Radical Numerics opens restricted Omnii-1 biodefense API and laptop deployment

Radical Numerics has opened restricted access to its Omnii-1 biodefense API and packaged the sequence-analysis model for local laptop use, according to an October 20 research preview. Vetted partners and external testers can use the release to analyze samples when connectivity is unreliable or sequence data must remain on site.

Partners can submit DNA or protein sequences, including metagenomic reads and assembled sequences, and receive flagged results tied to model predictions, taxonomic assignments and other bioinformatic evidence. Radical Numerics describes Omnii-1 as a function-aware layer: it predicts what a sequence may do, then combines that signal with conventional alignment, database search and taxonomic analysis. This retains reference-based evidence for close matches while scoring sequences that differ substantially from known references.

For an edge test, the company paired an Oxford Nanopore MinION portable sequencer with a MacBook Pro using an Apple M3 Max chip and 128 GB of memory. Radical Numerics said data streamed directly from the sequencer through basecalling, bioinformatics and Omnii-1. A separate workflow snapshot replayed time-stamped data from an earlier sequencing run. The demonstration used company-run tests and spiked sequences, so it did not establish independently observed field performance or operational error rates.

The company reported that Omnii-1 reached 98.5% recall at a 0.5 probability threshold on 16,307 likely functional pathogen-paraphrase proteins, compared with 83.3% for DIAMOND. In a separate viral-read evaluation using simulated mutations across 24 species, it reported mean recall gains of about 31% for arbitrary mutations and 52% for synonymous mutations over a comparison metagenomic workflow, with false-positive rates no higher than 0.5% in the tested background conditions.

Those benchmark and pathogen-detection results are issuer-reported and have not been independently validated. No completed third-party benchmark was identified in the research sources. The company has not identified the vetted API users or independent evaluators, and the opened sources did not provide pricing, a general-availability date or public API documentation.