Tachyon accelerates RNA-seq alignment while preserving STAR outputs in a multispecies benchmark

STAR is a widely used RNA-seq aligner, and a laboratory that changes its aligner or even its aligner version may have to validate its pipeline again. I developed Tachyon from STAR 2.7.11b to shorten alignment time without changing the alignments, so I kept a change only when Tachyon produced the same alignment records as STAR on the same input. The work took ten days, from 20 to 29 October 2026. I timed both programs on whole public RNA-seq samples from six organisms with the genome index already in memory. On a machine with eight physical AMD cores Tachyon 0.4.1 was 2.53 to 5.70 times as fast as STAR, and the mean of the six ratios was 3.95. On sixteen physical AMD cores five of the samples gave ratios of 2.72 to 5.86, and on eight physical Intel cores three of them gave 2.50 to 4.51. A separate CPU profile of 500,000 read pairs from each organism showed that most of the CPU time was saved in the step that joins seeds into alignments. Over the six organisms this step fell from 1,075 to 268 CPU seconds. A whole human sample of 51.7 million read pairs ran on sixteen physical AMD cores, and there Tachyon was 3.16 times as fast as STAR with 8 threads and 3.20 times as fast with 16. In the same campaign I compared 190 Tachyon runs with STAR. They came from the release and from experimental builds of the same source, and all of them matched STAR in alignment records, splice junctions, gene counts and mapping statistics. An earlier development benchmark of 21 workloads gave ratios of 1.78 to 7.69 with matching output fingerprints. In that benchmark, however, Tachyon 0.4 also used a precomputed lookup table and loaded the index in a different way. I ran all the tests myself.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-12
DOI
https://doi.org/10.5281/zenodo.24155617
Primary Topic
Genomics and Phylogenetic Studies
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Tachyon accelerates RNA-seq alignment while preserving STAR outputs in a multispecies benchmark

Antonello Colangelo
Zenodo (CERN European Organization for Nuclear Research)
Genomics and Phylogenetic Studies
preprint

Tachyon accelerates RNA-seq alignment while preserving STAR outputs in a multispecies benchmark

Antonello Colangelo
preprint en

Abstract

STAR is a widely used RNA-seq aligner, and a laboratory that changes its aligner or even its aligner version may have to validate its pipeline again. I developed Tachyon from STAR 2.7.11b to shorten alignment time without changing the alignments, so I kept a change only when Tachyon produced the same alignment records as STAR on the same input. The work took ten days, from 20 to 29 October 2026. I timed both programs on whole public RNA-seq samples from six organisms with the genome index already in memory. On a machine with eight physical AMD cores Tachyon 0.4.1 was 2.53 to 5.70 times as fast as STAR, and the mean of the six ratios was 3.95. On sixteen physical AMD cores five of the samples gave ratios of 2.72 to 5.86, and on eight physical Intel cores three of them gave 2.50 to 4.51. A separate CPU profile of 500,000 read pairs from each organism showed that most of the CPU time was saved in the step that joins seeds into alignments. Over the six organisms this step fell from 1,075 to 268 CPU seconds. A whole human sample of 51.7 million read pairs ran on sixteen physical AMD cores, and there Tachyon was 3.16 times as fast as STAR with 8 threads and 3.20 times as fast with 16. In the same campaign I compared 190 Tachyon runs with STAR. They came from the release and from experimental builds of the same source, and all of them matched STAR in alignment records, splice junctions, gene counts and mapping statistics. An earlier development benchmark of 21 workloads gave ratios of 1.78 to 7.69 with matching output fingerprints. In that benchmark, however, Tachyon 0.4 also used a precomputed lookup table and loaded the index in a different way. I ran all the tests myself.

Zenodo (CERN European Organization for Nuclear Research)
Genomics and Phylogenetic Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon