AI plans gene insertion
Basecamp Research unveiled an AI model that the team says can plan programmable gene insertions for therapeutic uses, a reported first in its class. (x.com) The model is presented as a design tool to guide gene‑editing strategies rather than a clinical treatment on its own. (x.com)
Gene editing is a way to rewrite DNA, but adding an entire working gene at an exact spot has been much harder than making small cuts. Large insertions are the part Basecamp Research says its new artificial intelligence system can now help plan. (nature.com, nature.com, prnewswire.com) Basecamp Research announced on January 12, 2026 that it had built what it called the first artificial intelligence models for programmable gene insertion. The company said the system is meant to design enzymes that place large therapeutic DNA sequences at chosen sites in the human genome. (prnewswire.com, genengnews.com) The model family, called EDEN, was developed with NVIDIA and trained on about 9.7 trillion biological tokens from BaseData, Basecamp’s proprietary genomics dataset. Basecamp said BaseData was collected over five years from more than 150 locations across 28 countries and five continents. (genengnews.com) The basic problem is that many diseases would be easier to treat by swapping in a full healthy gene, not by repairing one letter at a time. Existing approaches for large insertions have often run into low efficiency, off-target activity, size limits, or complex delivery requirements. (nature.com, nature.com) Scientists have been pushing on that problem with several toolkits, including Programmable Addition via Site-specific Targeting Elements, or PASTE, and engineered recombinases. In 2025, a Nature Biotechnology paper reported up to 53% integration efficiency and 97% genome-wide specificity for engineered recombinases at an endogenous human locus, while a 2024 Nature Protocols paper described PASTE inserting cargoes of at least 36 kilobases with efficiencies of up to 60%. (nature.com, nature.com) Basecamp’s claim is different from those studies because it says the artificial intelligence model designs the insertion enzyme from the target DNA sequence alone. In its January 2026 announcement and accompanying preprint, the company said EDEN produced active insertion proteins for 100% of tested disease-relevant target sites and demonstrated insertion at more than 10,000 disease-related locations in the human genome. (prnewswire.com, biorxiv.org, genengnews.com) The work is still early. GEN reported that the EDEN paper was posted as a bioRxiv preprint and had not yet been peer reviewed as of January 12, 2026. (genengnews.com, biorxiv.org) Basecamp also said it used the system to integrate cancer-fighting DNA into primary human T cells at new safe-harbor sites, producing chimeric antigen receptor T cells that cleared more than 90% of tumor cells in lab assays. Those are preclinical laboratory results, not evidence of safety or benefit in patients. (genengnews.com) The company paired the scientific announcement with financing news. Basecamp said NVentures, NVIDIA’s venture capital arm, invested in its pre-Series C round, and the Financial Times reported in January that NVIDIA and Microsoft were backing the company’s gene-therapy push. (prnewswire.com, ft.com) What happens next is less about one software model than about whether outside groups can reproduce the enzyme designs and move them through animal studies and human trials. Until then, Basecamp’s January 2026 announcement is a research claim about a design tool, not a marketed treatment. (biorxiv.org, genengnews.com)
