SeqDesk · revised predictions

How many human genomes would be sequenced by 2025?

In 2015, Stephens et al. projected 100 million to 2 billion human genomes by 2025. We rebuilt their figure from public archives and put the number that actually happened on it.

Stephens’ Figure 1, rebuilt on a log axis with our measured data laid over his. Following the rest of the tracker, solid = measured, dashed = forecast. The dashed vertical marks 2015, when the forecast was made; the shaded wedge is the range his three doubling rates projected to 2025. His headline was a head-count: 100M–2B human genomes (people). The solid black line is his own recorded growth (recreated from his figure); the two grey solid lines are ours, measured from the ENA/SRA archive. The darker grey line is the like-for-like — distinct people actually whole-genome-sequenced (~2.0M) — which fell 50–1000× short. The lighter grey line is the most generous reading: every genome-equivalent of human sequence data produced (~6.2M) — more than the number of people, because a 30× genome is ~30 genome-equivalents and much sequencing isn’t whole genomes — and even that is 16–322× short.

SeqDesk · revised predictions. Stephens' Figure 1 recreated from the published log-scale figure (values approximate); the three projection lines are computed from their 2015 baseline as N(2025) = N(2015) x 2^(120 / doublingMonths). The realized count reuses the curated 'Human genomes sequenced' tracker metric (a deliberately conservative ~2M lower bound; Berkeley Genomics). The paper's headline (100M-2B human genomes by 2025) is a head-count of people, so the like-for-like is our count of distinct individuals actually whole-genome-sequenced (~2M). Our curve runs below Stephens' recorded line because he counted genomes generously — his Figure 1 sits close to total sequencing capacity, not strict distinct individuals — so the gap peaks at ~100x around 2012 and narrows to ~4x by 2015. Early reconstruction anchors: HGP draft (2001), first individual genomes (Venter 2007; Watson and Wang/YH 2008), 1000 Genomes pilot (179 individuals, 2010) and Phase 1 (1,092, 2012). The discrepancy is reported against the paper's stated 100M-2B range; treat any straight-line extrapolation on this page with the same caution. As the MOST GENEROUS possible reading we also count every genome-equivalent of human sequence data produced (total bases / one genome) from the ENA/SRA read archive — this exceeds the number of distinct people (a 30x genome is ~30 genome-equivalents, and much sequencing isn't whole genomes), yet still reaches only roughly 6 million by 2025, ~16-322x under the projection. That data-volume measure corresponds to Figure 1's other axis (sequencing capacity); it counts only data submitted to the public archive, so it too is a lower bound.

• Three quantities share one axis. The dark line is Stephens' own (generous) count of human genomes; copper is our strict count of distinct PEOPLE whole-genome-sequenced; teal is genome-equivalents of sequencing DATA (total bases / one genome). The 100M-2B headline is a head-count, so copper is the like-for-like. Teal sits above copper because a 30x genome is ~30 genome-equivalents and much human sequencing is exome, transcriptome, amplicon or single-cell rather than whole genomes — so reading data-equivalents as a count of people is a category error.
• The capacity curve is a lower bound on data produced. It sums only human reads submitted to the open ENA/SRA read archive. A large and growing share of clinical and biobank sequencing sits in controlled-access repositories (EGA, dbGaP) or never leaves a hospital/company, so it is not counted. Stephens measured data PRODUCED; we can only measure data SUBMITTED, which is smaller. first_public also dates archive release, not when the sample was sequenced.
• Genome-equivalents are a volume proxy, not a genome count. We take total human bases (ENA tax_eq(9606), all read types) divided by one 3.2 Gb genome. That is the right unit for Stephens' data-volume comparison, but it is not a number of sequenced individuals: redundant resequencing, deep coverage and non-WGS assays all inflate bases per person. Choosing a different divisor (e.g. a 30x genome) would shift the curve by that factor.
• Stephens' Figure 1 is recreated, not the original data. His historical points and the three projection lines are read off the published log-scale figure (values approximate) and computed from a single 2015 baseline. His 100M-2B headline is a stated range, not a point estimate, and the paper's 'genomes' unit is itself ambiguous between sequencing capacity and a count of people.
• The ~2M genome count is conservative and ultimately unknowable. The distinct-genome curve reuses a deliberately conservative ~2M lower bound (Berkeley Genomics). Component cohorts (UK Biobank, All of Us, gnomAD, Genomics England) must NOT be summed — severe double-counting — the true figure is likely 2-5M, and because most clinical genomes never become public it can't be counted exactly. Early-year anchors are read off milestones, so the curve is sparse before 2015.
• How we read it. The 2015 forecast overshot either way — ~16-322x in data terms, ~50-1000x in distinct people — so the headline number was too optimistic on both measures. But the paper's core thesis held: genomics did become an exabyte-scale field. The miss was the timeline and the human-genome count, held back by cost, clinical value, consent and reimbursement rather than sequencer throughput. Treat every extrapolation on this page as a trend, not a destiny — that caution is the whole point of revisiting it.

• Stephens et al. 2015, PLOS Biology — Big Data: Astronomical or Genomical? (100M-2B by 2025 projection, Figure 1)
• Berkeley Genomics — “How many human genomes have been sequenced?” (~2M lower bound)
• UK Biobank WGS, Nature 2025 — 490,640 genomes at 32.5x
• gnomAD v4 — 76,215 WGS + 730,947 WES
• 1000 Genomes Project Phase 1 — 1,092 genomes
• 1000 Genomes Project pilot — 179 individuals (Nature 2010)
• First individual genomes — Venter (Levy et al. 2007); Watson (Wheeler et al. 2008)

The first of our revised predictions: rebuild an influential, ageing forecast from public archives and check it against what actually happened. Only aggregate counts are published. · back to all research data