Skip to main content

Candidate studies to curate

A working backlog of real, public M. gigas datasets, initially screened on 2026-08-28 and expanded in dated updates below. Sample counts and group structures come from deposited metadata unless a paper is explicitly cited. For current registration and harmonization status, use the dashboard and study YAMLs; historical screening notes below are not a live status table.

This is a working backlog, not a commitment. Update it as studies are curated or rejected.

The selection criterion that actually matters

The obvious filter — “does the paper publish full statistics?” — is the wrong one to lead with.

HESSER2024_VCOR harmonizes cleanly, resolves 87.2% of its identifiers, and contributes nothing to any meta-analysis, because its supplementary table reports only log2FoldChange and padj. No standard error, no unadjusted p-value, so no inverse-variance weight. That is not an unlucky draw: a significance-filtered table with adjusted p-values only is the normal format for supplementary DE tables in this literature. Selecting studies by hoping their supplementary files are richer will mostly reproduce the same dead end.

The reliable route to poolable evidence is raw reads in SRA. When AREE reanalyzes raw data itself, its own DESeq2 run produces lfcSE and an unadjusted p-value by construction — the study becomes poolable regardless of what the authors chose to publish. The raw_reanalysis Nextflow path has now run at full depth on real data (see implementation_status.md).

So the screen applied here, in priority order:

  1. Raw reads deposited and public (SRA/ENA), not just a supplementary table.
  2. Replicated design, n ≥ 3 per group. See the rejected candidates below — this is not a formality, and it is checked twice: distinct sample_alias values before download, then replicate dispersion after quantification (scripts/check_replicate_dispersion.py). Distinct sample names do not prove distinct animals.
  3. Resilience-relevant contrast. Prefer a tolerance/resistance phenotype (resistant vs susceptible lineages) over pure exposure (treated vs control), per resilience_vs_exposure.md. Judge this on whether a phenotype was measured, not on how the title is worded — see the correction above.
  4. RNA-seq first, because the intake converter currently handles only differential-expression tables and the RNA-seq workflow is the most complete.
  5. Stressor and phenotype spread, so that pooled groups have more than one study in them. Two studies of the same stressor beat five of five different ones — a meta-analysis of k=1 is not a meta-analysis.

Tier 1 — curate these first

#1 is registered as CALLA2026_OSHV — three of six comparisons have full-depth reanalysis and harmonized evidence. See first_raw_reanalysis.md.

#3 failed QC as IOCAS2022_OA_ENERGY (registered 2026-09-30, reanalyzed 2026-10-01, QC decision 2026-10-07). Its replicates are technical, not biological — see Rejected, and why it matters below.

#7 is registered and harmonized as LUTIER2022_OA_TIPPING (PR #11, merged 2026-10-09), AREE’s first usable real ocean-acidification study. Two things on this page were wrong about it. First, the 31 “AMPLICON” runs are RNA-seq, mislabelled at deposit: all 76 runs make the paper’s 15 tanks × 5 oysters. Second, the design is one tank per pH, so tank, not oyster, is the unit of replication. The prespecified contrast (pHT ≤ 6.7 vs ≥ 7.6, 4 vs 3 tanks) is fitted on pseudo-bulk tanks (--replicate_unit tank). It is classified exposure_only: the physiology was measured per tank, and no oyster was scored for tolerance.

Correction from the first draft of this page. #1 was listed here as a study whose “contrast is a resilience phenotype, not an exposure”, on the strength of its BioProject title (“Evaluating Pacific oyster lineages for tolerance to Ostreid herpesvirus”). Curating it showed that is not what the deposited data contains: there is no survival, mortality, or viral-load measurement for these animals, and the publication frames its results as groundwork for future tolerance breeding. It is registered as disease_susceptibility, not disease_resistance. A resilience-sounding project title is not a resilience phenotype — check for a measured outcome before promising one.

# BioProject Runs Design (from run metadata) Resilience context Why first
1 ✅ PRJNA1329250 42 RNA-seq 2 populations × 4 viral-strain levels (Control/Australia/France/USA), n=5 challenged / n=6 control OsHV-1 challenge in two hatchery populations Well replicated across all eight groups, and the same stressor class as the study already registered, so the two can eventually pool. Calla et al. 2026 (10.1016/j.fsi.2026.111154).
2 PRJNA593309 43 RNA-seq OsHV-1 × temperature (21/26/29 °C) × timepoint, n=3 Disease resistance under thermal modulation Multi-stressor, well replicated, and published open access — Delisle et al. 2020, J Exp Biol (10.1242/jeb.226233). Pairs with #1 on pathogen challenge.
3 ❌ PRJNA826964 18 RNA-seq control vs OA × 3 timepoints (7/28/56 d), n=3 Ocean acidification, energy metabolism Clean 2×3 factorial, small enough to reanalyze quickly, and opens a second stressor class.

Doing #1 and #2 together is the point: they give the pathogen-challenge group k ≥ 2 with real standard errors, which is the first time random-effects pooling would run on anything but simulated data.

Tier 2 — good candidates, specific caveats

# BioProject Runs Design Context Caveat
4 PRJNA678408 10 WGBS 5 diploid + 5 triploid Desiccation + acute heat Would be AREE’s first real methylation study. But the intake converter does not handle region tables yet, and coordinate-based evidence is exposed to the assembly change described in handling_genome_versions.md.
5 PRJNA762441 18 RNA-seq diploid/triploid × 3 timepoints, n=3 Thermal stress, ploidy contrast Clean design; ploidy is a useful resilience covariate. No linked publication confirmed.
6 🟡 PRJNA913164 72 diploid/triploid, marine heatwave Thermal tolerance Tag-seq (3′ counts), not standard RNA-seq. Run it with --tag_seq (no length correction in Salmon or DESeq2; see workflows/rnaseq/README.md). Largest n on the list. Registered as GEORGE2023_HEATWAVE_PLOIDY (#18): it is published, George et al. 2023, Glob Chang Biol (10.1111/gcb.16880); the BioSamples carry no design, which comes from the authors’ NCBI submission sheet.
7 ✅ PRJNA735889 76 RNA-seq (31 mislabelled AMPLICON) 15 tanks, one per pH, × 5 oysters OA “tipping point” Registered as LUTIER2022_OA_TIPPING. Do not filter on ENA library_strategy: the AMPLICON runs are RNA-seq. One tank per pH, so fit on tanks, not oysters.
8 ❌ PRJNA1196326 6 RNA-seq 2 groups × n=3 Transgenerational OA Registered as IOCAS2024_OA_TRANSGEN (#17) with qc_status: failed: each arm is one larval pool sequenced three times. See Rejected below.
9 ❌ PRJNA877226 6 RNA-seq infected only, 12 h vs 48 h, n=3 Vibrio × high temperature Rejected: no control arm. See Rejected below.

Adjacent, but a different evidence type

  • PRJNA1190893 — low-salinity adaptation, 102 WGS runs. Population genomics, not differential expression. Registering it would mean a new harmonizer for variant/selection-scan evidence and a new feature_type. Worth doing eventually — salinity tolerance is a real breeding target and AREE currently has no real salinity evidence — but it is a feature, not a curation task.

  • Heat-resistant vs heat-susceptible families (CIBNOR breeding program, Baja California Sur), Arredondo-Espinoza et al. 2023, Comp Biochem Physiol Part D Genomics Proteomics 47:101089 (10.1016/j.cbd.2023.101089, PMID 37269757). Earlier drafts of this page cited it as “Escobedo-Fregoso et al.” — she is the last and corresponding author; the first author is Arredondo-Espinoza. Author list and journal name verified against PubMed on 2026-09-05.

    The framing is exactly what AREE wants — RR vs SS phenotypes from a breeding program under an oscillatory thermal challenge (26–34 °C, 30 days, sampled at day 0 and day 30). It would be AREE’s first real study with a measured resilience outcome, and its first thermal stressor; every real candidate in the current pool is disease_associated for want of exactly this. Per-group replication is not stated in the abstract and must be confirmed before registration.

    No public raw-data accession could be located as of 2026-09-05: BioProject searches on organism plus thermal-challenge terms and on author/CIBNOR both returned nothing, and the Europe PMC record reports no associated data or supplement with no open-access full text to read the availability statement from. Note that sr320 is a co-author, so lab storage and a direct ask to the corresponding author are better first moves than cold outreach. Tracked in issue #6.

    Update 2026-09-30: CIBNOR’s WGBS project PRJNA690951 covers the same RR/SS design (see M1 below), so the design has public raw data now. The RNA-seq reads for this paper are still not located.

DNA methylation and proteomics candidates

Screened 2026-09-30. Sources: ENA read-run search for every Bisulfite-Seq/MeDIP-Seq/MBD-Seq run under taxon 29159 (20 BioProjects); PRIDE Archive search for C. gigas / M. gigas (19 projects); PubMed title/abstract search (37 methylation papers, 66 proteomics papers). Group sizes come from run labels or BioSample attributes, not from paper text. A publication is “confirmed” only where Europe PMC finds the accession in the paper’s full text, or PRIDE records it as the dataset reference. Otherwise the link is inferred from the title and design, and must be checked at intake.

Neither assay can go straight into the processed-table converter. That converter handles DE tables only (see implementation_status.md). So methylation studies need assay-specific raw processing; CIBNOR2021 has now completed Bismark plus pooled-family limma analysis. Proteomics can use the quantified-matrix workflow, as TIMMINS2014 does, with per-study intake for published tables. Coordinate evidence from methylation is also exposed to the assembly change in handling_genome_versions.md.

