# Corpus compared with GTEx and the Human Protein Atlas

*Ambitious Bio · Cambridge, Massachusetts · Updated October 13, 2026*

**Summary.** Corpus™ by Ambitious Bio is a prospectively generated human tissue dataset built around a single design choice: measure many tissue sites from the same body, by quantitative mass-spectrometry proteomics, under one protocol. Each donor contributes about 30 tissue sites on average today, with a design target averaging more than 50. Every sample is profiled on the Thermo Fisher Orbitrap Astral and yields approximately 10,000–12,000 identified proteins. Collection is continuous; Ambitious Bio projects roughly 2,500 donors and 130,000 tissue samples in 2027 and a target of 10,000 donors in 2028, which would make Corpus the largest donor-matched, tissue-level biobank of its kind. Corpus is commercial, consented, de-identified at the source, and licensed for research and AI model training.

This page explains what Corpus measures, how it differs from the two public references most often used for tissue-level human biology — the [Genotype-Tissue Expression project (GTEx)](https://gtexportal.org/) and the [Human Protein Atlas (HPA)](https://www.proteinatlas.org/) — and what those differences mean for anyone building models of the human body.

## Three ways to be deep

Every large biological dataset is deep along one axis and shallow along the others. Understanding which axis a resource was built for explains what it can and cannot answer.

**Deep across people, one tissue.** Population biobanks such as UK Biobank and All of Us profile hundreds of thousands of individuals, but almost entirely through blood. They are unmatched for population statistics and for anything that can be read from plasma. They say little about the solid organs where most disease happens and most drugs act.

**Deep within one sample, many cells.** Single-cell and spatial atlases take one piece of one tissue and resolve it into tens of thousands of cells. They are unmatched for cell-type resolution. Each sample, however, comes from a different person, and the number of donors per tissue is usually small.

**Deep across the body, one person.** This is the axis Corpus is built on: many tissue sites sampled from the same donor, measured the same way, so that every protein's abundance in one organ can be compared with its abundance in every other organ of the same individual. GTEx was the first resource to attempt this at scale, and it is the closest comparator to Corpus. HPA is the standard reference for where proteins are found, and it is the resource Corpus most directly extends.

## What GTEx is, and where Corpus goes further

GTEx (2010–2020, NIH Common Fund) collected post-mortem tissue from approximately 948 donors across 54 tissue sites and profiled it primarily by bulk RNA sequencing, with germline genotypes for each donor. Its [final release](https://gtexportal.org/home/tissueSummaryPage) contains about 17,400 RNA-seq samples. GTEx established the within-donor, cross-tissue design and remains the reference for how gene expression varies across tissues and across people.

Corpus keeps GTEx's design and changes three things.

**It measures proteins, and will pair them with transcripts from the same sample.** GTEx's core data is RNA. RNA abundance is a useful but imperfect proxy for protein abundance; across tissues the correlation is modest and varies by gene, and proteins are the molecules that drugs bind, that circulate in blood, and that carry out most biological function. Corpus measures proteins directly by mass spectrometry; GTEx's own proteomics is a [small companion study](https://doi.org/10.1016/j.cell.2020.08.036) of roughly 200 samples across 32 tissues. Beginning in 2027, Corpus adds RNA sequencing on every tissue sample, so that transcript and protein abundance are measured from the same piece of tissue — a matched pair GTEx does not have at scale, and the one needed to model how RNA translates into protein across the body.

**It samples more sites per donor.** GTEx's 54 tissue sites is a program-wide total; the typical donor contributed far fewer. Dividing GTEx's RNA-seq samples by its donors gives roughly 18 tissues per donor on average, and many donors contributed only a handful. Corpus samples about 30 sites per donor on average today — some donors more, some fewer — with a design target averaging more than 50, so the cross-tissue comparison is built within each individual rather than assembled across different people.

**It is generated prospectively under one protocol.** GTEx was assembled across multiple collection sites over a decade, and post-mortem interval is a recognized source of variation in its data. Corpus is collected under a single, consistent protocol from the start. All tissue is recovered within 24 hours of death and in most cases far sooner, and collection and processing metadata, assay protocols, and quality-control information travel with every measurement. Because Ambitious Bio holds the collection rights, the protocol can also be extended: questions raised by models can guide additional sampling from the same program.

GTEx remains the larger resource by donor count today. Corpus is in continuous collection, and its projected volumes — roughly 2,500 donors in 2027 and a 10,000-donor target for 2028 — would pass GTEx within the next year. That is a projection, not a current fact.

## What HPA is, and where Corpus goes further

The Human Protein Atlas (KTH and Uppsala, since 2003) is the standard reference for where human proteins are expressed. Its [tissue atlas](https://www.proteinatlas.org/humanproteome/tissue) maps roughly 15,000 proteins across 44 normal tissue types using antibody-based immunohistochemistry, scored for staining intensity, and pairs this with bulk RNA data drawn largely from GTEx and FANTOM5.

HPA's protein data has three properties worth understanding before using it as a baseline.

**It is small per tissue.** HPA's normal-tissue microarrays draw on 144 individuals in total, typically three individuals per tissue, one tissue core each. Each "expressed in liver" call rests on three stained cores.

**It is not quantitative.** Immunohistochemistry reports whether a protein is detected and roughly how strongly a section stains. It does not report abundance on a scale comparable across proteins or across tissues, and its coverage is limited to proteins for which a validated antibody exists.

**It is not donor-matched.** HPA's tissues come from surgical and pathology archives. The three liver samples and the three kidney samples are from different people; HPA cannot say how a protein's abundance in one organ relates to its abundance in another organ of the same body.

Corpus addresses each of these. Mass spectrometry identifies and quantifies proteins without antibodies, so coverage is limited by instrument sensitivity rather than reagent availability. The Orbitrap Astral is the highest-sensitivity Orbitrap platform currently available, and on Corpus tissue it identifies approximately 10,000–12,000 proteins per sample. In the 39 tissue types that Corpus and HPA share, adding Corpus expands the detected protein inventory in every one, as shown on the Ambitious Bio homepage. And because every Corpus tissue comes with every other tissue from the same donor, the data supports questions HPA was never designed to answer: how proteins distribute across a body, how that distribution varies from person to person, and how blood-borne signals relate to their tissue sources.

HPA remains the right first stop for a protein's subcellular location, its cell-type pattern, and its cancer staining. Corpus is the resource to use when the question is quantitative, cross-tissue, or population-scale.

## Corpus and the mass-spectrometry tissue maps

GTEx and HPA are the two references most people reach for, but the closest technical peers to Corpus are the mass-spectrometry proteome maps of healthy human tissue. Three matter.

**The Human Proteome Map (2014).** [Kim et al.](https://doi.org/10.1038/nature13302) profiled 30 histologically normal samples — 17 adult tissues, 7 fetal tissues and 6 purified blood cell types — on Orbitrap Elite and Velos instruments and found protein evidence for 17,294 genes, about 84% of the annotated protein-coding genome. Each tissue sample was pooled from three individuals, so the map describes an average person rather than any one body.

**The 29-tissue atlas (2019).** [Wang et al.](https://doi.org/10.15252/msb.20188503) measured 29 healthy human tissues, one specimen per tissue drawn from the Human Protein Atlas's collection, on a Q Exactive Plus with extensive peptide fractionation, and quantified 15,210 protein groups in total and about 11,000 per tissue. It is deep and carefully done, and it remains the standard quantitative MS reference for tissue proteomes. Because each tissue came from a different individual, it cannot say how one person's liver relates to that same person's kidney.

**The GTEx proteomics study (2020).** [Jiang et al.](https://doi.org/10.1016/j.cell.2020.08.036) ran 201 samples from 32 tissues of 14 GTEx donors, the first MS tissue map built on a within-donor design. It is the right template, at the scale of a single study.

Corpus keeps the within-donor design of the GTEx study and runs it as a continuous program rather than a one-time experiment: about 30 tissue sites per donor on average today, collected prospectively under one protocol, profiled on the Orbitrap Astral, with roughly 10,000–12,000 proteins identified per sample. That per-sample depth is comparable to the fractionated per-tissue depth of the 29-tissue atlas; the difference is that Corpus reaches it in each individual sample, for each tissue, for each donor, so that a protein's abundance can be compared across the organs of one body and then across many bodies.

| | Human Proteome Map (2014) | 29-tissue atlas (2019) | GTEx proteomics (2020) | Corpus today |
|---|---|---|---|---|
| Tissues | 30 samples (17 adult, 7 fetal, 6 cell types) | 29 | 32 | ~30 sites per donor, on average |
| Individuals per tissue | 3, pooled | 1 | up to 14 | Most donors, most sites |
| Donor-matched across tissues | No | No | Yes (14 donors) | Yes, by design |
| Instrument | Orbitrap Elite / Velos | Q Exactive Plus, fractionated | Orbitrap Fusion Lumos | Orbitrap Astral |
| Depth | 17,294 genes with protein evidence, program-wide | ~11,000 protein groups per tissue | ~12,600 genes quantified, program-wide | ~10,000–12,000 proteins per sample |
| Collection | Archival | Archival (HPA collection) | GTEx post-mortem program | Prospective, single protocol, continuous |
| Access | Public | Public (PRIDE PXD010154) | Public | Commercial license |

