouse	292 immune cell types	Array-based methodology Older version used in Finucane 2018: phase 1 (GSE15907) and phase 2 (GSE37448) data.	https://www.immgen.org/Databrowser19/DatabrowserPage.html Older version by Finucane 2018, available in Google Icloud bucket.
Chiou et al. 2023 Allen Institute BRAIN Initiative	Rhesus macaque	Brain	5 animals 30 brain regions ~2.58 million cells	https://cellxgene.cziscience.com/collections/8c4bcf0d-b4df-45c7-888c-74fb0013e9e7
Zhang et al. 2023 Allen Institute BRAIN Initiative	Mouse	Brain	4 mice ~8.4 million cells	https://cellxgene.cziscience.com/collections/0cca8620-8dee-45d0-aef5-23f032a5cf09

 

[ME/CFS Science Blog]

Genotype-Tissue Expression (GTEx)
This analysis was done in the DecodeME preprint using the MAGMA tool on the FUMA website. The preprint only posted some of the most significant regions, but the entire plot looks something like this if you upload the DecodeME data. All significant tissue types are in the brain.




GTEx is an NIH funded project with an explanation and background provided in this paper:
https://www.science.org/doi/10.1126/science.aaz1776

Here's the website with the GTEx data available for download:

GTEx Portal

The Genotype-Tissue Expression (GTEx) project is an ongoing effort to build a comprehensive public resource to study tissue-specific gene expression and regulation. Samples were collected from 53 non-diseased tissue sites across nearly 1000 individuals, primarily for molecular assays including...

gtexportal.org

The own used for DecodeME was an older version, version 8, while the most recent one is version 11. But I don't think the results would change much by redoing the analysis. It also handy that the FUMA website prepared the data in suitable format. It says:

“Gene expression values are log2 transformed average RPKM per tissue type after winsorized at 50 based on GTEx RNA-seq data. Tissue expression analysis is performed for 30 general tissue types and 53 specific tissue types separately. MAGMA was performed using the result of gene analysis (gene-based P-value) and tested for one side (greater) with conditioning on average expression across all tissue types.”

Source: https://fuma.ctglab.nl/tutorial

You can download it from their website:

Functional Mapping and Annotation of Genome-wide association studies

 

[ME/CFS Science Blog]

GTEx6 + Franke Lab
An older version of the GTEx (version 6) was combined with RNA data from Franke's lab in the study by Hilary Finucane et al. 2018. This is the paper where they introduced the LDSC-SEG approach to testing cell type enrichment and where they tested multiple diseases and traits.

Heritability enrichment of specifically expressed genes identifies disease-relevant tissues and cell types - Nature Genetics

A new method tests whether disease heritability is enriched near genes with high tissue-specific expression. The authors use gene expression data together with GWAS summary statistics for 48 diseases and traits to identify disease-relevant tissues.

www.nature.com

The Franke Lab data refers to a collection of older human, mouse and rat RNA datasets that were available on the (former) online tool DEPICT. It was grouped to 33 tissues. More info about it, is available in the supplementary material of Finucane et al. 2018 and this paper from Lude Franke's group (Fehrman et al. 2015):

Gene expression analysis identifies global gene dosage sensitivity in cancer - Nature Genetics

Rudolf Fehrmann, Lude Franke and colleagues report a method for capturing the variation present within mammalian transcriptomes in a limited number of 'transcriptional components' and demonstrate widespread correlation between gene copy number and expression levels. The method allows for the...

www.nature.com

I and Trafalmadorian97 have tried to replicate the Finucane et al. 2018 pipeline but with the DecodeME dataset. The results look like this, again only highlighting the brain tissue:s


Source: S-LDSC - ME/CFS Biostatistics Home

I also compared the ME/CFS analysis to those from other diseases provided in supplementary table 6 of the Finucane paper. It looks like this:

Thread:

Post in thread 'Heritability enrichment of specifically expressed genes identifies disease-relevant tissues and cell types, 2018, Finucane et al.'

Friday at 8:46 PM

One interpretation could be that what sticks out in the (still underpowered) DecodeME results is what every ME/CFS patient has in common. So this may point to the neural networks that produce sickness behavior symptoms such as fatigue. Hence the strong overlap with fibromyalgia.

Behind there may be multiple causes for the activation/importance of those networks, with different (immune) pathways in subgroups that are currently too small to be picked up by DecodeME.

• ME/CFS Science Blog

• 

 

[ME/CFS Science Blog]

Human Brain Cell Atlas v1.0
This is currently the most extensive dataset on brain cells in humans. Published in 2023 by the lab of Sten Linnarsson at the Karolinska Institute, one of the important teams that have been making multiple RNA datasets over the years.

