What Happened
Sequencing technology · Genomics · Citizen science · Workflow briefhome-seq blog, April 2026 · record to August 2026
"How I sequenced my genome at home" — read, checked where it could be, and set against the clinical standard
One Flow Cell and Four Hours of Handwork Buys About 10× Coverage
A hobbyist wrote up doing whole-genome sequencing at home on an Oxford Nanopore MinION: about four hours of hands-on work, 72 hours end to end, and a single flow cell yielding around 30 Gb — about 10× coverage of a human genome. They reported spending roughly US$1,100 on reagents in April 2026; Oxford Nanopore's own list prices for the same items have since moved, some down and one up. 10× coverage is enough to explore variants common in the population, but it sits well below the 30× that population-scale efforts such as the UK Biobank use for confident variant calls. What follows is the workflow, the costs as they stand today, and exactly where the result runs out.
- $1,100reported cost per run, Apr 2026 (author's own account)
- 4 hhands-on time
- 72 hstart to finish
- ~10×whole-genome coverage from ~30 Gb — below the 30× standard, below
How Firmly Each Part Stands The blog and its coverage are confirmed. What happened inside it is self-reported.
| How firm | What | Who says so |
|---|---|---|
| Confirmed | A blog titled "How I sequenced my genome at home" exists at a real, working address, and its own page metadata names its author's handle. | home-seq blog, read directly (HTTP 200); corroborated by Tom's Hardware, 21 Apr 2026, read directly |
| Confirmed | The MinION Mk1D device, the R10.4.1 flow cell, and the SQK-LSK114 ligation kit are real, currently-sold products, at the list prices given further down this page. | Oxford Nanopore's own price list and product pages, captured 24 Aug 2026 |
| Confirmed, but not verifiable here | That the author personally performed cheek-swab extraction, library prep, and a 48-hour MinION run at home, and obtained around 30 Gb of data at roughly 10× coverage. | The blog's own first-person account; no raw data, lab notebook, or third-party replication is available to check it against |
| Confirmed, but not verifiable here | The author's own reported cost (~US$1,100) and bench timings (~4 h hands-on, 72 h total). | Same first-person account |
Timeline
From a 2024 Device to an April 2026 Kitchen Table Four dated points, two years of hardware apart
- 22–23 May 2024Oxford Nanopore announces the MinION Mk1D at its "London Calling" conference. It supersedes the Mk1B the workflow below was run on, and is the generation the cost table prices.
- ~18 Apr 2026The run the blog describes appears to take place — implied by a dated file path in its own worked command example, not stated outright.
- Before 21 Apr 2026The blog post, "How I sequenced my genome at home", goes up. It carries no visible byline or dateline; the exact date could not be established.
- 21 Apr 2026Tom's Hardware (Mark Tyson) reports on the blog, naming its equipment list and its author's stated family motivation.
The Argument
Why Do This, and Where the Claim Runs Out The author's own three reasons
- Curiosity, and the wish to take it apart — the author describes owning a Raspberry Pi and a Jetson Nano for the same reason: a system they can touch, change, break and rebuild, which they say biology never felt like to them growing up.
- A family history of autoimmune disease — the author gives this as the motivation to look for answers in their own genomic data. This is reported as personal context, not as a finding this page can verify or build on.
- Understanding the workflow itself — not just reading a report at the end, but knowing the purpose and failure mode of every step, which is the actual substance of what follows on this page.
One Flow Cell, Two Ways to Spend Its Budget Breadth or depth — the choice decides what can later be claimed
| Measure | Option A — shallow whole genome | Option B — adaptive sampling |
|---|---|---|
| Coverage | ~10× | ~30–50× over the target region |
| Scope | The whole genome | A specified panel, as a BED file |
| Target size | — | Under 1% of the genome is ideal; under 5% keeps 30× reachable |
| Rare variant detection | Not dependable — cannot reliably separate a real one-in-ten-reads signal from noise | Reasonably confident, within the panel |
| Panel design effort | None — no panel needed | Coordinates must be looked up and a BED file built (an LLM helps with this) |
| Suits | A broad first look at the genome | Pharmacogenes, HLA, autoimmune loci — a specific, named question |
Adaptive sampling is what makes a home MinION run more than a curiosity: the sequencer reads the first ~500 bases of every fragment, checks them against a reference, and — if the fragment is off-target — reverses the voltage to eject it, handing the pore's time back to the regions of interest. It is, in effect, free targeted enrichment, with no custom DNA probes, no PCR, and no specially designed library. But neither option changes the underlying arithmetic: the depth generally used for confident, clinical-grade interpretation is higher than either gets on its own from a single flow cell. The UK Biobank's 490,640-genome resource was sequenced to an average depth of 32.5×, with at least 23.5× per individual — read directly from the paper describing it — and Illumina's own published throughput tables assume more than 120 Gb per sample to reach that same 30× figure. A home run's 10× whole-genome pass, or its 30–50× on a narrow panel, is a real result and not the same tier of result.
What Others Add
How a Nanopore Reads DNA Bases recovered from changes in current
- A flow cell's array holds about 2,048 pore sites; Oxford Nanopore guarantees at least 800 active before a run is worth starting. As DNA threads through an active pore, the change in current is decoded by a neural network into A, C, G and T.
- Reads can run to tens of thousands of bases, far beyond the 150 bp that a standard short-read platform such as Illumina's NovaSeq runs, which helps with difficult regions and structural variants.
- One flow cell run for 48 hours produces about 30 Gb (Oxford Nanopore's own figure for a fresh cell is 20–40 Gb). A human genome is about 3.2 Gb, hence roughly 10× coverage.
- Where a methylation-capable model is used, 5mC and 5hmC calls are written straight into BAM tags — a capability that arrived by a software update to the basecaller, with no new chemistry and no new flow cell.
Equipment and Consumables, Priced Twice What the author reported paying, and what Oxford Nanopore lists today
| Item | Author reported, Apr 2026 | Current list, Aug 2026 | Note |
|---|---|---|---|
| MinION Mk1D | ~$3,200 | $3,150 | The sequencer itself, reusable; ONT's price includes 12 months of standard support |
| R10.4.1 Flow Cell (FLO-MIN114) | ~$900 | $840 | One per run, single use |
| SQK-LSK114 Ligation Kit | ~$100/rxn (6-rxn pack ~$610) | $720 / 6 rxn ≈ $120/rxn | Ligation-based library preparation |
| NEBNext Companion Module v2 | ~$55/rxn | Could not be established | End repair and ligation enzymes; NEB's own site returns a bot check to an automated request |
| Monarch T3010 gDNA Kit | ~$3/prep | Could not be established | Buccal DNA extraction; same access limit as above |
| Flow Cell Wash Kit (EXP-WSH004) | ~$17/wash | $120/kit | For a mid-run wash and reload; ONT does not state washes per kit |
| Sundries | ~$50 | Varies by supplier | LoBind tubes, PBS, ethanol and so on |
The author's own reported total for one run in April 2026 was about US$1,100. That figure is not this project's own re-derived total: of its six line items, the device and the flow cell are cheaper on Oxford Nanopore's current list, the ligation kit's pack price is higher, and two NEB reagent lines could not be independently re-priced from a source this project could read. Most reagents also come in laboratory-scale packaging — the ligation kit's pack covers six reactions and the NEB module's 24 — so anyone running this once wastes most of a pack, which is the hardest part of the cost to avoid regardless of which year's prices are used.
Six Steps at the Bench From setup to alignment
- 01Set up, check poresat least 800 active
- 02Extract DNA~30 min
- 03Prepare library~70 min
- 04Loadhighest risk
- 05Sequence, monitor48 h
- 06Basecalland align
Where Library Preparation Goes Wrong Five specific failure modes, from the author's own account
| Step | What matters |
|---|---|
| Drying AMPure beads | Thirty seconds is enough. Longer and they crack and stick to the tube wall, losing DNA irreversibly. |
| The second cleanup | Use Long Fragment Buffer. Ethanol destroys the motor protein on the adapter. |
| Enzyme mixes | Never vortex. Glycerol-containing enzyme solutions lose activity once they foam; flick to mix instead. |
| Ligation buffer | It is viscous. Pipette-mix slowly, or it will look mixed while still layered, and ligation efficiency suffers. |
| A good yield | 150–450 ng of library in 15 µL; load 12 µL and keep the rest for a reload. |
Loading: Air Is the Enemy And three things to watch
The greatest risk is a bubble entering the flow cell: the author reports that a bubble on their own first run stopped every pore lighting up, and was drawn back out just before it reached the array. Draw back no more than 30 µL of storage buffer, add liquid slowly, and stop the moment a bubble appears. Watch Oxford Nanopore's own priming and loading tutorial through before starting.
