The Best DNA Sequencer for Sequencing Yourself

If you are sequencing a single genome, your own, and you intend to work with the data yourself, we would choose Oxford Nanopore: R10.4.1 flow cells, Dorado super-accuracy basecalling, and roughly 30x coverage on a PromethION. Illumina still wins on raw per-base accuracy, and we will come back to that. The reason to reach for long reads anyway is that a single long-read run delivers four things at once that short reads give you only partially or not at all: structural variants, repeat expansions, long-range phasing into haplotypes, and 5mC methylation read from the same molecules, with no bisulfite step and no additional assay.

There is a reasonable alternative. If what you want is the cheapest possible callset of single-nucleotide variants (SNVs, single-base changes) with the deepest benchmarking history behind it, buy an Illumina 30x whole genome and stop reading. That is a defensible choice. It is also a genome that is missing a specific and enumerable set of features, and once you have looked closely at what is missing, most people do not want to go back.

Why long reads win for a personal genome

To understand where short reads fall short, it helps to recall how they work. Short-read sequencing fragments DNA into pieces of 150–300 base pairs, sequences both ends of each piece, and maps those fragments back to a reference genome ¹. That approach works well when the variant you are looking for is a single base substitution sitting in unique sequence. It degrades badly when the variant is larger than a single read, or when the surrounding sequence is repetitive enough that a read maps to several locations equally well.

The consequences fall into four categories, each of which affects what you can learn about your own genome.

Structural variants. Deletions, duplications, inversions, and translocations in the kilobase-to-megabase range are systematically under-called by short reads, which must infer them indirectly from discordant read-pair orientations and dips in coverage. Long reads span the breakpoints directly. In a study of four individuals with autism spectrum disorder, long-read sequencing resolved structural variants whose internal architecture was not recoverable from short-read data ². That architecture included nested duplications, inverted segments, and the actual order of events. That level of resolution matters when a clinical report tells you there is a “duplication at chr15q11” and gives you no sense of how it is arranged.

Repeat expansions and imprinting. Expansions in short tandem repeats are invisible to reads shorter than the repeat itself. Long reads also carry methylation state alongside sequence, which makes it possible to separate mechanisms that look identical at the level of sequence alone. Nanopore sequencing has been used to detect and subtype Prader-Willi and Angelman syndromes. It distinguishes deletion, uniparental disomy, and imprinting defects in a single assay ³. Workflows now exist to profile the human repeatome, sequence and methylation together, directly from nanopore reads ⁴.

Phasing. A short read can tell you that you carry two variants in the same gene, but it usually cannot tell you whether they sit on the same chromosome copy (cis) or on opposite copies (trans). A long read carries many heterozygous sites on a single molecule, so haplotype blocks extend for megabases. A haplotype block is the stretch over which you know which variant travels with which. Allelic configuration is an interpretable layer in its own right: long-read haplotype phasing resolves the cis and trans relationships that change the functional reading of paired variants ⁵. For a personal genome, phasing is what turns “two variants in one gene” into a statement about whether any intact copy of that gene remains.

Methylation at no extra cost. The raw nanopore signal encodes base modifications as well as bases. Dorado calls 5mC and 5hmC in CpG context in the same pass as the basecalls. You obtain a genome-wide methylome without a separate bisulfite library and without the DNA damage that bisulfite treatment causes, already phased by haplotype ⁶.

Where nanopore is worse, plainly

A fair comparison has to include the places where nanopore loses, and there are several worth stating clearly before you commit money to a run.

Residual error is concentrated in homopolymers and small insertions and deletions. R10.4.1 chemistry with super-accuracy basecalling brings simplex modal accuracy into the Q20+ range. Duplex basecalling, in which both strands of the same molecule are called together, pushes higher still. Even so, you will see more 1–2 bp indel false positives in long homopolymer runs than Illumina produces. If your question is whether you carry one specific single-base pathogenic variant in a well-characterized gene, short reads or targeted Sanger confirmation are the more conservative answer. Sanger remains the confirmation method for precisely this reason, offering high per-read accuracy over long single amplicons and a fifty-year track record, at the cost of handling one amplicon at a time ⁷.

Cost is the second consideration. A 30x nanopore genome on a PromethION flow cell is not cheaper than a 30x Illumina genome at current pricing, and it becomes considerably more expensive if you run a flow cell poorly and have to repeat it. Illumina’s advantage comes from manufacturing scale and the depth of short-read benchmarking, which is why it dominates population-scale projects ¹⁸.

