Step 01
Start with run and sample metadata
Barcodes, sample type, extraction batch, run ID and client context are captured before analysis. This prevents the most common failure mode: a correct classification attached to the wrong sample or report.
Microbiome & pathogen detection Nanopore 16S
Case study · Microbiome · Genics Pty Ltd
One lead bioinformatician, one MinION: sample to pathogen report in 2 days.
Genics depended on external providers for microbiome analysis: 8 to 20 weeks from sample receipt to a client report, on someone else's timeline. As the company's lead bioinformatician, I brought the whole chain in-house on Oxford Nanopore and owned it end to end — what now runs as Microbiome Health Analysis (MHA).
For the specialist Oxford Nanopore 16S workflow: raw signal → basecalling → mapping → variant calling → pathway analysis → statistical testing → database integration → automated client reporting. OTU and ASV approaches to absorb Nanopore's higher per-read error rate. Validated against the former provider's Illumina results.
2 days
Turnaround, from 8–20 weeks, as routine
Lead
Lead bioinformatician; built every computational stage
Nanopore
16S, OTU/ASV, pathogen calls
Jan–Aug 2026
Lead Bioinformatician, Genics
The result
The turnaround number is real, and it is routine, not a best case.
This is the same timeline, drawn to scale. The outsourced bar spans the full 8–20 week range Genics used to wait on; the in-house bar is what sample-to-report now takes, on a normal run, not a demo run.
Outsourced
8–20 weeks
In-house
2 days
048121620 weeks
Solid bar = the shortest wait (8 weeks); hatched = the range up to 20 weeks.
2 days is roughly 1–4% of the old timeline — the number is not exaggerated for effect.
Sample-to-report: one pipeline, five checkpoints
- Sample + metadataBarcode, extraction batch, run ID and client context captured before any analysis runs.
- Basecall + QCYield, read length, per-barcode depth and contamination controls checked before taxonomy.
- Taxonomy + pathogen flagOTU/ASV taxonomic assignment plus pathogen-specific flagging, tuned for Nanopore's error profile.
- SQL storageEvery run, sample, taxon and flag written to a queryable, auditable database.
- Client reportExecutive summary, detected organisms, caveats and confidence, plus a technical appendix.
What a run looks like
One run, many sample types, two targets.
Run anatomy
2 to 56
barcoded samples per run
~1,500 bp
full-length 16S amplicon on a MinION
16S + ITS
bacteria (SILVA) and fungi (ITS reference), via wf-16s
2 days
receipt to report sent, as routine
WaterSoilAnimal faecesAnimal remainsBloodTissue
Run capacity and sample types only; client-specific counts and commercial figures are not disclosed.
The method
One person, the whole chain, six checkpoints.
As lead bioinformatician at Genics, I designed the workflow, coordinated lab staff (DNA extraction, enrichment, Nanopore sequencing) and IT (infrastructure), and built every computational stage myself.
Step 02
QC before interpretation
Yield, read length, per-barcode depth and contamination controls are checked before taxonomy. Bad runs fail early instead of producing polished but misleading reports.
Example command shape
nanopore-qc --run RUN_ID --barcodes samples.csv --min-reads 5000 classify-16s --reads fastq_pass/ --db curated_16s_db --out taxonomy.tsv report-qc --taxonomy taxonomy.tsv --controls controls.yaml
Step 03
Handle Nanopore's error rate with OTU and ASV, not brute force
Nanopore's per-read error rate is higher than short-read Illumina data. Rather than fight it with coverage alone, the pipeline combines OTU clustering and ASV resolution so taxonomic and pathogen calls stay reliable at commercial turnaround.
Step 04
Persist results in SQL
Each run writes structured outputs to a database so old reports can be regenerated, audited and compared against later database versions.
Schema sketch
runs(run_id, flowcell, started_at, db_version, status) samples(sample_id, run_id, barcode, client_ref, sample_type) taxa(sample_id, taxon, rank, reads, relative_abundance) flags(sample_id, flag_type, severity, review_status)
Step 05
Generate a report, not a data dump
The final artefact is shaped for non-specialists: executive summary, detected organisms, caveats, confidence and a technical appendix for review. Lab-facing dashboards give the team visibility without needing to read raw output.
Step 06
Write it down so it survives without you
SOPs, threshold definitions, failure modes and rerun instructions were written and centralised specifically because one person understood the whole chain. The workflow was designed not to depend on that being permanent.
“I was the only person in the company who understood the chain from A to Z, so I was also the one translating between the lab, IT and the client.”
Adrien Oliva · Lead Bioinformatician, Genics Pty LtdValidation and limits
Nanopore was checked against the old Illumina pipeline, and the two don't fully agree.
The profiles differ. They were never claimed to match.
Overall taxonomic profiles from Nanopore and from the former provider's Illumina pipeline differ — two different platforms and two different pipelines, which is expected. What was validated, and holds up, is pathogen detection: Nanopore calls are nearly as accurate as the Illumina baseline despite the platform's higher raw error rate. That distinction is deliberate; overclaiming concordance is exactly the kind of thing a commercial report cannot afford to get wrong.
Species-level calls are not always defensible
Short amplicons can fail to separate close species. The report exposes confidence and avoids overcalling rather than forcing a single species name.
Relative abundance is not absolute load
Composition shifts are meaningful, but they are not direct cell counts unless the assay includes controls designed for that purpose.
Database versioning matters
A result can change because the reference database changed, not because the biology did. Production pipelines need database provenance, not just software provenance — part of why results are stored in SQL rather than flat files.
Output
What the client receives.
Microbiome report package
pipeline handoverReportHTML/PDF customer summaryreview-ready
Tablestaxa, flags, QC, metadatamachine-readable
Databaserun/sample/taxa schemaauditable
SOPrerun, failure modes, thresholdshandover
No client names, sample counts or commercial figures are disclosed here — only the pattern that made 2-day turnaround routine.
Need microbiome reporting that does not collapse at handover?
Start with a diagnostic, then build the smallest production path from raw reads to report.
Keep reading
Other case studies
Case studyGeneGuardian
Dynamic consent for genomic data, live at a pilot hospital
Read it →Case studyPrecision medicineReference bias in underrepresented populations
Read it →Case studyPharmacogenomicsStar-allele calling in 2,503 people
Read it →Case studyNGS & WGSRaw reads to a decision, across platforms
Read it →