AI for gene insertion
Basecamp Research unveiled an AI model designed to programmatically suggest gene insertions for cell and gene therapies, a first-of-its-kind approach being discussed in the research community. The announcement frames the model as a tool for designing therapeutic insertions rather than a finished clinical treatment. (x.com)
Gene insertion is the problem of adding a long stretch of DNA at a chosen address in a cell’s genome. Basecamp Research said on January 12 it built an artificial intelligence system to suggest the enzymes for that job. (prnewswire.com) In cell and gene therapy, that kind of insertion could be used to replace a faulty gene or reprogram immune cells with new DNA instructions. Basecamp said its models are aimed at those design tasks, not at a finished treatment ready for patients. (prnewswire.com) The company’s system, called EDEN, is a family of foundation models trained on DNA sequences and evolutionary data. A January 12 bioRxiv preprint said the largest version has 28 billion parameters and was trained on 9.7 trillion nucleotide tokens from a proprietary dataset. (biorxiv.org) The basic technical hurdle is size and precision at the same time. Researchers have spent years trying to place large DNA cargo at exact genome sites without cutting both strands of DNA and creating unwanted byproducts. (nature.com) Several methods already pursue that goal. PASTE, published in Nature Biotechnology in 2023 and expanded in a 2025 Nature Protocols paper, reported insertion of DNA cargo as large as about 36 kilobases in mammalian cells. (nature.com 1) (nature.com 2) Another line of work uses CRISPR-associated transposases, which are bacterial insertion systems adapted for human cells. A 2024 Nature Biotechnology paper reported targeted DNA integration in human cells without double-strand breaks using a Type I-F CRISPR-associated transposase. (nature.com) Basecamp’s claim is different from those earlier platforms because it says the model designs the insertion enzyme from a target DNA sequence prompt. In its preprint and announcement, the company said EDEN designed active large serine recombinases for 100% of the disease-relevant target sites it tested. (biorxiv.org) (prnewswire.com) Basecamp also said it showed insertion at more than 10,000 disease-related genomic locations and used the approach to integrate cancer-fighting DNA into primary human T cells. The company said those engineered cells cleared more than 90% of tumor cells in laboratory assays. (prnewswire.com) The outside validation is still early. The January 12 paper is a preprint, which means it has not been peer reviewed, and the evidence now in public is limited to lab data and company disclosures rather than clinical results. (biorxiv.org) Basecamp said NVIDIA helped train the models and that NVentures, NVIDIA’s venture arm, invested ahead of the company’s Series C round. That leaves the next test in plain view: whether an artificial intelligence system that can propose insertion tools on paper can keep working under the slower standards of peer review, reproducibility, and eventually drug development. (prnewswire.com)