The relevant paper is Siletti et al. 2023:

https://www.science.org/doi/10.1126/science.add7046

And the data is available here:

Cellxgene Data Portal

Find, download, and visually explore curated and standardized single cell datasets.

cellxgene.cziscience.com

There are different ways to analyse this data. The FUMA website standardizes gene expression per dissection, creating multiple smaller datasets. Paolo Maccalini used this in his meta-analysis of DecodeME and the Million Veterans Program (MVP) dataset. The results mainly pointed to eccentric medium spiny neurons. His paper is discussed here:
https://s4me.info/threads/biologica...encing-of-me-cfs-2026-maccallini-et-al.50225/

The FUMA approach of working per dissection means that the gene expression per cell type is influenced by whatever else is in the dissection dataset. This makes it more variable and perhaps less robust. When I applied it to DecodeME only (not the meta-analysis) or used a different MAGMA window size, I got different results.
https://s4me.info/threads/biologica...026-maccallini-et-al.50225/page-6#post-699621

The Duncan et al. 2025. paper standardizes gene expression by the entire dataset. Trafalmadorian97 applied this analysis to the DecodeME dataset and got the following results, also finding eMSN as the top hit.

Source: https://trafalmadorian97.github.io/...nalysis/ME_CFS/DecodeME/h_MAGMA-HBA-DecodeME/

I got a similar result and an even stronger signal when using the DecodeME + MVP meta-analysis. It looked like this:

Source: https://s4me.info/threads/biologica...026-maccallini-et-al.50225/page-6#post-699621

We can also compare ME/CFS results to other diseases analyzed by Duncan et al using the same approach. For example the plot below highlighted the results for eMSN in different diseases.

Thread: https://s4me.info/threads/eccentric-medium-spiny-neuron-emsn.50276/post-695111

 

[ME/CFS Science Blog]

DESCARTES
This is also a major dataset with a lot of strong points: multiple tissues, in humans, 4 million cells. The main limitation is that it is from foetuses so may not directly apply to the cell types in adults.

The main paper is Cao et al. 2020:

https://www.science.org/doi/10.1126/science.aba7721

And the raw data is available here:

Cellxgene Data Portal

Find, download, and visually explore curated and standardized single cell datasets.

cellxgene.cziscience.com

This dataset was applied to DecodeME in a recent preprint by Jun Hyun Lee. It analysed level 2 which has 77 different cell types. The strongest signals were for inhibitory neurons. Here's a link to the full results, posted on the forum:

Preprint Post in thread 'Global and local genetic overlap among ME/CFS, irritable bowel syndrome and psychiatric traits: a hypothesis-generating analysis, 2026, Lee'

Jun 11, 2026

From Supplementary Table 4, data for the full DecodeME-chort.

I think table 4 is showing the MAGMA results from the less precise sensitivity analysis, where instead of using continuous expression values, the author created a binary gene set in which the top 10% highest value genes were labeled as 1 and all others labeled as 0.

Supplementary Table 7 includes the main MAGMA results. These are all the FDR significant ME/CFS-all cases cell types:

...

• forestglip

• 

 

[ME/CFS Science Blog]

A note on the FUMA platform datasets
The FUMA platform contains many of these dataset and makes it easy to test for them by uploading your data. Their approach is explained in this preprint:

Preprint Post in thread 'Identification of cell types associated with 14 brain phenotypes from more than 10 million single cells, 2025, Phung et al.'

Saturday at 2:04 PM

This is a recent preprint from the researchers who maintain the FUMA website on the cell type analysis they do, matching GWAS data to RNA databases. They created 388 datasets from 36 studies and tested these in 14 brain traits and diseases.

Here's one of their graphs. I noticed that oligodendrocyte precursor were associated with multiple brain disorders such as depression and schizophrenia. These oligodendrocyte precursor also came up as a significant finding In the analysis I did with the decodeME + MVP data (the one where I used a 30,15 window for MAGMA). I couldn't make sense out of...

• ME/CFS Science Blog

• 

We have previously found significant links to white matter cells from the Seeker dataset (in the Maccalini paper), striatal cells from Dropviz (also Maccalini paper) and neurons from the Braun fetal dataset using FUMA. But FUMA divides these datasets in smaller bits, often by dissection. I think it would be worthwhile to try to re-analyze them by standardizing gene expression for the entire dataset, similar to what Duncan et al. did with the Human Brain Atlas.

 

[paolo]

Than you for this overview!

I think one of the reasons behind the FUMA approach to cell-type analysis is to allow the combination of datasets from different sources.

 

[ME/CFS Science Blog]

A recent and useful lecture on the limitations of cell-type enrichment analysis with GWAS data. It's from Rachel Brouwer, a member of the Posthuma group that produced FUMA, MAGMA, FLAMES, etc.


The conclusion is a bit sobering. These analyses work well for broad cell types (like neurons, glia, B-cells, etc.) but probably not yet for more specific cell types.

There are multiple choices to make in the statistical analysis, and many of these choices have a substantial influence on the results. In addition, there is often a strong correlation among cell types, which makes it difficult to differentiate them. Third, many of these RNA datasets are based on a very limited number of donors (medium of 5).

She says (minute 13:10): "....for now, my conclusion would be: broad cell types are fine, but if you ask me to give you a fine-grain cell type, I don't have too much confidence in using only this as a reason to do lab work, which is very unfortunate. I've been hesitating a long time to say this out loud but here we are. I don't think this is this should be the only reason to pick a cell type."