fallingstar10/otter-data

OTTER reproducibility data

This public dataset contains versioned, real-data fixtures and provenance records used to reproduce the executable validation claims for OTTER (Orchestrated Transcriptomic, Tumor-xenograft, and Epigenomic Reporting Workflow).

Scope

The dataset contains compact, deterministic derivatives of accessioned sequencing data; BAM/BAI fixtures; expected outputs; Gate A–D oracles; RNA-PDX and BS-PDX mixture inputs; and Xenofilx parameter-grid inputs. It does not contain manuscript materials, private Slurm controllers, full-scale production runs, or full reference-genome snapshots.

GitHub Actions consume this dataset for fixture-based checks only. They do not run full FASTQ mapping/alignment. Normal CI reference acquisition uses official upstream FASTA/GTF sources and pinned reference contracts. The separate `fallingstar10/xdxtools-genomes` dataset is the offline reference snapshot and is updated only after CI validation.

Layout

provenance/
  schema.json
  files.jsonl
  checksums.sha256
fastqcx/rna-pdx-srr30880970/
bamdriver/gate-a/rna-pdx-srr30880970/
pairbam/gate-b/rna-pdx-srr30880970/
xenofilx/gate-c/{rna-hg38,bs-pdx-graft-hg38,bs-pdx-srr36187610-host}/
xenofilx/gate-d/bs-pdx-srr36187610/
xenofilx/mixture/rna-pdx/<cell>/{graft.bam,host.bam,truth.tsv}
xenofilx/mixture/bs-pdx/<cell>/{graft.hg19.bam,host.mm10.bam,truth.tsv}
xenofilx/benchmark/tables/{note4-summary-table.csv,note4-confusion-matrix-table.csv,table-s5-gradient.tsv,table-s6-gradient.tsv,table-s7-parameter-grid.tsv}

Each line in provenance/files.jsonl conforms to provenance/schema.json. The required owner object identifies the repository, workflow, validation gate, and assay. The required derivation object records the SRA accession, Paracloud source path, method, seed when applicable, and input/output counts.

Mixture gradients (Supplementary Note 4)

xenofilx/mixture/bs-pdx/ holds the complete 0–99% bisulfite composition gradient behind Supplementary Table S6: 43 cells, each with a human-graft alignment to bisulfite-converted hg19, a mouse-host alignment to bisulfite-converted mm10, and a per-fragment construction truth. Cell labels are human0-replicate1 (pure-mouse control) and human0p<fraction>-replicate<n>, for example human0p5-replicate1 (50% human).

Sources are human RRBS SRR31480456 and mouse RRBS SRR10025242, aligned with Bismark 3.1.0; each cell carries 1,000,000 complete fragments. The three replicates of the 60% cell (human0p6-replicate1..3) are also the inputs of the Supplementary Table S7 threshold/penalty audit. Every file's SHA-256 matches the value recorded by the original evaluation manifests, so the published fixtures are provably the same bytes those analyses consumed.

xenofilx/mixture/rna-pdx/ holds the complete 0–99% RNA-seq composition gradient behind Supplementary Table S5: 43 cells, each with a human-graft alignment to hg38 (STAR), a mouse-host alignment to mm10 (STAR), and a per-fragment construction truth. Cell labels are human0-replicate1 (pure-mouse control) and human0p<fraction>-replicate<n>, for example human0p5-replicate1 (50% human).

Sources are human RNA-seq SRR1039508 and mouse RNA-seq SRR037954, aligned with STAR; each cell carries 1,000,000 complete fragments (n=3 replicates per non-zero fraction). Every file's SHA-256 matches the value recorded by the evaluation manifests.

Benchmark tables and confusion matrices (Supplementary Note 4)

xenofilx/benchmark/tables/ publishes the machine-readable summary tables, confusion matrices, and audit manifests corresponding to Supplementary Tables S5, S6, and S7:

• —note4-summary-table.csv — Unified 117-row x 24-column cross-assay benchmark summary covering the RNA-seq 0–99% gradient (n=3), BS-seq 0–99% gradient (n=3, three arms), BS-seq 60% parameter grid (n=3, three arms), and authentic PDX accessions (SRR30880970, SRR36187610).
• —note4-confusion-matrix-table.csv — Comprehensive 446-row x 32-column table recording the full 2x2 confusion matrix (TP, FN, FP, TN exact integer counts and derived rates) for every single replicate (329 rows) and stratum mean (117 rows).
• —table-s5-gradient.tsv — RNA-seq 0–99% gradient benchmark table (14 non-zero points n=3).
• —table-s6-gradient.tsv — BS-seq 0–99% gradient benchmark table with modern, patched legacy, and conventional unpatched paths (all n=3).
• —table-s7-parameter-grid.tsv — 4x3 mismatch threshold and unmapped penalty factorial grid across all three paths (n=3).
• —table-s8-reference-grid.tsv — 2x2 factorial reference genome combination benchmark (hg19/hg38 x mm10/mm39) on 60% BS-seq mixtures across all three tool paths (n=3).
• —note4-reference-grid-summary.csv — Reference genome combination summary table (12 stratum mean rows).
• —note4-reference-grid-confusion-matrix.csv — Complete 48-row x 28-column table recording the full 2x2 confusion matrix (TP, FN, FP, TN integer counts and derived rates) for all 36 individual evaluations and 12 stratum means.
• —per-cell-expected-s5.json, per-cell-expected-s6.json, per-cell-expected-s7.json — Machine-auditable per-cell reference metric targets used by CI parity gates.
• —rna-mixture-cells.json — Immutable fixture contract for all 43 RNA-PDX mixture cells.

Validation policy

• —All supplied files are checksum-verified before use.
• —RNA-PDX and BS-PDX evidence is independently scoped.
• —Xenofilx parameter recommendations apply only to the documented assay and fixture.
• —Results and logs from scientific CI are published as GitHub Actions artifacts, including failure cases.
• —Full raw-read mapping, manuscript-scale benchmarking, and private Paracloud operations are outside GitHub-hosted CI.

Citation

Use the future OTTER software release tag together with the exact Hugging Face dataset revision SHA shown by the workflow or provenance record.