St. Jude Cloud and the Pediatric Cancer Genome Project — raw paediatric leukaemia genomes a small lab can get without a collaboration
also called St. Jude Cloud · PCGP · Pediatric Cancer Genome Project · St. Jude Genomics Platform · PeCan · Pediatric Cancer Knowledgebase · stjude.cloud
St. Jude runs a cloud platform holding raw genome and RNA sequencing data from more than ten thousand children with cancer, and any researcher can browse it for free and request the underlying data through a signed agreement. It is the one resource on this list a lab can start using this week without knowing anybody.
The same thing, for a scientist
The St. Jude Children's Research Hospital – Washington University Pediatric Cancer Genome Project (PCGP), launched in 2010, was the first large-scale effort to sequence paediatric cancer genomes systematically [E1]. Its output, plus St. Jude's ongoing prospective clinical genomics programmes, is distributed through St. Jude Cloud: a cloud-based data-sharing ecosystem holding **1.25 petabytes** of harmonised genomic data from more than 10,000 paediatric cancer patients and survivors — **12,104 whole genomes, 7,697 whole exomes and 2,202 transcriptomes** — across three interconnected applications (Genomics Platform, Pediatric Cancer Knowledgebase, Visualization Community) [E2]. Browsing data and workflows is free and requires no account; access to raw sequence requires a data access agreement, after which storage is not charged [E3]. The PCGP/TARGET joint pan-cancer analysis of 1,699 paediatric leukaemias and solid tumours established that paediatric cancer drivers overlap adult ones by only 45% and that structural variants and copy-number changes constitute 62% of events [E4].
What is actually there, in numbers1 deeper
The 2021 Cancer Discovery description is the authoritative published account and its figures are worth stating exactly rather than approximately 2:
| Quantity | Value |
|---|---|
| Paediatric cancer patients and long-term survivors | more than 10,000 |
| Paediatric sickle cell patients | more than 800 |
| Harmonised genomic data volume | 1.25 petabytes, freely available |
| Whole genomes | 12,104 |
| Whole exomes | 7,697 |
| Transcriptomes | 2,202 |
| Paediatric cancer subtypes classified by expression profiling (demonstration use case) | 135 |
| Subtypes with mapped mutational signatures (demonstration use case) | 35 |
Three things distinguish this from a typical repository.
It is harmonised. Data are processed under a common framework, which is the difference between a dataset and a pile of files. The pan-cancer analysis of 1,699 paediatric tumours explicitly ran WGS, WES and transcriptome data 'under a uniform analytical framework' 4; a small lab that downloads heterogeneous data from several sources spends its first six months on alignment and QC harmonisation it should not have to do.
It is growing. The resource expands with regular uploads from St. Jude's prospective clinical genomics programmes 2, which means it is not a frozen legacy deposit in the way TARGET is.
It is paediatric by construction. This matters more than it sounds. Paediatric cancers arise in developing mesodermal tissue, and the pan-cancer work found only 45% of paediatric driver genes overlap adult pan-cancer studies, with structural variants and copy-number alterations accounting for 62% of events 4. Adult-derived resources systematically mis-serve a KMT2A-fusion question, because fusions are precisely the class adult catalogues under-represent.
The access model, stated precisely
Free, immediately, with no agreement and no account: browsing the data and the analysis workflows. St. Jude Cloud's own documentation states that 'Browsing our data and workflows is free and does not require an account' 3. This includes the Pediatric Cancer Knowledgebase and the Visualization Community layers described in the published account 2. For a laboratory generating hypotheses about KMT2A-fusion expression programmes, this is a same-day resource.
Raw sequence requires a Data Access Agreement. St. Jude operates a formal data request process with a DAA, described as protecting the data even though it is de-identified 3. Available types are whole genome, whole exome and RNA-seq.
Cost: once a request is approved, researchers 'will not be charged for any associated storage fees' 3. Compute is a separate matter and the documentation reviewed here did not state compute pricing; the platform runs on cloud infrastructure (DNAnexus and Microsoft are named collaborators on the published platform 2), so cloud compute charges should be assumed to exist until confirmed otherwise.
