Answers are grounded in this paper and cite their sources.
DNA data storage is judged by bits stored per synthesized nucleotide, yet it is rarely known where an experiment loses the bits it does not store. Here we fit position-dependent insertion–deletion–substitution channels to the raw reads of 17 public datasets, prove capacity bounds for them and measure information rates on the reads. A model-based accounting indicates that sampling dominates the losses of low-error array pools read by Illumina: at low sequencing depth, unread strands, uneven coverage and indexing cost 0.2–0.35 bits per nucleotide, against less than 0.01 for sequence errors. Error-prone chemistries leave 0.04–0.26 bits between genie information and achieved rates, and designs that forbid repeated bases cap density at 1.59 bits per nucleotide. In a prespecified, paired test on fresh files, an iterative soft receiver with sparse pilot bases needed 7–42% fewer reads than Trellis BMA, the best published method, on eight of nine datasets. The released software places any dataset against its own losses.
No takes yet. Share an insight, caveat, or question.
Khodaiemehr et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: