08/05/2026
How do you benchmark sequencing accuracy without the benchmark itself being biased? 🤔 Reference genomes like GRCh38 were built predominantly from short-read SBS sequencing data. That means errors baked into the reference are invisible, and real biology can get mislabeled as "sequencer error”.
A new preprint from the Northwest Genomics Center and collaborators across NIH's SMaHT Network addressed this limitation by developing donor-specific assemblies (DSAs), diploid reference genomes built entirely from long-read + Hi-C data, with zero short-read influence on base calls. That gave them a technology-agnostic yardstick to compare the accuracy of nine short-read platforms.
What they found:
📊 After standard quality filtering, the Element AVITI™ showed significantly lower error rates, particularly with UltraQ™ chemistry which had the lowest error of any platform tested.
🔍 In low-complexity regions, where many high-value variants live, AVITI held the lowest SNV and indel error rate.
The full breakdown covers the methodology, the figures, and the comparison results. Read it: https://bit.ly/4xlbT6H.