Methylation: Tier 1

M1 is registered as CIBNOR2021_HEAT_RRSS (2026-09-30) — AREE’s first real methylation study. All 12 libraries share one BioSample, so the design was parsed from experiment_title with the new --label-field / --label-pattern options of aree fetch-samplesheet. Reanalysis completed on 2026-10-16 using family-level pooling and limma: 2 vs 2 families, not 6 vs 6 independent oysters. The 132,587 harmonized regions yielded no BH-significant differences; this is a low-power null result.

# Accession Runs Design (from deposited metadata) Context Why
M1 ✅ PRJNA690951 12 WGBS 2 thermal-resistant (RR52, RR59) + 2 susceptible (SS05, SS35) families × n=3, gill, day 30 of 26–34 °C oscillation (CIBNOR) Thermal tolerance, measured phenotype Same breeding program and challenge as Arredondo-Espinoza et al. 2023, the RR/SS study in issue #6. This is the first public raw data from that design. No methylation paper cites it yet. It is the methylation arm: the RNA-seq reads are still not located. Effective replication is 2 vs 2 families, not 6 vs 6 oysters.
M2 PRJEB81880 40 EM-seq 5 resistant vs 5 susceptible families, gill + mantle, before (T0) and after (T1) POMS, one oyster per family × tissue × time Disease resistance (POMS) A measured R/S phenotype with a before/after design, in AREE’s best-populated stressor class (OsHV-1/POMS: CALLA2026_OSHV, DELISLE2020_OSHV_TEMP). Inferred to be Valdivieso et al. 2025, Sci Total Environ (10.1016/j.scitotenv.2025.178385).
M3 PRJNA562805 12 WGBS intertidal vs subtidal origin × control/heat, n=3, gill Thermal response by habitat origin Confirmed: Wang et al. 2021, Heredity (10.1038/s41437-020-0351-7). Pairs with M1 to give thermal methylation k=2.
M4 PRJNA682817 24 WGBS diploid/triploid × pH 8.2/7.7, n=6 per cell, adult ctenidia (UW) Ocean acidification (exposure) Best-replicated OA methylation design found. Its natural RNA-seq partner, IOCAS2022_OA_ENERGY, failed QC; LUTIER2022_OA_TIPPING is OA but juvenile whole tissue at far lower pH, so not a like-for-like pairing. Lab-internal, so metadata recovery is easy. No publication located. Sister project of PRJNA678408 (Tier 2 #4 above).
M5 PRJNA806944 8 WGBS female gonad, ambient vs low pH Ocean acidification (exposure) Confirmed: Venkataraman et al. 2022, BMC Genomics (10.1186/s12864-022-08781-5). Group labels are not in the run metadata, so recover the split from the paper.

Methylation: Tier 2

Accession Runs Design Caveat
PRJEB105019 60 WGBS 2 families (F14R, H2D) × 3 age cohorts at T0, plus post-challenge R vs S individuals, n=6 (Ifremer DECICOMP) Best design on this list: individual-level resistance outcome, n=6. Released 2025-12 with no publication yet, so treat as unpublished and check the embargo/reuse terms. The EM-seq arm of DECICOMP: its metabolomics arm is MTBLS13463, registered as DECICOMP2025_POMS_AGE (#21). Run titles give the cohort codes: C3 = 4, C7 = 16, C8 = 28 months. Its 24 resistant/susceptible libraries are all from the 28-month cohort, at 500 hpc.
PRJNA609264 47 WGBS early microbial exposure vs control, F1/F2, 2 families, n=3 at most timepoints Confirmed: Fallet et al. 2022, Microbiome. Transgenerational disease protection is directly relevant to breeding. However, the design is a sprawling time course with pools and mixed assays, so intake needs a curated subset.
PRJNA807732 24 WGBS intertidal F0/F1/F2 and subtidal F0, control vs heat, n=3 Probably Wang et al. 2023, Sci Total Environ (transgenerational intertidal). Not confirmed.
PRJNA1113357 9 WGBS V. alginolyticus, gill, 0/6/48 h, n=3 Probably Li et al. 2024, Fish Shellfish Immunol. Small and exposure-only, but pairs with HESSER2024_VCOR on Vibrio.
PRJEB58545 48 WGBS pesticide mixture, F0 × F1 exposure (E/T), gastrula and metamorphosis, n=4 Probably Sol Dourdin et al. 2024, Environ Sci Technol. Opens pollutant_exposure, but no resilience phenotype.
PRJEB60400 246 WGBS POMS-adapted vs naive populations Confirmed: Gawra et al. 2023, Sci Adv. This is population-epigenetic differentiation, not a treatment contrast, so it needs the same new evidence type as the WGS salinity project above.

