Recursion showcases AI atlas
- Recursion highlighted AI that integrates patient data with cell-level models to improve oncology translation at AACR. - The company presented a CellNeighbor atlas technique to bridge preclinical models and patient data. - The work underscores industry focus on data integration and model-driven translational research in cancer R&D (x.com).
Recursion used the American Association for Cancer Research meeting this month to show a new AI mapping method that matches lab cancer models to real tumors more closely. (aacrjournals.org) Cancer drug teams often test compounds in cell lines, which are tumor cells grown in the lab for years. Recursion’s AACR 2026 abstract says that long in vitro culturing can change those cells enough to raise doubts about whether they still represent the patients they originally came from. (aacrjournals.org) Its answer is a transcriptomic atlas, a map built from gene-activity patterns rather than microscope images alone. The company said CellNeighbor combines cell-line data from DepMap with tumor data from The Cancer Genome Atlas and deidentified patient tumor data from Tempus. (aacrjournals.org; recursion.com) The method first strips out gene-expression signals tied to the tumor microenvironment, the non-cancer cells and surrounding tissue that can blur comparisons. It then uses Harmony, a data-integration tool, and a nearest-neighbor approach to place each cell line inside a “patient neighborhood” of transcriptionally similar tumors. (aacrjournals.org) That gives researchers a ranked list of which cell lines look most like a target patient group before they run compound screens. Recursion said it also built a neighborhood-tissue homogeneity score to judge how confident those matches are and support more automated model selection. (aacrjournals.org) The presentation landed at AACR’s April 17-22, 2026 annual meeting in San Diego, where companies and academic labs use posters to signal which tools they want oncologists and partners to notice. AACR says the annual meeting is the main gathering for the cancer research community. (aacr.org; aacr.org) Recursion has been building toward this patient-data push for at least a year. In a March 12, 2025 post, the company said it was combining machine-learning models trained on cell data with multimodal, deidentified patient data from partners including Tempus and Helix, and said the Tempus dataset alone was about 20 petabytes. (recursion.com) The company’s oncology emphasis is also broader than one poster. Bayer and Recursion said on November 9, 2023 that they had refocused their collaboration on precision oncology, with the deal covering up to seven oncology programs and up to $1.5 billion in potential milestone payments plus royalties. (bayer.com) Recursion told investors on February 25, 2026 that it had $754 million in cash and cash equivalents and was advancing multiple clinical and preclinical programs while continuing to invest in its AI platform. That leaves CellNeighbor looking less like a one-off conference demo and more like infrastructure for how the company wants to choose cancer models going forward. (ir.recursion.com; aacrjournals.org)
