Alphabiome

Alphabiome is an AI company applying machine learning to raw genetic sequencing data for drug development. Its core claim is that reference-based genomics tools can only interpret DNA that has already been mapped, while the majority of the microbial and viral material in a biological sample has no reference genome. The company states that reference genomes represent roughly 1% of microbial diversity. Alphabiome's engine, described as a "Genetic Radar," tokenizes raw sequence directly across host, bacterial, viral and fungal DNA, learning statistical structure from k-mers without labels, dictionaries or prior assumptions, and converting raw reads into AI-usable tokens for prediction, patient stratification, mechanism discovery and therapeutic decision-making.

The stated first application is drug development: predicting which patients will respond to a therapy, which will not, and why, with the aim of compressing translational research timelines and allowing clinical trials to run with fewer, better-selected patients. The company positions this representation layer as sitting upstream of biological foundation models, and says its engine has scanned trillions of DNA fragments and discovered more than 100,000 novel signatures, opening a discovery space it describes as 1,000 times larger than traditional workflows. Company communications also cite a trial in which microbiome samples from ulcerative colitis and Crohn's disease patients treated with six different drugs were analyzed to identify over 10,000 microbe-derived biomarkers associated with treatment success.

Scientific and executive leadership includes founder and CEO Dr. Yaniv Altshuler, an MIT researcher who has published three books, more than seventy academic papers and twenty patents in AI, alongside MIT Prof. Alex (Sandy) Pentland and Stanford Nobel laureates Prof. Roger Kornberg and Prof. Michael Levitt. The company's technical lineage is described as an application of decentralized and scalable AI, swarm intelligence and social physics methods to genomics.

Founding story

The underlying AI technology was developed by founder and CEO Dr. Yaniv Altshuler, an MIT researcher who worked with MIT Prof. Alex Pentland on swarm intelligence and social physics; those decentralized, scalable AI methods were later applied to genomics, with the R&D team's work on predicting feed additive efficacy from rumen sequencing published in CRC Press's Applied Swarm Intelligence. The founding team is described as including MIT AI researchers and Stanford biochemistry Nobel laureates.

Business model

Alphabiome develops a proprietary AI representation and prediction layer intended for use in pharmaceutical drug development and clinical treatment selection. Public materials describe value delivered to pharmaceutical companies through faster and more effective drug development, to doctors and hospitals through prediction of medicine efficacy, and to insurers and governments through reduced spending on ineffective treatment, but do not specify pricing or contracting terms.

Traction

Company-published items report a series of validation results: a peer-reviewed study across 13 commercial dairy herds predicting additive response from raw genetic signal (April 2025); a controlled trial validating the engine on a second compound (September 2025); a peer-reviewed study across 10 farms and 339 animals (November 2025); an extension of response prediction to human disease (November 2025); a leading-hospital clinical study predicting response across seven biologic drugs with up to 3x higher accuracy than state-of-the-art models (February 2026); and an extension of the representation layer to a new domain covering three indications and eight therapies (April 2026). A press release also describes a trial in ulcerative colitis and Crohn's patients across six drugs that identified over 10,000 microbe-derived biomarkers.

Latest developments

In February 2026 the company reported a hospital-based clinical study in which its AI predicted response across seven biologic drugs, citing up to 3x higher accuracy than state-of-the-art models. In April 2026 it announced an extension of its representation layer to a new domain, covering three indications and eight therapies.