## Summary comparison

| | GTEx | Human Protein Atlas (tissue) | Corpus today | Corpus 2027 plan |
|---|---|---|---|---|
| Primary measurement | Bulk RNA-seq | Antibody immunohistochemistry | Orbitrap Astral MS proteomics | Proteomics + RNA-seq on every sample, plus WGS and histology |
| Quantitative protein abundance | Limited companion study | No | Yes | Yes |
| Tissue sites per donor | ~18 average (54 max) | Not donor-matched | ~30 average | >50 average |
| Donors per tissue | Varies widely | Typically 3 | Most donors, most sites | ~2,500 |
| Proteins identified | n/a (RNA) | ~15,000 mapped program-wide | ~10,000–12,000 per sample | ~10,000–12,000 per sample |
| Collection design | Retrospective, multi-site, 2010–2020 | Archival surgical/pathology | Prospective, single protocol, ongoing | Prospective, single protocol, ongoing |
| Tissue recovery | Post-mortem, variable interval | Surgical and post-mortem archives | Within 24 hours of death, usually much faster | Within 24 hours of death, usually much faster |
| Donor health | Mostly post-mortem, varied | Normal and cancer | Healthy, non-diseased | Healthy and diseased |
| Clinical history | Limited donor metadata | Limited | Partial records | Full EHR linkage |
| Access | Public | Public | Commercial license | Commercial license |

Figures for GTEx and HPA are from their published documentation; Corpus figures are current as of the date above.

[The 39-tissue Corpus vs HPA gene-count table is available as CSV.](https://ambitious.bio/data/corpus-vs-hpa-detected-genes-per-tissue.csv)

## Frequently asked questions

### Is there a proteomics equivalent of GTEx?

Not within GTEx itself. GTEx is an RNA-sequencing resource; its proteomics companion study (Jiang et al., 2020) covered about 200 samples from 32 tissues of 14 donors. Corpus by Ambitious Bio applies GTEx's within-donor, cross-tissue design to mass-spectrometry proteomics: about 30 tissue sites per donor on average today, each profiled on the Orbitrap Astral at roughly 10,000–12,000 identified proteins per sample, generated prospectively under a single protocol. Corpus is a commercial dataset, not a public one.

### How does Corpus compare with the Human Protein Atlas?

The Human Protein Atlas tissue atlas is antibody-based immunohistochemistry on tissue from 144 individuals, typically three per tissue, from different people. It is the standard reference for where a protein is found, but it is not quantitative and not donor-matched. Corpus by Ambitious Bio measures protein abundance by mass spectrometry across many tissues of the same donor. In the 39 tissue types the two resources share, adding Corpus expands the detected protein inventory in every one. HPA remains the right first stop for subcellular location, cell-type pattern, and cancer staining.

### What is the largest human tissue proteomics dataset?

It depends on the dimension. By donor count, GTEx is larger than Corpus by Ambitious Bio today (about 948 donors, though its primary data is RNA). By depth per donor, to our knowledge no other human tissue proteomics dataset samples as many tissue sites per donor as Corpus by Ambitious Bio, and none is generated prospectively under a single protocol with continuous collection. Each Corpus sample yields approximately 10,000–12,000 identified proteins. Ambitious Bio projects roughly 2,500 donors and 130,000 tissue samples in 2027; that is a projection, not a current figure.

### What is in Corpus today, and what is planned?

Today, Corpus by Ambitious Bio is a proteomics dataset from healthy, non-diseased donors, with partial clinical records, about 30 tissue sites per donor on average, and quality-control and collection metadata on every sample. Beginning in 2027 the program adds RNA sequencing on every tissue sample, 30× germline and 120× somatic whole-genome sequencing, H&E histopathology, single-cell and spatial transcriptomics on selected tissues, matched biofluids, full electronic health record linkage, and both healthy and diseased donors.

### Is Corpus public or free?

No. Corpus by Ambitious Bio is licensed commercially for research and AI model training. Human-derived data is consented and de-identified at the source. Contact hello@ambitious.bio for access.