Not verified here: eligibility criteria for the DAA (institutional signatory requirements, whether non-US or non-academic applicants qualify), review turnaround time, and compute pricing. These were sought but not found in the documentation pages read on 2026-09-18 and should be confirmed before being planned around.
What a small lab can do with this today, without a collaboration
This is the point of the node, and it is worth being concrete because it is the only entry in this landscape that requires no negotiation with anybody.
Immediately, free, no agreement: interrogate the Pediatric Cancer Knowledgebase and the visualisation layers for KMT2A-fusion cases — which fusion partners appear, at what ages, with what co-occurring lesions — and compare expression-defined subtypes across the 135 the platform's own demonstration use case classified 2. Generate the hypothesis before requesting the data to test it.
After a DAA, at the cost of a signature and a wait: raw WGS, WES and RNA-seq on paediatric leukaemia, harmonised, including the PCGP cases 23. This supports breakpoint-level analysis of KMT2A fusions, allele-specific expression of the fusion transcript, and definition of the HOX/MEIS programme in primary infant blasts rather than in cell lines. A laboratory that has only ever worked in SEM, MV4-11 and RS4;11 can move to primary human data without acquiring a single patient.
What it cannot give. No treated-versus-untreated menin-inhibitor comparison — this is diagnostic and clinical-genomics material, not trial correlative material. No prospective specimens, no cells, no ability to do an experiment. And the paediatric leukaemia cases are dominated by St. Jude's own case mix, so infant KMT2A-rearranged ALL is a small subset of a large resource rather than a purpose-built cohort.
Strategic reading. St. Jude Cloud is the resource that makes a small laboratory credible before it asks COG or a company for anything. The strongest version of a specimen request to COG is one where the applicant has already done the analysis on public data and is asking for the specimens that test the resulting prediction. This node is how that precondition gets met.
Is a collaboration realistically possible — and is one even needed?
Mostly not needed, which is the unusual and valuable thing here. The entire value proposition of St. Jude Cloud is that it substitutes a signed agreement for a relationship. A laboratory with no connections in paediatric oncology can obtain harmonised paediatric leukaemia genomes by filling in a form.
Where a genuine collaboration would add something: St. Jude is also an active clinical site with its own menin-inhibitor trial — the phase 1 of revumenib with azacitidine and venetoclax in patients aged 1 to 30 (NCT06177067), sponsored by St. Jude with Syndax as collaborator. That is treated-patient material, which the cloud platform does not contain. The institution is therefore both an open-data provider and a closed-specimen holder, and the two require entirely different approaches: a form for the first, a colleague for the second.
The honest limitation on ambition. St. Jude is one of the largest and best-resourced paediatric cancer research institutions in the world, with its own computational biology department that produced the platform. A small laboratory does not partner with St. Jude as an equal on genomics; it uses St. Jude's data and distinguishes itself on the biology or the assay that St. Jude is not running.