Skipped: PRJNA684407 and PRJNA833956 (triploid infertility and growth, with no stressor). Also skipped are PRJNA324546 (developmental MeDIP) and the 2013–2016 reference methylomes (no contrast).

Proteomics

The raw-data pool is thin. PRIDE holds 19 C. gigas projects, and most Chinese-lab proteomics papers on the stressor list have no PRIDE deposit. They may be in iProX, which was not searched. Harvesting processed tables from the papers is the realistic route. The Roberts-lab projects make that easiest.

# Accession Design Context Notes
P1 PXD002316 4 treatments (temperature × pH) × 3 tank replicates, larvae OA × warming (multi-stressor) Registered, not harmonized as HARNEY2016_OA_WARM_LARVAE (2026-10-17). The design is sound, but this is a 2-DE study with no deposited spot volumes or gel images; PRIDE holds only spot identifications. The paper reports 12 of ~650 spots, selected at P < 0.02, with no SE, and 9 of 11 accessions are now deleted from UniProtKB. Worth reanalyzing if the SameSpots data are ever released.
P2 PXD015434 early juveniles, temperature Thermal Registered, QC failed as CRANDALL2022_HEAT_SEED (2026-10-17): one silo per temperature, and the two silos differed beyond temperature (see Rejected below).
P3 PXD013262 seed, time × temperature Thermal Registered, QC failed as WANAMAKER2020_HEAT_SEED (2026-10-17): one silo per temperature, technical replicates only (see Rejected below). With P2 also failed, no public thermal proteomics study in this list is analyzable.
P4 PXD000835 ctenidia, ambient vs high pCO₂ Ocean acidification Reanalyzed as TIMMINS2014_OA_GILL (2026-10-17), the first real proteomics study in AREE. 4 individual oysters per pCO₂ × mechanical-stress group, from the paper’s per-oyster NSAF table. Low-power null: 639 proteins testable, none at padj < 0.05. It gives an OA protein layer with SEs alongside LUTIER2022_OA_TIPPING (RNA-seq) and M4/M5. QC passed with warnings.
P5 PXD011365 gill mitochondria, hypoxia–reoxygenation Hypoxia tolerance Registered, not reanalyzed as SOKOLOV2019_HYPOX_MITO (2026-10-17). It is the only public omics dataset found for hypoxia_tolerance, but not usable as deposited: the phosphoproteome has no run-to-condition key; the total proteome is 3 oysters per condition, acquired in condition blocks; and the only quantitative files are Scaffold mzIdentML exports over oyster_v9 RefSeq IDs. Reopen if the authors supply the phospho key and per-oyster NSAF tables.
P6 PXD065668 / PXD065706 gill proteome + phosphoproteome, ERK inhibition Thermotolerance divergence Registered, not reanalyzed as WANG2026_ERK_HEAT (2026-10-17). The quantification is only in proprietary Spectronaut .sne archives, and the paper’s per-pool tables list only significant features. Proteins are the authors’ own gene models, and the runs were acquired in condition blocks. Reopening needs a DIA re-search of the .raw files. It is a pharmacological perturbation, so mechanism support, not population evidence.
P7 PXD064651 mucus, Vibrio infection, iTRAQ Pathogen challenge Registered, cannot be reanalyzed as JIN2025_VIBRIO_MUCUS (2026-10-17). IBT 16-plex with no channel-to-sample key, no quantification table, and a Mascot search run without quantitation. The infected and control channels cannot be told apart. Unpublished.
P8 PXD000905 low vs high OsHV-1 load, 2-DE Pathogen Registered, not harmonized as CORPOREAU2014_OSHV_LOAD (2026-10-17). It is in the same situation as P1. PRIDE holds only identifications of the 25 excised spots; no spot volumes or gel images were deposited. The paper reports only average fold changes for those significant spots.

No PRIDE deposit was found for these papers, so they are processed-table only. Leprêtre et al. 2020, Front Immunol (OsHV-1 proteomics in two families with contrasting susceptibility) is the strongest proteomic resilience phenotype found. Kim et al. 2025, CBP-D covers hypoxia + heat in hemocytes, with transcriptomics and proteomics together. Chen et al. 2023, Fish Shellfish Immunol covers salinity tolerance with multi-omics.

Rejected: PXD008057 (two populations × 0/6/24 h heat). Its sample protocol pools 10 oysters per timepoint per population, which leaves six pooled samples and no replicates. This is the proteomic version of PRJNA623063 below.

Registered, QC failed: PXD013262 (P3, WANAMAKER2020_HEAT_SEED). The 26 runs are 13 pools in technical duplicate, and the authors’ sample metadata on figshare puts every 23 °C pool in one silo and every 29 °C pool in another. Temperature is confounded with silo, and no replicate is biological. The paper’s ASCA found no temperature effect (P = 0.19). The authors’ code also mentions a second 23 °C silo that was never deposited. If those samples exist, re-screen.

