[study] CIBNOR2021_HEAT_RRSS (PRJNA690951): first reanalysis

[sr320]

Claimed on Chlo (sr320's workstation), 2026-10-09, for the first reanalysis of a study registered in #9. aree start-study refuses registered accessions, so this branch was created by hand. Run finished 2026-10-17. Draft until #67 merges.

Study

• study_id: CIBNOR2021_HEAT_RRSS
• Accession(s): PRJNA690951 (SRR13482027–SRR13482038)
• Mode: raw_reanalysis (WGBS, Bismark → methylKit)
• Machine and storage used: Chlo. FASTQ on blue-block (/Volumes/blue-block/aree/raw/CIBNOR2021_HEAT_RRSS/, 129.1 GB), Nextflow work dir and results on the Samsung T5.

Checklist

• aree validate-study registry/studies/CIBNOR2021_HEAT_RRSS.yaml passes (registered in [study] CIBNOR2021_HEAT_RRSS (PRJNA690951) #9)
• Sample sheet and FASTQ manifest generated by aree fetch-samplesheet, not hand-written ([study] CIBNOR2021_HEAT_RRSS (PRJNA690951) #9)
• FASTQ MD5s verified against the manifest before the run (24/24; 12 corrupt resumed transfers re-fetched)
• Workflow manifest(s) committed under data/studies/CIBNOR2021_HEAT_RRSS/, with parameters, tool versions and input checksums
• Standardized result tables and the HTML report committed; no raw reads, work/ or results/
• qc_status and analysis_status in the YAML reflect what was actually run (passed_with_warnings, harmonized)
• Limitations and quality flags record anything that weakens the evidence (replication, phenotype definition, excluded samples)
• Curation tests in tests/test_cibnor2021_study.py pin the decisions that are easy to erode later
• No edits to shared code, vocabularies or docs; those went in a separate PR (Fit methylKit DMRs on families, not libraries (methylation.replicate_unit) #15, Test pooled family tiles with limma (or DSS), not per-region methylKit MN #67)

Results

AREE's first complete real methylation run. Code: #15 (merged) and #67 (must merge first).

• Alignment. All 12 libraries: 50.0–53.9% unique mapping with --score_min L,0,-0.6. Bismark's default mapped 13–15%: the threshold is too strict for this genome, and the library is directional (bismark_scoring_diagnostic.txt). Duplicates removed: 6.3–9.4%.
• Methylation-bias trim. Read 2's first 20 bases are end-repair artifacts (84% CpG methylation at position 1), so the extractor ignores them. After the trim, CHH methylation is 0.6–1.1%, i.e. at least 99% bisulfite conversion (library_qc.tsv).
• Replicates. Within each family, libraries vary 6.7–8.1× more than binomial sampling, consistent with distinct animals despite the shared BioSample (replicate_dispersion_qc.tsv).
• DMR test. Strands merged, libraries pooled by family, CpGs kept at ≥10× in all 4 families. Units: 31,166 genes (body + 1.5 kb promoter) and 101,421 intergenic 1 kb tiles. Test: limma moderated t, 2 RR vs 2 SS families.
• Result: a low-power null. Nothing reaches q < 0.05 (smallest q 0.57). 25 genes have p < 0.001, about the 31 expected by chance. Every unit carries an effect, a standard error and a p-value, so it pools when a second thermal methylation study arrives.
• Harmonized. 132,587 records. 30,732 gene-linked candidates, all emerging and resilience_associated.
• Pages rebuild with this evidence: no errors, 0 spurious multi-omics candidates, and the 889 high-priority candidates unchanged.
• Run: Nextflow session cdc10126-dbdf-465e-b876-71efcae13f47. The curation notes have every first-run fix and the decisions behind them.

🤖 Generated with Claude Code