| group | St. Jude Children's Research Hospital — Department of Computational Biology (St. Jude Cloud); St. Jude / Washington University Pediatric Cancer Genome Project (PCGP) |
|---|---|
| institution | St. Jude Children's Research Hospital, Memphis, Tennessee, USA; PCGP jointly with Washington University School of Medicine, St. Louis. Platform built with DNAnexus and Microsoft Research. |
| focus | Open sharing of harmonised paediatric cancer and sickle cell genomic data (WGS, WES, RNA-seq), cloud analysis workflows, a paediatric cancer knowledgebase and multi-omic visualisation |
| access model | Two tiers. (1) Free, no account, no agreement: browse data and workflows, use the knowledgebase and visualisation apps. (2) Raw sequence (WGS/WES/RNA-seq): formal data request plus a signed Data Access Agreement; storage not charged after approval. NOT VERIFIED: DAA eligibility criteria, approval turnaround, and compute pricing — assume cloud compute costs exist until confirmed. |
| what a small lab can do today | Without any agreement: hypothesis generation on KMT2A-fusion cases via PeCan and the visualisation layers, including comparison across expression-defined paediatric subtypes. After a DAA: breakpoint-level analysis of KMT2A fusions, fusion-transcript allele-specific expression, and definition of the HOX/MEIS programme in primary paediatric blasts rather than in SEM/MV4-11/RS4;11 cell lines. What it cannot give: any on-treatment or menin-inhibitor-exposed material, any prospective specimen, any viable cells. |
| partnering angle | Uniquely, mostly no partnership required — a signed Data Access Agreement substitutes for a relationship, which is what makes this the correct first move for a laboratory with no existing connections. Use the free tier to generate a specific, testable prediction about KMT2A-fusion biology, use the DAA tier to test it on primary paediatric genomes, and then approach COG or a company with a result rather than a request. A genuine human collaboration with St. Jude is warranted only for its trial-treated material (NCT06177067), which the cloud platform does not contain and which requires a colleague rather than a form. |
what it controls
[ "1.25 petabytes of harmonised paediatric genomic data: 12,104 whole genomes, 7,697 whole exomes, 2,202 transcriptomes from >10,000 paediatric cancer patients and survivors", "The Pediatric Cancer Knowledgebase (PeCan) and Visualization Community — free to browse with no account", "Cloud analysis workflows co-located with the data", "Separately, as a clinical institution: treated-patient material from its own menin-inhibitor trial NCT06177067 (revumenib + azacitidine + venetoclax, ages 1-30), which is NOT in the cloud platform" ]
key papers — the numbers behind the chart
[
{
"title": "St. Jude Cloud: A Pediatric Cancer Genomic Data-Sharing Ecosystem",
"year": 2021,
"venue": "Cancer Discovery",
"pmid": "33408242",
"doi": "10.1158/2159-8290.CD-20-1230"
},
{
"title": "The Pediatric Cancer Genome Project",
"year": 2012,
"venue": "Nature Genetics",
"pmid": "22641210",
"doi": "10.1038/ng.2287"
},
{
"title": "Pan-cancer genome and transcriptome analyses of 1,699 paediatric leukaemias and solid tumours",
"year": 2018,
"venue": "Nature",
"pmid": "29489755",
"doi": "10.1038/nature25795"
}
]- Open access to data removes the access problem and thereby removes the competitive moat: every other group can run the same analysis, so a result derived only from St. Jude Cloud is by construction one anybody could have got first.
- The data are diagnostic and clinical-genomics material, not trial correlative material. No amount of it will answer why a specific infant relapsed on a menin inhibitor.
- Harmonisation is done to St. Jude's pipeline choices. Anyone whose question depends on a different aligner, caller or fusion-detection method inherits decisions they did not make and may not be able to undo at petabyte scale.
- A cohort assembled from one institution's case mix plus research consent is not a population sample; rare subtypes will be over- or under-represented in ways that are invisible in the headline totals.
- What are the DAA eligibility criteria and the actual approval turnaround? Not stated in the documentation read here, and this determines whether the resource is usable in weeks or quarters.
- What does cloud compute cost on the Genomics Platform? Storage is explicitly free after approval; compute pricing was not found, and for petabyte-scale reanalysis this is the decisive number.
- How many infant (under 12 months) KMT2A-rearranged ALL cases with RNA-seq are actually in the resource? The headline counts are cohort-wide and the infant subset could be small enough to be underpowered for the questions this project cares about.
- Is any treated-on-trial material from St. Jude's own revumenib study (NCT06177067) destined for the cloud platform, or does it remain institution-controlled indefinitely?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The Pediatric Cancer Genome Project is a joint St. Jude Children's Research Hospital – Washington University initiative to identify somatic mutations driving paediatric cancer through genome sequencing, established as a deliberate community resource.
