Who it is for
Research teams that need a transcriptomics pipeline they can defend: deterministic species and platform gates, audit-grade provenance, and a clean handoff to downstream analysis or manuscript figures.
strict-omics · project lane
LLM proposes. Deterministic gates decide.
Run the first four ingestion gates in your browser right now — parse, QC, species gate, trim. Production alignment, RO-Crate provenance, and the full pipeline run on our Nextflow / Snakemake backend.
Fail-closed ingestion gateBrowser-side QC + speciesContainer-pinned productionRO-Crate provenance
Up to 200K reads processed locally per run. No upload. No server pipeline call.Have a CSV instead? Try the multi-omics tool →
What we do
We run a two-branch transcriptomics factory (microarray vs RNA-seq) behind a fail-closed ingestion gate. The LLM proposes candidate datasets and metadata; a Pydantic-validated gate decides what enters the pipeline.
What you get
A versioned run manifest, an RO-Crate provenance bundle, MultiQC-aggregated QC, a species-verified sample list, container-pinned preprocessing outputs, and a decision-ready brief you can act on.
Operating steps
The workbench above covers steps 1, 2, and 5. Steps 3 and 4 (alignment and containerised production QC) run on the backend.
01
Ingestion gate
Repository metadata, platform/assay fields, publication cross-check. Fail-closed: conflicting evidence moves to manual review or rejection. The browser workbench above runs the first 4 gates locally.
02
Empirical species check
FastQ Screen / Kraken2 for RNA-seq; verifyBamID2 + CrosscheckFingerprints for human samples. The browser workbench uses a k-mer index to fail-closed if the dominant species is not on the allow-list.
03
Technology-specific branch
Microarray (RLE, NUSE, percent present) and RNA-seq (FastQC, RNA-SeQC 2, RSeQC) are never mixed. Batch effects are detected before they are corrected. Production runs only.
04
Containerised QC
Pinned Nextflow / Snakemake runs, MultiQC aggregation, batch-aware thresholds. ENCODE-style read depth and replicate standards where applicable. Production runs only.
05
Provenance & handoff
DataLad-versioned data, RO-Crate workflow-run provenance, Git-versioned code, and a decision-ready brief that links every output back to a study accession. Production runs only.
Stack
Production-grade tools, pinned by digest.
Every component is selected for portability, auditability, and the ability to rerun a clean pipeline and reproduce the output.
Nextflow
Production orchestration for HPC, cloud, and workstation.
Snakemake
Leaner alternative, especially for R-heavy custom workflows.
nf-core conventions
Style guide and quality floor for reusable pipelines.
MultiQC
Aggregated QC report across all modules.
DataLad
Git-annex versioning of large raw and derived datasets.
Workflow Run RO-Crate
Captures execution provenance with inputs and outputs.
Pydantic + XML prompts
Local schema validation so the LLM cannot relax scientific constraints.
FastQ Screen / Kraken2
Empirical species verification before alignment.
verifyBamID2
Human-sample contamination and identity check.
Standards we enforce
Repository-native metadata, minimum-information standards, and a bilingual controlled vocabulary. We use these as hard gates, not as guidelines.
MIAME / MINSEQE
Minimum information standards that pin what a usable study must report.
MAGE-TAB / SOFT
Repository-native formats for ArrayExpress, BioStudies, and GEO sample and platform metadata.
ENCODE bulk RNA-seq
Read length >= 50 bp, two or more replicates, ~30M aligned reads, Spearman >= 0.9 isogenic / >= 0.8 anisogenic.
Bilingual controlled vocabulary
Canonical English ontology terms for tissue, disease, strain, and perturbation; Korean mirror for ops.
Boundaries
The LLM proposes candidate inclusions and never relaxes scientific constraints. Final inclusion is a deterministic validator decision.
Microarray and RNA-seq never share a preprocessing branch. Different metadata, raw files, QC, and batch behaviour.
Container digests, reference builds, and annotation releases are pinned in the run manifest. A clean rerun reproduces the output or the pipeline is not yet production.
Route preview
Request a Paid Brief
Send a short note and we will return a route preview, an owner, and a fit score. Project-tier engagements start at \u20A98M.
Which organism and assay type are you working with? (Homo sapiens, Mus musculus, microarray, bulk RNA-seq, single-cell, spatial, etc.)
Do you have raw data, and in what format? (FASTQ / SRA / BAM for RNA-seq, CEL / IDAT for microarray, or processed matrices only.)
What decision does this pipeline need to support? (target ID, cohort selection, manuscript figure, audit-grade dataset, regulatory submission, etc.)
24h response targetShort intake, clear next step
Submissions are routed into the Brown Biotech Notion intake hub.
Triage preview
Send a concise project brief
Share just enough context to route the request well. You'll see the route, owner, approval gate, and next action after submit.