Registered, QC failed: PXD015434 (P2, CRANDALL2022_HEAT_SEED). The 16 runs are 4 pools × 4 DIA m/z-range injections. The pools come from two silos, the only 23 °C control and the 29 °C control, each sampled on two days. The hatchery notebook in the authors’ Zenodo release shows the 23 °C silo started at 16–18 °C, set poorly and died out after a ciliate outbreak. The paper’s survival contrast (0 % vs 86 %) is therefore confounded with that history, and no protein reaches adjusted P < 0.05. Both P2 and P3 come from the same Taylor Shellfish 23-vs-29 °C silo trials. Any further proteomics from that program is worth checking for silo replication before claiming it.

2-DE studies (P1, P8) are a separate problem. PRIDE deposits for gel studies usually hold only identifications of excised spots, not spot volumes, so they cannot be requantified. Check for deposited gel images or a spot-volume table before claiming one.

Non-coding RNA, chromatin, and metabolomics candidates

Screened 2026-10-11. Sources: - every ENA read run under taxon 29159 (12,994 runs), bucketed by library_strategy; - every RNA-Seq run with library_selection = size fractionation, or with miRNA/lncRNA/circRNA in its study or experiment title; - MetaboLights (EBI search for “gigas”); - Metabolomics Workbench (species and title search); - a PRIDE refresh; - PubMed searches for C. gigas lncRNA/circRNA and miRNA under stress (18 papers).

Group sizes come from sample_alias and experiment_title.

Non-RNA-seq run counts for M. gigas in ENA on this date:

Strategy Runs
Bisulfite-Seq 607
miRNA-Seq 117
RIP-Seq 44
ChIP-Seq 28
ATAC-seq 24
MeDIP-Seq 21
ncRNA-Seq 8

AMPLICON (3,329) and OTHER (2,069) runs are mostly microbiome and genotyping runs, but LUTIER2022_OA_TIPPING shows that mislabelled RNA-seq can hide there.

The methylation and proteomics findings above still hold: no new stress-relevant WGBS/EM-seq project has appeared since 2026-09-30, and the one new C. gigas PRIDE project (PXD082789, simulated digestion) is food science.

Metabolomics: the real find of this screen

AREE’s only metabolomics study is the simulated GIGAS_GROW06. Both of these are POMS studies, AREE’s best-populated stressor class, and both have a measured disease outcome:

Status 2026-10-11. X1 and X2 are registered as DECICOMP2025_POMS_AGE (#21) and DUPERRET2025_POMS_STARVE (#22). Both are draft PRs, because neither can be analyzed: every Metabolite Assignment File (MAF) in both deposits lists metabolite names but holds no abundance values. The only data are Agilent targeted-dMRM vendor files (.d), which no AREE workflow reads. Ask the submitters (J. Vidal-Dupiol, IHPE) to complete the MAFs, or add an MRM integration step. Curation also found, in the order it matters:

  • X2: diet is fully confounded with MS batch on all four platforms. Fed samples were acquired Dec 2023–Feb 2024 and starved samples May–Jun 2024, with no overlap. Even with abundances, a diet effect here cannot be separated from batch.
  • X2 samples are pools of 10 oysters per tank × family, with 3 tanks per diet, so replication is 3 tanks.
  • X1 age is confounded with challenge year: each age cohort was challenged in a different year. MS batch is not confounded with age.
  • X1 sampling time (0 hpc) is inferred, from the EM-seq arm’s identical 2 × 3 × 6 structure at 0 hpc. Whether the metabolomics and EM-seq samples are the same animals is not stated. The first draft of this table said they were; that was an assumption.
  • Neither deposit has database identifiers or MSI levels, so metabolite mapping confidence starts at unresolved.
  • The studies may not be independent: both use family F14R from the same 2020 Ifremer production, and X1’s MAFs carry X2’s sample columns.
# Accession Design Context Notes
X1 🟡 MTBLS13463 (+ PRJEB105019) DECICOMP (Ifremer): 2 families × 3 ages (4/16/28 mo) × 6 whole animals, 0 hpc before OsHV-1 cohabitation; older oysters are more tolerant of POMS Disease resistance, measured per cohort Registered as DECICOMP2025_POMS_AGE (#21): 28 vs 4 mo (primary) and 16 vs 4 mo. Same project as PRJEB105019 (methylation Tier 2), which is cited but kept out of accessions so it can be registered on its own. Manuscript in preparation; no DOI.
X2 🟡 MTBLS13313 2 families × fed/starved (3 tanks each) × 0–12 hpc, pools of 10; starvation reduced POMS mortality (F14R 34.8% vs 75.5%; F11N 93.3% vs 100%) Disease resistance × nutritional_limitation Registered as DUPERRET2025_POMS_STARVE (#22): starved vs fed at 0 hpc. Published: Duperret et al. 2025, iScience 29:114333 (10.1016/j.isci.2025.114333). Its RNA-seq arm is public and is the better study; see below.