The Pediatric Cancer Genome Project · Downing JR, Wilson RK, Zhang J, Mardis ER, Pui CH, Ding L, Ley TJ, Evans WE · Nature Genetics 2012What this evidence cannot show
A 2012 perspective written while the project was still generating primary sequence data; its descriptions of scope and early results are now historical. It establishes the origin and intent of PCGP, not its current holdings. According to PubMed; DOI https://doi.org/10.1038/ng.2287
Evidence that argues against this (1)
- The PCGP's design decisions are a description of how the project was set up. Its infant KMT2A-rearranged content is the operative fact for this graph, and that is 22 whole genomes in the discovery cohort — a number this project has had to caveat repeatedly elsewhere. Institutional ambition and usable denominator are different things. Cancer Discovery 2021
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
St. Jude Cloud is a cloud-based data-sharing ecosystem holding 1.25 petabytes of harmonised genomic data — 12,104 whole genomes, 7,697 whole exomes and 2,202 transcriptomes — from more than 10,000 paediatric cancer patients and long-term survivors plus more than 800 paediatric sickle cell patients, freely available, expanding with regular uploads from St. Jude's prospective clinical genomics programmes, and organised into three interconnected applications: Genomics Platform, Pediatric Cancer Knowledgebase and Visualization Community.
St. Jude Cloud: A Pediatric Cancer Genomic Data-Sharing Ecosystem · McLeod C, Gout AM, Zhou X, Thrasher A, Rahbarinia D, Brady SW, Macias M, Birch K, Finkelstein D, et al. · Cancer Discovery 2021What this evidence cannot show
Published 2021; the quoted counts are a snapshot and the resource is explicitly described as expanding, so current holdings are larger and the exact composition differs. Written by the platform's builders, so its framing of usability and value is not disinterested, although the data counts are concrete and checkable. 'Freely available' refers to the absence of a fee, not to the absence of a data access agreement. The paediatric leukaemia fraction, and the infant KMT2A-rearranged fraction within it, are not broken out. According to PubMed; DOI https://doi.org/10.1158/2159-8290.CD-20-1230
Evidence that argues against this (1)
- Answers this entry's third open question with a number, and the number is the argument against the resource as well as for it. The entry asks 'How many infant (under 12 months) KMT2A-rearranged ALL cases with RNA-seq are actually in the resource?' and notes that the headline counts are cohort-wide. The flagship PCGP infant cohort is Andersson et al.: whole-genome, exome, RNA and targeted DNA sequencing of 65 infants (47 MLL-rearranged, 18 non-MLL-rearranged) plus 20 older children with MLL-rearranged leukaemia. So the order of magnitude for infant KMT2A-rearranged cases with multi-platform sequencing in this resource is tens, not thousands - roughly 0.4% of the 12,104 whole genomes the platform advertises. The same paper supplies the reason this matters more than usual: the predominant leukaemic clone carries a mean of 1.3 non-silent mutations, among the lowest somatic mutation frequencies of any sequenced cancer, and the kinase-PI3K-RAS pathway mutations found in 47% of cases were often subclonal and frequently lost at relapse. At n=47 with 1.3 mutations per clone, any recurrence-based discovery analysis in this cohort is underpowered, and a null result would be uninterpretable. The honest revision to this entry's framing: St. Jude Cloud is genuinely open and genuinely large, but for the specific population this project cares about it is a small cohort behind a low barrier rather than a large cohort behind a high one, and the binding constraint is statistical power, not the data-access agreement. Caveats: the resource has grown since 2015 through prospective clinical genomics uploads, so the current infant count may exceed 47 and was not enumerated here; the abstract does not separate the whole-genome subset from exome or RNA; and the cohort's age is a limitation for anything depending on modern read lengths or pipelines. Read: abstract only. Andersson AK et al. 2015
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
St. Jude Cloud's access model has two tiers: browsing data and workflows is free and requires no account, while access to the underlying whole genome, whole exome and RNA-seq data requires a formal data request and a signed Data Access Agreement, after which approved researchers are not charged storage fees.
