# Genome sequencing cost analyses

ProteinIQ analyses supporting [How much does it cost to sequence a genome?](https://www.proteiniq.io/guides/genome-sequencing-cost), completed September 24, 2026. Run `python analyze.py` in this directory (Python 3.13, pandas, numpy, xlrd) to regenerate `summary.json`, `nhgri-halving-times.csv` and `core-facility-results.csv`.

## 1. Halving time of NHGRI sequencing costs

Input: `raw/Sequencing_Cost_Data_Table_May2022.xls`, the National Human Genome Research Institute's cost table (79 dated values, September 2001 to May 2022), downloaded on September 24, 2026 from https://www.genome.gov/about-genomics/fact-sheets/DNA-Sequencing-Costs-Data.

Method: we fitted an ordinary least-squares line to log2(cost per megabase) against time within four eras and report the halving time (negative reciprocal of the slope, in months). Era boundaries are January 2008, when NHGRI's data switch to next-generation platforms, and the ends of 2011 and 2015, where the rate of decline visibly changed. The halving times depend on these boundaries and should be read as era averages.

We fitted cost per megabase rather than cost per genome because NHGRI's genome definition changes: dividing its cost per genome by its cost per megabase gives 18,000 Mb (about 6x coverage of a 3,000 Mb genome) before 2008 and 90,000 Mb (30x) afterwards. The published per-genome series therefore overstates the fall by a factor of five across the 2008 transition, and the per-megabase decline (about 908,000-fold) is larger than the per-genome decline (about 182,000-fold).

Inflation: `raw/bls-CUUR0000SA0-*.json` hold the monthly US CPI-U (series CUUR0000SA0, not seasonally adjusted) from the BLS public API, retrieved September 24, 2026. Each NHGRI value is scaled by CPI for August 2026 divided by CPI for the value's month. BLS did not publish some months; none coincides with an NHGRI date.

## 2. Survey of US university core facility prices

Input: `core-facility-prices.csv`, hand-entered from 17 public price lists found by web search and read on September 24, 2026. The file records each URL, the SHA-256 of the page or PDF we read, the stated price-list date, the option used and notes. Facilities were included when they publish either Illumina NovaSeq X 2x150 sequencing prices or a whole-genome library price. It is a convenience sample, not a random or complete census.

Rules:

- A 30x human genome is 90 Gb of sequence, NHGRI's definition, which equals 300 million 2x150 read pairs. The summary also reports sequencing at 100 Gb as a sensitivity check.
- For sequencing, we used the cheapest lane-level 2x150 option one laboratory can buy (usually a 25B lane; a 10B lane where no 25B lane is listed; a full flow cell where no lane is listed; Case Western's shared 1.25B-read option). Yield is the facility's stated read or data count, using the midpoint of a stated range, or Illumina's nominal rating (25B flow cell, 3.125 billion pairs per lane) when only the flow-cell type is named.
- The University of New Hampshire sells shared-run sequencing per million read pairs and quotes dedicated flow cells on request. We report it separately and exclude it from lane-level statistics.
- Library price is the standard whole-genome DNA library at the smallest batch tier. Internal rates are each university's own users; external rates are the external academic or non-profit tier.
- Totals add library and sequencing only. They exclude DNA extraction, sample QC beyond what a library price includes, bioinformatics, storage and shipping.

Exclusions: TGen (page blocked by a bot check), NC State (NovaSeq X by quote), Michigan (page returned 403), and facilities whose prices are available only on request. The University of Iowa contributes a library price only. The University of Minnesota's all-in Ultima UG 100 price and the Toronto Centre for Applied Genomics' all-in price (CAD 650) are discussed in the article but not included in the Illumina-based statistics.

Limitations: list prices change and may differ from negotiated quotes; stated yields are conservative at some facilities; four facilities send NovaSeq X runs to external providers; the sample is US-centric and self-selected by what is published online.