Beyond the empty MAFs, there is no converter for real MTBLS m_*.tsv tables: metabolomics harmonization exists only for the demo format.

The RNA-seq arm of X2

The “integrative omics” in X2’s paper includes RNA-seq, deposited as two ENA projects split by diet: PRJEB86618 (starved) and PRJEB86525 (fed). The design is 2 families × 2 diets × 7 times (0–72 hpc) × 3 tank pools, 83 of 84 libraries deposited, single-end 75 bp. Unlike the metabolomics, library numbering interleaves the two diets.

Registered as DUPERRET2025_STARVE_RNASEQ (#23), one study across both projects. All three comparisons now have harmonized evidence. Comparisons: starved vs fed at 0 hpc (primary, resilience), matching X2’s metabolomic contrast, and 24 vs 0 hpc within each diet (disease). It needed two workflow changes, now merged: single-end input (#25) and, as a general feature, --covariates (#27).

It is fitted on tanks, not with a family covariate. Both families share each tank, so ~ family + condition would count 6 libraries per diet when 3 tanks were treated. Each tank’s two family libraries are summed (--replicate_unit tank, or tank_hpc for the time contrasts), giving 3 vs 3. This is the same lesson as LUTIER2022_OA_TIPPING: find the unit the treatment was applied to before counting replicates.

It shares tanks with X2, so the two are convergent evidence across omics layers, not independent replication.

Metabolomics Workbench has no C. gigas study. Its oyster entries are C. virginica predator/shell studies.

Chromatin accessibility (ATAC-seq)

# Accession Runs Design Context Notes
C1 PRJNA970954 6 ATAC + 6 RNA-seq adult gill, heat (HC) vs no heat (NC), n=3, sample labels shared across both assays (HC.gigas1–3 / NC.gigas1–3) Thermal (exposure) Two-species study (C. gigas vs C. angulata). Paired ATAC + RNA-seq gives a clean, linkable multi-omics pair (peak → nearest gene → DE). Needs a new feature_type for accessible regions and a peak-calling workflow.
C2 PRJNA1212587 6 ATAC intertidal vs subtidal, n=3 Thermal / habitat origin Probably the same group as M3 (PRJNA562805, Wang et al. 2021, intertidal vs subtidal WGBS). Would give methylation + accessibility on one contrast. No publication confirmed.
— PRJNA1113803 4 ATAC fast vs slow growth strains, n=2 Growth Skip: n=2, no stressor.
— PRJNA417263 ATAC + 1 RNA hepatopancreas, 2 technical reps — Skip: reference chromatin map, no contrast.

ChIP-seq / CUT&Tag: annotation, not evidence

PRJNA577897 (HSF1 ChIP, gill, heat), PRJNA1292992 (BTF3L4 CUT&Tag, HSP regulation, 2026), and PRJNA1193684 (CLOCK DAP-seq, thermal adaptation) are all n=1 vs input. They cannot produce effect sizes. They can, however, annotate candidates: “HSF1 binds this promoter” is useful mechanism context on an evidence card. Treat them as a reference layer (data/reference/), not as studies. PRJNA1292166 is a 2026 ChIP method-optimization set (H3K4me3, gill and mantle) and has the same use.

Non-coding RNA

The deposited small-RNA record is weak for AREE’s purposes. Most miRNA-Seq projects are atlases, developmental series, or reproduction/triploidy studies. The stress-relevant ones are mostly unreplicated:

Accession Runs Design Verdict
PRJNA836789 6 miRNA + 6 ssRNA selectively bred strains HH vs CC, n=3, soft tissue Out of scope (checked 2026-10-15). The BioProject describes it as a study of inbreeding depression (one inbred strain with growth depression), not a stressor or resilience contrast.
PRJNA750516 2 miRNA + 6 lncRNA-labelled RNA two groups (RZ_GH vs KTD_GL), one sample alias each Skip: replicates are re-sequenced libraries of one sample.
PRJNA694837 6 miRNA + 6 RNA BM vs OM mantle, n=3 Probably shell colour/biomineralization. No stressor, so low priority.
PRJNA341808 ❌ 3 “miRNA” + 9 RNA larvae: control / bacteria / heat / bacteria+heat, Ion Torrent Rejected 2026-10-14 (see Rejected below): the replicates are technical, and the “miRNA-Seq” control runs are mislabelled RNA-seq.
PRJNA484109 1 miRNA + 9 RNA OsHV-1 infection (Rosani et al. 2020, BMC Genomics, 10.1186/s12864-020-07026-7) Only one small-RNA run is in this project. The paper’s paired miRNA/mRNA libraries must be located before this is usable. Checked 2026-10-15: the six gill RNA-seq runs (S1–S6) carry no group labels in the deposit, so the design is not recoverable from it.
PRJNA266361, PRJNA259995, PRJNA232734 1–3 miRNA stress/atlas, single pooled sample Skip: unreplicated.

