Noetik

Noetik is a privately held AI-native biotechnology company in South San Francisco, founded in 2022 by Ron Alfa, its chief executive and formerly head of research at Recursion, and Jacob Rinaldi. It builds foundation models of cancer biology, led by the OCTO family, trained on tumor data it generates itself. It appears here not as an AI adopter but as an AI model builder: its product is the models, which it licenses to drug developers rather than selling self-serve software, and its website names no drug candidate of its own.

Its defining pattern is data-first model licensing. Noetik produces its own training data, pairing spatial protein and gene-expression maps, pathology slides, DNA sequencing and clinical records from the same patient tumors. A September 2024 technical report described a lung cancer atlas of more than 1,000 cases collected on six continents, and an in vivo CRISPR platform called Perturb-map, launched in November 2023, tests hypotheses in mouse models. The company describes the result as the largest spatial biology dataset in oncology. It raised a 14 million dollar seed round led by DCVC in September 2023 and a 40 million dollar Series A led by Polaris Partners in August 2024.

The models followed the data. OCTO, short for Oncology Counterfactual Therapeutics Oracle, was introduced in a July 2024 technical report as a multimodal transformer that simulates a patient's tumor. The virtual cell model, OCTO-VC, followed in December 2024: it places computational cells at chosen spots in real patient tissue and predicts their gene expression, and a public web demo called Celleporter shows how that works. In June 2025, Agenus described OCTO as a 1.5 billion parameter model trained on about 200 million tumor and immune cells. A separate model, TARIO-2, reads routine hematoxylin and eosin (H&E) pathology slides, and Noetik says it was trained on samples from 19 cancer types.

In January 2026, GSK took a five-year, non-exclusive license to the OCTO-VC models for non-small cell lung cancer and colorectal cancer. Noetik says the deal carries 50 million dollars in upfront payments and near-term milestones plus an annual license fee, and in August 2026 it reported the first milestone: model inference and fine-tuning delivered to GSK and meeting the agreed testing criteria. Agenus signed a June 2025 collaboration to find patients likely to respond to its immunotherapy combination of botensilimab and balstilimab. In June 2026, Noetik reported that TARIO-2 picked out patients with better response and survival from pretreatment slides in a retrospective analysis, with what it calls roughly three times the expected response rate of earlier trials.

The honest framing: Noetik's value rests on a proprietary tumor dataset built in its own labs, with the models as the way that data is sold. The GSK license shows a large drug developer paying a recurring fee for access to models alone, but it is non-exclusive and limited to two cancers, and the TARIO-2 result is a retrospective analysis rather than a prospective trial. Among peers, Owkin builds on patient data from hospital partners and sells an agent, K Pro; Bioptimus trains pathology foundation models; Tempus AI pairs a diagnostics business with data licensing; and Insilico Medicine has taken its own drug candidates into clinical trials, a step Noetik has not announced.