Whole Genome Mapping: What the Term Means and What It Costs

If you are asking what it costs to get your genome mapped, the short answer is $300 to $1,000 for a 30x short-read whole genome. That price is from a consumer or research provider that returns raw FASTQ. A clinical-grade genome ordered through a physician with a signed report costs a few thousand dollars. Optical genome mapping is a different assay entirely, and it costs $500 to $1,500 more if you separately want it. The word “mapping” is doing a lot of work in that question, and which of its four meanings you want changes both the price and the file you end up with.

The four things called genome mapping

Linkage and physical maps are the historical sense. Before cheap sequencing, you located a gene by tracking recombination between markers across families, or by ordering overlapping BAC clones along a chromosome. The Human Genome Project produced a physical map before it produced a sequence. The first genome-wide variation map of this kind gave 1.42 million SNPs with roughly one marker every two kilobases of the accessible genome ¹. Nobody orders this for themselves today.

Read mapping (alignment) is what a bioinformatician means. Your sequencer produces short reads. Mapping is the step that assigns each read a coordinate on a reference genome. This is the step that turns FASTQ into BAM or CRAM. If you buy a whole genome and get raw data back, this is the compute you will run or pay someone to run.

Optical genome mapping (OGM) is a physical assay. It is not sequencing. Ultra-high-molecular-weight DNA (megabase-scale molecules) is labeled at a specific 6-mer motif, linearized in nanochannels, and imaged. You get a pattern of label distances per molecule, assembled into consensus maps and compared against a reference map. It resolves large structural variants, balanced translocations, and inversions that short reads miss. The assay does not read bases.

“Getting your genome mapped” in ordinary speech means whole-genome sequencing. That is what most people asking about cost want.

What you get for $300 to $1,000

A 30x short-read human genome. Concretely: two gzipped FASTQ files totaling 60 to 100 GB, about 800 million to 1 billion paired 150 bp reads, mapping to roughly 90 GB of BAM or 15 to 25 GB of CRAM after alignment. Variant calling yields around 4 to 5 million small variants against GRCh38, of which roughly 3.5 million are SNVs and 500,000 to 800,000 are indels, plus 10,000 to 25,000 structural variant calls depending on caller and thresholds.

The price curve that got us here was forecast almost exactly. The $1,000 genome was a stated engineering target in 2005, when a genome cost on the order of $10 million ². It arrived, and the consequence is that the bottleneck moved from generating sequence to interpreting it, which is where you will spend your time ³.

Three things to check before you buy:

• Do you get FASTQ, or only a VCF? A VCF from a vendor’s pipeline is a summary. If you ever want to recall variants with a different tool, call structural variants, estimate mitochondrial heteroplasmy, or check coverage over a specific gene, you need the reads.
• What is the actual mean coverage, and what fraction of the genome is covered at 10x or more? 30x nominal often lands at 27x delivered. Run mosdepth -n --fast-mode --by 1000 sample your.cram and look at the distribution. Do not stop at the mean.
• Short read or long read? PacBio HiFi or Oxford Nanopore at 15 to 30x costs several times more but resolves repeats, phases variants across tens of kilobases, and finds far more structural variation.

The pipeline we would run

Start from FASTQ. Check it first: fastp -i R1.fq.gz -I R2.fq.gz --json fastp.json for adapter content, duplication estimate, and per-base quality.

Align with bwa-mem2:

bwa-mem2 mem -t 32 -K 100000000 -Y \
  -R '@RG\tID:L1\tSM:you\tPL:ILLUMINA\tLB:lib1' \
  GRCh38_full_analysis_set_plus_decoy_hla.fa R1.fq.gz R2.fq.gz \
| samtools sort -@ 8 -m 2G -o you.bam -

-K 100000000 fixes the input chunk size so the output is deterministic regardless of thread count. -Y uses soft clipping for supplementary alignments, which structural variant callers want. Mark duplicates with samtools markdup or Picard, then convert to CRAM with samtools view -T ref.fa -C --output-fmt-option archive. CRAM with the reference specified is the format to keep long term.

For small variants we would use DeepVariant over GATK HaplotypeCaller for a single sample: run_deepvariant --model_type=WGS --num_shards=32. It is a single step, it needs no base quality recalibration, and it is more accurate on indels in homopolymer runs. GATK still wins if you want joint calling across a family or cohort, where gVCFs and GenotypeGVCFs have no equivalent.

Then sanity-check before you believe anything. verifyBamID2 or somalier for contamination and sample swaps: cross-sample contamination above 1 to 2% will produce false heterozygous calls across the genome. bcftools stats for the transition/transversion ratio, which should sit near 2.0 to 2.1 genome-wide and closer to 3.0 in coding regions. If Ti/Tv is 1.6, something in the pipeline is wrong.

GRCh38 or T2T-CHM13

Map to GRCh38 as the primary, because every annotation resource, allele frequency database, and clinical variant table is keyed to it. Use the analysis set with decoy and HLA contigs, and make sure your aligner handles alt contigs correctly (bwakit’s postalt script, or a pipeline configured for alt-awareness). Mapping to a version of GRCh38 without decoys pushes reads from unplaced repeats onto real chromosomes and manufactures false variants.

Then, if you care about the hard regions, map a second time to T2T-CHM13v2.0. The complete assembly adds centromeric satellite, acrocentric short arms, and a number of segmental duplications and medically relevant genes that were collapsed or missing in GRCh38. Compare the two, do not replace one with the other.

Where mapping fails, and how you see it

Short-read mapping has three systematic failure modes, and all three are visible in your own BAM.

MAPQ 0 pileups: in segmental duplications and paralogous gene pairs, reads map equally well to two places, the aligner assigns mapping quality 0, and variant callers throw them away. samtools view -c -q 1 versus total count gives you the fraction. Genes with near-identical pseudogenes are the usual casualties.

Coverage dropout in high-GC promoters: PCR-based library preps deplete GC-rich first exons. PCR-free libraries are the fix and are worth asking about.

Structural variants below the detection floor: a short-read caller such as Manta or smoove infers SVs from discordant pairs and split reads. It does well on deletions of a few hundred bases to a few kilobases and poorly on insertions, inversions with repetitive breakpoints, and anything inside a repeat array. Long reads with Sniffles2 or optical mapping cover that gap. This is the case for OGM: it sees balanced events and megabase-scale rearrangements that no short-read pipeline will report.

Getting from a VCF to meaning

Annotate with VEP or SnpEff against a fixed Ensembl release, join to gnomAD allele frequencies, and filter. Most of the 4 to 5 million variants are common and uninformative. Coding variants that are rare and predicted damaging are a manageable list, usually a few hundred. Noncoding variants are the harder problem: the ENCODE pilot found biochemical activity across a large fraction of the genome, including transcription and regulatory signal far outside protein-coding exons, which means a variant outside a gene is not automatically inert and also not easily interpretable ⁴. Population-scale sequencing has made the statistical methods better, but the interpretation gap remains the limiting step. The data is not what holds things back ³.

One boundary to keep. A research or consumer genome is a measurement. It is not a medical result. Variants in clinically actionable genes need orthogonal confirmation in a CLIA laboratory and interpretation by a clinical geneticist or genetic counselor before they mean anything about your health. Read your data, build your own annotations, and take anything that looks consequential to a clinician.

Questions people also ask

How much does it cost to get a genome mapped? $300 to $1,000 for 30x short-read WGS with raw FASTQ returned, several thousand for a clinical genome with a signed report, and a few thousand for long-read sequencing. Optical genome mapping is priced separately and typically runs $500 to $1,500.

Is genome mapping the same as genome sequencing? No. Sequencing reads bases. Mapping either means aligning those reads to a reference, or a separate imaging assay (optical genome mapping) that measures long-range structure without reading bases at all. In casual use, “get my genome mapped” means sequencing.

What is optical genome mapping used for? Detecting large structural variants: deletions, duplications, inversions, translocations, and repeat expansions above roughly 500 bp to 5 kb, in a single assay. It is used clinically in hematologic malignancy and constitutional cytogenetics as an alternative to karyotyping and FISH. It does not give you SNPs.

How much compute do I need to map my own genome? A 32-core machine with 64 GB of RAM aligns a 30x genome in about 3 to 6 hours with bwa-mem2. DeepVariant adds a few hours on CPU, far less with a GPU. Budget 400 GB of scratch disk, and keep the final CRAM plus the reference FASTA.

How many variants will I have? Roughly 4 to 5 million small variants against GRCh38, about 10,000 to 20,000 structural variants with short reads and 20,000 to 25,000 with long reads, and on the order of tens of thousands of rare coding variants of which a small number have any documented interpretation.

Footnotes

1. The International SNP Map Working Group, Cold Spring Harbor Laboratories:, Ravi Sachidanandam, et al. A map of human genome sequence variation containing 1.42 million single nucleotide polymorphisms. Nature, 2001. https://doi.org/10.1038/35057149 ↩

2. Simon T. Bennett, C.L. Barnes, Anthony J. Cox, et al. Toward the $1000 Human Genome. Pharmacogenomics, 2005. https://doi.org/10.1517/14622416.6.4.373 ↩

3. Tuuli Lappalainen, Alexandra J. Scott, Margot Brandt, et al. Genomic Analysis in the Age of Human Genome Sequencing. Cell, 2019. https://doi.org/10.1016/j.cell.2019.02.032 ↩ ↩²

4. Ewan Birney, Paul Flicek, Damian Keefe, et al. Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature, 2007. https://doi.org/10.1038/nature05874 ↩