Several published stress miRNA studies have no accession in their abstracts. Check their data-availability statements before treating them as raw-data candidates: - exosomal miRNAs after heat or Vibrio (2024, Dev Comp Immunol, 10.1016/j.dci.2024.105174); - miRNAs in haemocyte immune priming (2026, Front Immunol, 10.3389/fimmu.2026.1753252); - haemocyte miRNAs after bacteria and heat (2014, PLoS One, 10.1371/journal.pone.0088397).

lncRNA and circRNA almost never have their own deposits. The C. gigas lncRNA stress papers found re-quantify ordinary RNA-seq: - heat, built from 27 public RNA-seq sets (10.1007/s10126-022-10140-0); - OsHV-1 antiviral response (10.1016/j.fsi.2018.07.032); - growth at low temperature (10.1186/s12864-023-09543-7).

The realistic route to lncRNA evidence is therefore a workflow feature, not a curation task: add lncRNA loci to the quantification index and run them through the existing RNA-seq path for studies already registered (CALLA2026_OSHV, DELISLE2020_OSHV_TEMP, LUTIER2022_OA_TIPPING). PolyA-selected libraries will under-sample non-polyadenylated lncRNAs and miss circRNAs entirely, so record library selection as a caveat.

Other non-RNA-seq projects noted, not shortlisted

  • PRJNA894789 — 5-Aza-dC demethylation × heat, gill, n=3 (0 h control / 6 h H₂O / 6 h 5-Aza). The runs are RNA-seq despite the size fractionation label. It is a pharmacological perturbation of methylation, so it is mechanism support for M1/M3 rather than population evidence, like P6.
  • PRJNA291364 — diuron-exposed parents → spat, WGBS + RNA-seq, n=2. This is transgenerational pollutant_exposure, and too small to pool.
  • PRJNA689936 — “wild oysters” WGBS, 6 runs from one sample alias. Skip.
  • PRJNA748038 — m6A RIP-seq (MeRIP) across development. It has no stressor, and RNA modification is a new evidence type.
  • PRJEB77044 — 96 runs, BC field stations × mantle/hemolymph. Labelled RNA-Seq with size fractionation; whether it is small RNA or standard RNA-seq was not checked. If standard, it belongs on the RNA-seq list as a field/environmental gradient design.

Rejected, and why it matters

PRJNA623063 — “Transcriptome of the Pacific Oyster, Crassostrea gigas Larvae after Vibrio alginolyticus Challenge”, 12 RNA-seq runs.

On title alone this looked like the ideal second study: same life stage, same stressor class, and nearly the same phenotype as HESSER2024_VCOR. The run metadata says otherwise — all 12 runs share a single sample_alias (Oyster_M49) across 12 timepoints (T01–T12). It is an unreplicated time course, so no valid differential-expression contrast can be computed from it.

Recording this because the failure is invisible from the abstract, and because a curator working from titles would have spent real effort before finding out. Check sample_alias before downloading anything.

PRJNA826964 — “The compromised energy management of Pacific oysters (Crassostrea gigas) under ocean acidification conditions”, 18 RNA-seq runs. Registered as IOCAS2022_OA_ENERGY and kept in the registry with qc_status: failed, so the reason stays on record.

This one passed the sample_alias check: 18 distinct aliases, labelled “biological replicate 1–3” in every group. It failed only after full-depth reanalysis. Within each group, replicates vary about as little as read sampling alone would make them — median dispersion 0.0002–0.0016, against 0.03–0.05 for CALLA2026_OSHV’s real biological replicates. Each group is most likely one RNA sample sequenced three times. The symptoms, in the order they appeared: 43–48% of genes “DE” at every timepoint; PC1 = 96.5% of variance within a 3 vs 3 contrast; three OA groups that agree with the controls as well as with each other; and fold changes that barely replicate between timepoints.

The danger is not wasted compute. Harmonized, such a study carries an lfcSE far too small, so inverse-variance pooling would let it outweigh every real study it joins. Run scripts/check_replicate_dispersion.py on every raw_reanalysis study before harmonizing it.

PRJNA1196326 — “Transcriptional acclimation following transgenerational exposure of oysters to ocean acidification”, 6 RNA-seq runs (IOCAS). Registered as IOCAS2024_OA_TRANSGEN (#17) with qc_status: failed. It is the same failure as PRJNA826964, from the same institute: dispersion 0.0008–0.0010 after a full-depth run, 42.6% of genes “DE”, and an expressed-SNP screen showing each arm is one larval pool sequenced three times (within-arm allele-frequency difference 0.061 against 0.051 for one RNA). The two arms also descend from different parents, so parental pH is confounded with parental genotype.