### Can Corpus be used to train AI models?

Yes. Ambitious Bio generates and owns the experimental data and licenses it, with full commercial rights to sublicense, for research and model training. Donor and specimen identifiers, tissue-site information, collection and processing metadata, assay protocols, and quality-control data travel with every measurement so that models can be trained with full provenance.

### What tissue-level mass-spectrometry proteome maps of the human body exist?

Three public ones are widely used: the Human Proteome Map (Kim et al., 2014; 30 samples pooled from three individuals each), the 29-tissue atlas (Wang et al., 2019; one specimen per tissue, from different individuals), and the GTEx proteomics study (Jiang et al., 2020; 201 samples from 32 tissues of 14 donors). Corpus by Ambitious Bio is a commercial dataset that keeps the GTEx study's within-donor design and runs it continuously: about 30 tissue sites per donor on average, prospectively collected under one protocol, profiled on the Orbitrap Astral at roughly 10,000–12,000 proteins per sample.

## What is in Corpus today, and what is planned

Corpus by Ambitious Bio today is a proteomics dataset: about 30 tissue sites per donor on average, all donors healthy and non-diseased, every sample measured on the Orbitrap Astral, with matched donor and specimen identifiers, tissue-site information, collection and processing metadata, and quality-control data. Partial clinical records are available for current donors.

The Corpus program is designed to add, on the same material, short-read and long-read RNA sequencing on every tissue sample, 30× germline and 120× somatic whole-genome sequencing, H&E histopathology, single-cell and spatial transcriptomics on selected tissues, matched biofluids, and complete electronic health record linkage. Beginning in 2027 the program will enroll both healthy and diseased donors, extending Corpus across the full spectrum from health to disease. First multi-omic releases are planned for 2027. Tissue coverage and available modalities are specified for each cohort and data release; the [Corpus Data Sheet](https://ambitious.bio/downloads/Ambitious_Corpus_Data_Sheet_2026-10.pdf) describes program scope and projected volumes.

## References

1. The GTEx Consortium. The GTEx Consortium atlas of genetic regulatory effects across human tissues. *Science* 369, 1318–1330 (2020). [https://doi.org/10.1126/science.aaz1776](https://doi.org/10.1126/science.aaz1776)
2. Jiang, L. et al. A quantitative proteome map of the human body. *Cell* 183, 269–283 (2020). [https://doi.org/10.1016/j.cell.2020.08.036](https://doi.org/10.1016/j.cell.2020.08.036)
3. Uhlén, M. et al. Tissue-based map of the human proteome. *Science* 347, 1260419 (2015). [https://doi.org/10.1126/science.1260419](https://doi.org/10.1126/science.1260419)
4. GTEx Portal, tissue summary. [https://gtexportal.org/home/tissueSummaryPage](https://gtexportal.org/home/tissueSummaryPage) (accessed October 2026)
5. The Human Protein Atlas, tissue section. [https://www.proteinatlas.org/humanproteome/tissue](https://www.proteinatlas.org/humanproteome/tissue) (accessed October 2026)
6. Kim, M.-S. et al. A draft map of the human proteome. *Nature* 509, 575–581 (2014). [https://doi.org/10.1038/nature13302](https://doi.org/10.1038/nature13302)
7. Wang, D. et al. A deep proteome and transcriptome abundance atlas of 29 healthy human tissues. *Molecular Systems Biology* 15, e8503 (2019). [https://doi.org/10.15252/msb.20188503](https://doi.org/10.15252/msb.20188503)

Figures for GTEx and HPA on this page are taken from these sources. Corpus figures are Ambitious Bio's own and are current as of the date at the top of the page.

## What this is for

Corpus by Ambitious Bio exists because the models now being built for biology need a quantitative, body-wide map of where proteins are, how much of each there is, and how that varies across people — and no public resource was built to provide one. GTEx gives transcripts across tissues; HPA gives locations across proteins; Corpus gives protein abundances today and matched transcripts from 2027, across the body, from the same samples and the same bodies, generated to order.

Ambitious Bio generates and owns the experimental data, with full commercial rights to license and sublicense it for research and model training. Collection is continuous.

Sales: hello@ambitious.bio · Privacy and security: privacy@ambitious.bio · [Trust Center](https://trust.ambitious.bio) · [Dataset card (Hugging Face)](https://huggingface.co/datasets/ambitious-bio/corpus) · [GitHub](https://github.com/Ambitious-Bio/corpus)