- Pore occupancyA decline over time is normal; at 30% or below, do a nuclease wash and reload.
- ~400 b/sThe normal translocation speed. A sharp drop means the pores are deteriorating.
- ~4 kbThe typical read-length peak for buccal cells, which reflects DNA fragment quality.
Choosing a Basecalling Model Fast, hac, and sup, in order of increasing accuracy and cost
| Model | Per-base accuracy (author's own figure) | Speed | When to use it |
|---|---|---|---|
| HAC | ~99% | Fast enough to run live | Throughout the run; Oxford Nanopore's own documentation recommends it for most users |
| SUP | ~99.5% | Slow — roughly ten times slower than HAC; about 31 hours on the author's NVIDIA machine against three, and over five days on the Mac | Re-run afterwards over regions of interest |
- minimap2 — align the Nanopore reads to GRCh38.
- samtools sort, index and flagstat — the author expects over 95% mapped.
- mosdepth — depth QC over the target region, confirming the adaptive sampling panel enriched as intended; the author's own rule of thumb is 5–6× enrichment on a 1% panel, with 1× meaning adaptive sampling silently was not active.
- The resulting
aligned.bamsupports variant calling, HLA phasing and pharmacogenomic typing. The author also describes an intention to run reads through DeepMind's AlphaGenome to predict the functional effect of non-coding variants — a research tool for prediction, not a diagnostic one. - For scale, on the author's own benchmark: one run produces about 49 GB of raw pod5 signal, 6 GB of BAM and 1 GB of logs, and on an NVIDIA GPU they measured HAC running about five times faster and SUP about four times faster than on an Apple Silicon Mac.
Conclusion
Who This Suits, and What It Cannot Do Three kinds of person, and three hard limits
Suited to
three kinds of people
- Anyone who wants to understand sequencing by doing it, rather than reading about it — the workflow's real strength is that every step's purpose and failure mode becomes concrete.
- Anyone with a specific, named question about a specific genomic region — adaptive sampling is genuinely good at exactly this, turning free targeted enrichment into 30–50× on a narrow panel with no custom probes.
- Researchers with bench experience who can already handle cold chain, QC and data analysis, and for whom the equipment and reagent lists above are familiar rather than novel.
Limits
three hard ones
- Cost, and where it lands is not fixed. A single run runs to roughly US$1,000 or more even on current list prices, and most reagents come in packaging sized for a lab running many samples a week, so a single run wastes most of a pack.
- Shallow coverage. One flow cell over a whole genome gives only about 10×, which is enough for common variants and well short of the 30× that population-scale efforts such as the UK Biobank use for confident calls.
- No DNA QC instrument in most home setups. Without a fluorometer to measure DNA concentration before committing it to library prep, a low yield and a failed prep look identical until the run itself fails.
Not a diagnostic procedure
A single home run's variant calls sit at roughly 10× coverage, well short of the 30× depth that population-scale efforts such as the UK Biobank use for confident calls — and a variant call, at any depth, is not the same thing as a diagnosis. Nothing on this page should be read as, or used as, a basis for a clinical decision. Anything that looks worrying belongs with a qualified clinician working from accredited, formal test data.
Understanding, not replacement
The workflow's genuine value, on the evidence here, is in understanding how sequencing actually works and in answering a specific, narrow question through adaptive sampling — not in replacing an accredited diagnostic test, and not in producing a comprehensive, confident account of rare variation. Both are honest uses of the same US$1,100 run; only one of them is what a home run can actually deliver.