Caltech researchers have developed a DNA assembly method that achieves one-in-a-million accuracy - four to five orders of magnitude better than existing techniques. Called Sidewinder, the approach solves a fundamental problem that has limited synthetic biology: reliably stitching together short DNA fragments into gene-length or even genome-length sequences.
How Sidewinder Works
Current DNA synthesis can only produce short oligonucleotides of 10-100 base pairs reliably. Building anything longer requires assembling these fragments, but previous methods suffered error rates of 1-in-10 to 1-in-30 misconnections. At those rates, assembling a gene (thousands of nucleotides) or genome (millions) becomes impractical.
Sidewinder solves this by adding “page numbers” to DNA fragments. The method uses three-way junctions (3WJ) that attach numbered tags to each oligonucleotide. These tags guide individual DNA pieces to assemble in the correct sequence, then are seamlessly removed afterward, leaving a continuous double helix.
The key innovation is separating sequence information from assembly information - the first DNA construction method to do so. Previous approaches relied on matching sequence endpoints, which became increasingly unreliable as assemblies grew larger.
Lead researcher Kaihang Wang, assistant professor of biology and biological engineering at Caltech, describes the significance: “DNA is the source code of all earthly life and biological functions.” Sidewinder enables “writing DNA faster, more easily, and cheaper than previously possible.”
What the Numbers Mean
The jump from 1-in-30 to 1-in-1,000,000 misconnection rates isn’t incremental improvement - it’s a qualitative change in what’s buildable.
Consider assembling a 10,000 nucleotide gene from 100 fragments. At 1-in-30 error rates, you’d expect roughly 3 assembly errors per attempt, making successful construction extremely unlikely. At 1-in-1,000,000, you’d need to run 10,000 assemblies before expecting a single error. This makes reliable construction of genes, gene clusters, and potentially entire genomes practical for the first time.
Frances Arnold, Nobel laureate in chemistry, assessed the impact: “Sidewinder addresses a key bottleneck in translating computational design into reality, with applications across health and sustainability.”
Applications
The immediate applications span medicine, agriculture, and materials science.
For personalized medicine, Sidewinder could enable rapid construction of custom cancer vaccines or engineered therapeutic proteins. In agriculture, complex genetic modifications that require multiple gene insertions become feasible. Materials science applications include engineering organisms that produce ultra-strong proteins for new biomaterials.
First author Noah Robinson, now a postdoctoral scholar who completed his PhD on the project, framed the broader significance: “To do that [cellular engineering], we need to be able to design and construct in the language of life, DNA.”
Commercialization and Access
Genyro, a biotechnology company co-founded by Wang, has secured exclusive licensing rights to Sidewinder. The company aims to make the technology available for next-generation DNA construction applications.
The research was funded by the National Science Foundation, Shurl and Kay Curci Foundation, National Institutes of Health, and Caltech’s Center for Environmental Microbial Interactions. The paper was published in Nature on January 21, 2026.
The Fine Print
Sidewinder addresses assembly accuracy but doesn’t solve other bottlenecks in synthetic biology. The oligonucleotides being assembled still need to be synthesized with their own accuracy limitations. Costs for synthesis remain significant for large-scale projects.
More fundamentally, being able to construct DNA precisely doesn’t mean we know what to construct. Designing functional genomes remains its own challenge, as the AI bacteriophage work (see related coverage) demonstrates - even with sophisticated models, most generated sequences don’t work.
Still, removing a 10,000x accuracy bottleneck changes what’s practically achievable. Combined with improving AI design tools, the gap between computational biology and functional organisms continues to narrow.