Sources

Utility of long-read sequencing for All of Us

Generated 2026-04-12 07:16 UTC

Metadata

title
Utility of long-read sequencing for All of Us
kind
paper
status
ingested
added
2026-04-07T09:46:12+09:00
raw source
raw/sources/mahmoud_2024_utility_of_long-read_sequencing_for.pdf
deep ingested
2026-04-07

Source

Study design

  • Samples: HapMap samples and All of Us control samples spanning eight datasets
  • Platform: comparison across short-read and long-read technologies, including PacBio HiFi and ONT, with scalable cloud pipelines
  • Aim: determine which long-read approaches are most useful for medically relevant genes, small variants, and SVs in a large-cohort program

Summary

  • This paper asks a deployment question: if a program the size of All of Us considers long reads, what do they buy?
  • The answer is that long reads materially improve coverage and variant recovery in medically relevant, technically challenging genes, and that HiFi gives the strongest overall variant-calling accuracy in this pilot.
  • The paper also contributes operational infrastructure by releasing a cloud-based long-read analysis pipeline.

Key findings

  • HiFi produced the most accurate results in the study for both small and large variants.
  • Long reads improved sequencing and variant calling in challenging medically relevant genes, including complex loci where short reads remain weaker.
  • The authors show that machine-learning-based callers and merged calling strategies can improve long-read small-variant performance.
  • The paper frames low-coverage long-read sequencing as a potential strategy for scaling sample counts, but treats that as a tradeoff rather than a solved recipe.

Strengths

  • Strong translational relevance to biobank-scale decision-making.
  • Focuses on clinically important gene sets rather than only genome-wide aggregate metrics.
  • Provides practical pipeline infrastructure instead of only benchmark figures.

Limitations and caveats

  • Pilot scale: the paper does not settle how long-read sequencing should be rolled out across a cohort of one million people.
  • Technology conclusions are sensitive to rapidly changing chemistry and caller performance.
  • The paper is richer on comparative performance than on end-to-end clinical outcome.

Relevance to this corpus

  • This paper sharpens the corpus-wide platform question.
  • It supports broad value for long reads, but its strongest claims are in hard medically relevant regions rather than in every part of the genome equally.

Open questions

  • At what coverage and cost point do long reads become justified for national-scale screening programs?
  • How durable is the HiFi-over-ONT advantage as ONT chemistry and duplex workflows improve?