SACRAMENTO, Calif., Oct. 20, 2026 /PRNewswire/ -- Sophont, a frontier medical AI startup, announces Holon-S1, a pan-cancer AI model that predicts spatial gene and biological pathway activity from routine hematoxylin and eosin (H&E) pathology slides. By using widely available pathology slides, Holon-S1 aims to broaden access to insights that typically require costly and time-consuming molecular testing, supporting biomarker discovery and personalized treatment research.

"Most AI pathology solutions infer patient risk profiles without showing clinicians and researchers what biology resulted in the prediction," said Tanishq Mathew Abraham, Ph.D., CEO & Co-Founder of Sophont. "Holon-S1 was designed to uncover the biology behind the answer, showing which genes are active, where in the tumor they are active, and how that is related to whether a patient is likely to respond to treatment. Pathology has historically told us what a tumor looks like. Holon-S1 goes beyond this and tells us what the tumor is doing and where it may be vulnerable."

In a retrospective osteosarcoma patient case study using publicly shared slides, Holon-S1 recovered several insights originally obtained through months of extensive molecular testing.

Holon-S1 identified a fibroblast-associated program including FAP, the gene targeted by a therapy that the patient responded to. Comparing pretreatment and post-treatment slides, quantified increased immune infiltration and decreased cancer proliferation. The model had not been trained on osteosarcoma, suggesting potential generalization to unseen cancer types and analysis of archived slides without new tissue.

In triple-negative breast cancer, a novel biomarker defined by the spatial organization of predicted EREG activity was associated with response to chemotherapy before surgery in discovery and separate held-out cohorts. EREG has been implicated in chemotherapy sensitivity in other cancers; for the first time, this analysis suggested potential relevance in breast cancer. Details appear in Holon-S1's technical report.

Sophont evaluated Holon-S1 across 16 organ types, more than 15 scanners, and formalin-fixed, paraffin-embedded and fresh-frozen tissue. In a controlled comparison, it outperformed two other state-of-the-art models released this year in spatial gene activity prediction and robustness.

Holon-S1 is currently available for research use only. Sophont is designing further retrospective and prospective studies to evaluate potential diagnostic use.

"Biomedical foundation models are advancing at an extraordinary pace, and we are entering an era in which feature extraction and prediction alone may no longer suffice," said Jorge Reis-Filho, Chief Strategy Officer and Operating Partner at Foresite Labs and Venture Partner at Foresite Capital. "Models that combine high predictive accuracy with biological interpretability could enable new approaches to translational research and help us reimagine disease taxonomy and segmentation, as well as play pivotal roles in patient stratification," he added. "I look forward to seeing rigorous, independent evaluations of Holon-S1 against other leading foundation models on biologically and clinically relevant benchmarks."

The company is inviting biopharma and life sciences companies as well as research institutions to collaborate on biomarker discovery, therapeutic target research, and analysis of existing and archived patient cohorts.

Explore Holon-S1: https://sophont.med/holon

About Sophont

Sophont is a frontier medical AI lab building open, multimodal medical foundation models. Founded in February 2025 by Tanishq Mathew Abraham, Ph.D., and Paul Scotti, Ph.D., Sophont has raised $9.22M to date from Kindred Ventures, Upfront Ventures, Jeff Dean (Discovery Loop), Logan Kilpatrick (Google DeepMind), Clem Delangue (Hugging Face), and others. Sophont's mission is to build open medical superintelligence: AI that sees the whole patient, and that every clinician and researcher can use, inspect and build on. Its open releases include OpenMidnight, CortexMAE and Medmarks. Learn more: https://sophont.med.

Media Contact

[email protected]

SOURCE Sophont

Share this article