St. Jude Cloud Genomics Platform — Getting Started / Overview documentation · St. Jude Children's Research Hospital · docs.stjude.cloud 2026What this evidence cannot show
Documentation authored by the data provider. The pages read on 2026-09-18 did NOT state DAA eligibility criteria (institutional signatory requirements, whether non-academic or non-US applicants qualify), approval turnaround time, or compute pricing — only that storage is not charged after approval. Compute charges should be assumed to exist, since the platform runs on commercial cloud infrastructure, until confirmed otherwise. Confirm all three before planning around them.
Evidence that argues against this (1)
- A two-tier access model means the free tier does not contain the data. A signed data access agreement requires institutional signature authority and review, and the same paper reports a median seven-day approval alongside an 18.8% rejection rate — so roughly one application in five does not proceed, which is schedule risk rather than latency. Cancer Discovery 2021
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The joint PCGP/TARGET pan-cancer analysis of 1,699 paediatric leukaemias and solid tumours, processed under a uniform analytical framework across whole-genome, whole-exome and transcriptome data, identified 142 driver genes of which only 45% match adult pan-cancer studies, with copy-number alterations and structural variants constituting 62% of events.
Pan-cancer genome and transcriptome analyses of 1,699 paediatric leukaemias and solid tumours · Ma X, Liu Y, Liu Y, Alexandrov LB, Edmonson MN, Gawad C, Zhou X, Li Y, Rusch MC, Easton J, et al. · Nature 2018What this evidence cannot show
Six histotypes only, so not a complete census of paediatric cancer; the cohort is a union of PCGP and TARGET cases and is therefore shaped by what those two programmes chose to sequence, which biases toward high-risk and research-consented disease. Driver-gene calls depend on the statistical framework used and are not a fixed list. 2018 data and methods. According to PubMed; DOI https://doi.org/10.1038/nature25795
Evidence that argues against this (1)
- Qualifies the independence of the two cohorts this item combines, and it is a methodological point rather than a challenge to the finding. The pan-cancer analysis draws on PCGP and TARGET together. TARGET's paediatric leukaemia arms were assembled in partnership with the Children's Oncology Group from COG trial cohorts, and its B-ALL discovery material came from COG frontline high-risk protocols whose registered minimum age is 1 year (COG-P9906, NCT00005603, n=276; AALL0232, NCT00075725, n=3,154). Two consequences. First, PCGP-plus-TARGET is not a discovery-plus-independent-replication design in the way a reader might assume, because the TARGET component and the COG component are the same patients; anyone designing a two-cohort analysis in this graph should not count them twice. Second, and specific to this project: the combined 1,699-sample pan-cancer resource contains very little infant leukaemia, because the TARGET half structurally excludes infants and the PCGP infant series is only 65 patients. The 45%-overlap-with-adult-drivers headline is therefore a statement about paediatric cancer generally and should not be read as a statement about infant KMT2A-rearranged disease. Caveats: the exact per-study composition of TARGET-ALL-P2 was not enumerated here, so the infant count is inferred from contributing-protocol eligibility rather than counted; and the PCGP and TARGET overlap concerns the leukaemia arms, not the solid-tumour arms. Read: registry records and abstracts; no full text. Children's Oncology Group; Pediatric Oncology Group 2026
the sequencing resource the co-mutation picture rests on
Counter-evidence (Andersson 2015, PMID 25730765): for the genomic questions the binding constraint is cohort size and a 1.3-mutation-per-clone burden, not permission -- 47 infant MLL-r whole genomes will not support a recurrence analysis however freely shared
Counter-evidence: TARGET was run in partnership with COG and drew its cases from COG trials, so PCGP+TARGET is not discovery-plus-independent-replication; and the infant cohort lives in PCGP (47 MLL-r genomes), not TARGET
Counter-evidence (Andersson 2015, PMID 25730765): the infant KMT2A-r subset is ~47 sequenced cases at a mean 1.3 non-silent mutations per clone -- the binding constraint on this resource is statistical power, not the data-access agreement
open-access paediatric genomic resource requiring no collaboration; start here
PeCan and ProteinPaint were UNREACHABLE by every public route tried