PRJNA341808: C. gigas D-hinged larvae, control / bacteria / heat / bacteria + heat, 3 Ion Torrent runs per group, 26.2 GB. Rejected before download, 2026-10-14, by scripts/prescreen_replicates.py. It is the first study turned away by the pre-screen, not after a full reanalysis.

Each group’s three runs share one BioSample (SAMN05728707–10) and one sample_alias, which was already a warning. The pre-screen, run on the first 1 M reads per run from NCBI, gave these median dispersions:

Group Median dispersion
heat −0.0010
bacteria 0.0039
control 0.0050
bacteria + heat 0.0058

Two groups fail and two sit at the 0.005 threshold. The pre-screen reads high (LI2022_HEAT_PLOIDY’s real replicates scored 0.019–0.022 on it), so all four groups are in the technical-replicate range. Each group is most likely one larval RNA sample sequenced three times.

Two things the screen also settled: - The three control runs deposited as miRNA-Seq are ordinary RNA-seq. They map to the transcriptome as well as the stress runs (55% vs 53–58%). - A genotype check could not have decided this case. Larval pools from one cross share parental genotypes, whether or not the pools are independent.

An expressed-SNP screen run independently on 2026-10-13 (first ~2 M reads of the control and heat runs, 167 sites) agrees but cannot decide alone: runs within a group differ in allele frequency by 0.062 against 0.057 expected from one RNA sample, and groups differ from each other by 0.146. That is what one RNA per group looks like, but large pools from one cross would look much the same, as noted above.

PRJNA694496 — Tan et al. 2021, Front Physiol (10.3389/fphys.2021.663023), larvae of a thermotolerance-selected line vs a control line, 6 runs. Rejected because the design cannot be recovered. All six runs share one BioSample, whose tissue (“gill”) also contradicts the experiment title (“larva”). The run aliases are vendor IDs (W056-01-T01…T06), and the paper says only “three replicates for each group” without mapping runs to groups. Assigning groups would be a guess. The authors also note that their survival difference “may not fully reflect the thermotolerance” of the lines.

PRJNA877226 — “RNA-seq of Pacific oyster infected with Vibrio pathogen under high temperature”, 6 runs (Ocean University of China). The groups are recoverable after all: the BioSample titles give V. alginolyticus-infected gill at high temperature, 12 h vs 48 h, n=3. But the deposit has no uninfected control and no normal-temperature arm. The run names (C33-n) match the 23 °C arm (C23-n) of the same lab’s PRJNA756710 (LI2022_LOWSAL_VIBRIO, #26). Using that study’s 30 ppt libraries as a control would cross BioProjects and probably sequencing batches, so it is not done. A within-infection time contrast is not stress-vs-control evidence.

PRJNA298285: larval OA × warming, University of Brest (2015). It has 24 RNA-seq runs: trochophore / early / late veliger × pH 7.9 / 8.1 × 20 / 22 °C, two runs per group. Each group’s two runs share one BioSample. Rejected before download, 2026-10-15, by scripts/prescreen_replicates.py (first 1 M reads per run from NCBI, the four 24 h groups). Every group’s median dispersion is negative (−0.0047 to −0.0065): the two runs vary less than read sampling alone predicts, which is the signature of one library sequenced on two lanes. Effective n is 1 per group, for all 12 groups.

Two lessons from these rejections. One BioSample per group, or a whole study under one BioSample, is the cheapest early warning of pseudo-replication. And replication can be checked before the full download: scripts/prescreen_replicates.py does it from a subsample. An expressed-SNP genotype check (check_replicate_genotypes.py, currently in sr320/AREE-co) complements it when libraries are single animals or small pools of unrelated animals. It cannot decide for large pools of siblings, which converge on their parents’ allele frequencies whether or not they are independent.

Sourcing notes

  • 93 candidate BioProjects were found for C. gigas / M. gigas across the stressor terms in AREE’s ontology (thermal, acidification, pathogen, salinity, hypoxia, desiccation). The shortlist above is the subset passing the screen.
  • GEO holds only 10 oyster expression series and is largely pre-2015; the useful corpus is in SRA/ENA. Do not treat a GEO search as sufficient.
  • Hypoxia is badly underrepresented: exactly one BioProject matched, and it has no reads in ENA. AREE’s phenotype ontology carries hypoxia_tolerance, but there may be no public transcriptomic dataset to populate it.

Bibliographic metadata for every cited paper was retrieved from PubMed — including the author order, which one of them had wrong until 2026-09-05.