Compute is the third. Dorado super-accuracy basecalling is GPU work, so budget an A100 or better, or plan on many hours on a consumer card. Basecalling a 30x human genome at sup is not a laptop job. We recommend keeping the POD5 raw signal files if you have the storage, roughly 1–2 TB for a human genome, because they are what allows you to re-basecall with an improved model in two years without re-extracting DNA.

Sample quality is the fourth, and it is the one people underestimate. Nanopore read length is bounded by your DNA, not by the instrument. Blood drawn into EDTA and extracted with a high-molecular-weight protocol (Circulomics/Nanobind or similar) yields read-length N50s in the tens of kilobases, where N50 is the length such that half the sequenced bases sit in reads at least that long. A cheek swab, or a badly vortexed extraction, gives you 5 kb reads. Most of the long-read advantage then evaporates.

What to ask a provider for

Providers vary in what they consider a standard run, so it pays to be specific. Ask for the following, and get them in writing before you pay:

• R10.4.1 chemistry (PromethION or MinION flow cells), ligation kit SQK-LSK114 or the current equivalent.
• At least 30x aligned coverage of GRCh38 or, better, T2T-CHM13v2.0. Ask for the figure measured after alignment and filtering, not the raw gigabase yield.
• Read N50 of 20 kb or greater. Push back if they quote mean read length instead.
• Basecalling with Dorado sup and the 5mCG_5hmCG modified-base model.
• Deliverables: POD5 raw signal, unaligned BAM with MM/ML modification tags preserved, aligned BAM and the caller VCFs. A BAM is the compressed binary file of aligned reads; a VCF is the tabular list of variant calls. Insist that the modification tags survive alignment, since this is the most common thing to lose.
• Duplex only if the provider can hit meaningful duplex rates. Otherwise it costs yield and buys little at 30x.

The pipeline we would run

Once the data arrive, the following sequence takes you from raw signal to phased variants and haplotype-resolved methylation.

# basecall with modifications, emit unaligned BAM with MM/ML tags
dorado basecaller sup@v5.0.0 pod5/ \
  --modified-bases 5mCG_5hmCG \
  --device cuda:all > calls.ubam

# align, keeping tags (-y carries them through)
dorado aligner -t 32 GRCh38.mmi calls.ubam | \
  samtools sort -@ 8 -o aln.bam - && samtools index aln.bam
# equivalently: minimap2 -ax map-ont -y --MD -t 32 ref.fa reads.fq

# small variants
run_clair3.sh --bam_fn=aln.bam --ref_fn=GRCh38.fa \
  --platform=ont --model_path=/opt/models/r1041_e82_400bps_sup_v500 \
  --threads=32 --output=clair3_out

# structural variants
sniffles --input aln.bam --vcf sv.vcf.gz --reference GRCh38.fa \
  --minsvlen 50 --output-rnames

# phase small variants and haplotag the BAM
longphase phase -s clair3_out/merge_output.vcf.gz -b aln.bam \
  -r GRCh38.fa -o phased --ont
whatshap haplotag --reference GRCh38.fa --output-haplotag-list htags.tsv \
  -o aln.haplotagged.bam phased.vcf aln.bam

# methylation, split by haplotype
modkit pileup aln.haplotagged.bam methyl_by_haplotype/ \
  --ref GRCh38.fa --cpg --combine-strands --partition-tag HP

A few words on why each tool was chosen. We prefer Clair3 to DeepVariant for ONT data mostly because the ONT-specific models track chemistry releases faster. Sniffles2 beats cuteSV for structural variant calling here because its genotyping is better behaved on a single sample. LongPhase is faster than WhatsHap for whole-genome phasing and can use structural variants as phasing evidence. We still run whatshap haplotag because its haplotag output plugs cleanly into modkit’s --partition-tag HP, which yields allele-specific methylation genome-wide. That combination is the one thing you cannot get any other way for the price, and it is where imprinted loci light up immediately.

Before believing any of the output, run a few sanity checks. Use samtools stats for coverage uniformity, mosdepth --by 1000 to find dropout regions, and whatshap stats --gtf to see your phase block N50. If the phase block N50 comes in under a megabase, your read N50 was too short.

One point is not negotiable. A variant in your own genome is not a diagnosis. Long-read callsets contain plenty of novel structural variants of unknown significance, and the error modes described above generate false positives that can look alarming. Anything you would act on belongs in front of a clinical geneticist, confirmed in a CLIA-certified lab, usually by an orthogonal method.

Questions people also ask

Which company is considered the best for DNA sequencing? Illumina leads by volume and by benchmarking depth, and it is the right default for large short-read projects ¹⁸. For a single personal genome that you intend to interrogate yourself, Oxford Nanopore gives more information per run. Element Biosciences and PacBio are real alternatives: Element competes with Illumina on short-read accuracy and price, while PacBio HiFi beats nanopore on per-base accuracy, with shorter reads and higher cost.

Why is Illumina better than Sanger? Throughput. Sanger sequencing reads one 500–1000 bp amplicon at a time, whereas massively parallel sequencing reads hundreds of millions of fragments in a single run, which is what made whole genomes economically possible ⁷⁹. Sanger remains the confirmation standard for a single known variant.

Is Sanger sequencing 100% accurate? No. It is very accurate in the clean middle of a trace and unreliable in the first ~30 bases, in long homopolymers, and through heterozygous indels, which produce overlapping traces that require deconvolution.

What is the most accurate DNA sequencing method? For per-base substitution accuracy on short fragments, Illumina and Element. For long-read per-base accuracy, PacBio HiFi. For completeness of a genome, meaning structural variants, repeats, phasing, and methylation, long reads lead, and the accuracy question changes shape, because a perfectly called base in a region you could not assemble is not much use ¹⁰⁵.

Who are Illumina’s main competitors? Element Biosciences, Oxford Nanopore, PacBio, MGI/Complete Genomics, and Ultima. The short-read price floor has been falling because of that competition.

Woolf Software builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

1. Rachel L Goldfeder, Dennis P Wall, Muin J Khoury, et al. Human Genome Sequencing at the Population Scale: A Primer on High-Throughput DNA Sequencing and Analysis. American Journal of Epidemiology, 2017. https://doi.org/10.1093/aje/kww224 ↩ ↩² ↩³

2. Sarah Dada, Katherine Dixon, Vahid Akbari, et al. Uncovering the complexity of structural variants in four individuals with autism spectrum disorder. Genome, 2025. https://doi.org/10.1139/gen-2024-0121 ↩

3. Vahid Akbari, Sarah Dada, Yaoqing Shen, et al. Long-read sequencing for detection and subtyping of Prader-Willi and Angelman syndromes. Journal of Medical Genetics, 2024. https://doi.org/10.1136/jmg-2024-110115 ↩

4. Brando Poggiali, Leena Putzeys, Jeppe Dyrberg Andersen, et al. ECHO: a nanopore sequencing-based workflow for (epi)genetic profiling of the human repeatome. Bioinformatics, 2026. https://doi.org/10.1093/bioinformatics/btag648 ↩

5. Josh N. Vo, Yi-Mi Wu, Rui Wang, et al. Long-Read Haplotype Phasing Resolves Allelic Configuration as a Missing Layer of Precision Oncology. Cancer Discovery, 2026. https://doi.org/10.1158/2159-8290.cd-26-0839 ↩ ↩²

6. Clarissa F. de Carvalho, James Ord, Paula Escuer, et al. Long-read sequencing for epigenomic studies in the fields of ecology and evolution. Philosophical Transactions of the Royal Society B: Biological Sciences, 2026. https://doi.org/10.1098/rstb.2025.0099 ↩

7. Kübra Eren, Nursema Taktakoğlu, Ibrahim Pirim. DNA Sequencing Methods: From Past to Present. The Eurasian Journal of Medicine, 2023. https://doi.org/10.5152/eurasianjmed.2022.22280 ↩ ↩²

8. Sang Tae Park, Jayoung Kim. Trends in Next-Generation Sequencing and a New Era for Whole Genome Sequencing. International Neurourology Journal, 2016. https://doi.org/10.5213/inj.1632742.371 ↩ ↩²

9. Jason M. Rizzo, Michael J. Buck. Key Principles and Clinical Applications of “Next-Generation” DNA Sequencing. Cancer Prevention Research, 2012. https://doi.org/10.1158/1940-6207.capr-11-0432 ↩

10. Bioinformatics for human long-read whole genome sequencing. Nature Reviews Methods Primers, 2026. https://doi.org/10.1038/s43586-026-00528-w ↩