Cancer AI Models Will Reach Clinical Trials Within a Year | Tanishq Abraham

Summary · 3 min read

Tanishq Abraham is co-founder and CEO of Sophont. The company built Holon-S1, a model trained on pathology tissue slides to extract biological insights from tumors, and also works with neuroimaging.

Abraham says Holon-S1 uses standard pathology slides to map tumor biology, which could help researchers and clinicians tailor cancer care to individual patients. He describes an osteosarcoma case where the model identified a biological insight and a gene the medical team targeted, while emphasizing that the model needs further validation. He argues that AI trained on human clinical data could help reveal which patients respond to treatments, and that diagnostic and treatment-response tools may affect patients sooner than drugs developed from scratch. He says clinical trials could provide evidence for these tools within a year, and points to open-source bio-AI projects as a way to contribute.

Minute by minute

1. 0:36

   Holon-S1 reads tumor biology from slides. Abraham says Holon-S1 was trained on pathology tissue slides to infer gene expression, cell interactions, pathways and other biology from a standard tumor biopsy. He argues that patients with the same diagnosis can have different underlying biology, while molecular tests can be costly and slow.

2. 3:29

   Slide analysis may help study rare cancers. Abraham says the model was not trained on osteosarcoma but identified biological insights in a patient’s tumor, including a gene the medical team targeted. He says the team had found related insights using expensive molecular tests and that the patient was successfully treated, while stressing that the result needs further validation.

3. 5:14

   Personalized care starts with mapping tumors. Abraham says he sees cancer care not as finding one magic drug, but as building a generalizable framework for personalized treatment. Holon-S1 focuses on mapping a tumor’s biology; he says those insights could then help guide treatment design.

4. 7:52

   Applications beyond cancer remain untested. Abraham says Holon-S1 was trained only on cancer data, so it is not known whether it works beyond cancer. He sees possible research directions in autoimmune disease and neuroimaging, where Sophont studies differences among patients with neurodegenerative diseases.

5. 9:32

   Research and pharma are potential early users. Abraham says researchers could apply Holon-S1 to archives of pathology slides to find biological patterns, while pharma companies could use its pathway features to study or predict treatment response. He says the insights could also support diagnostic or treatment-selection tests, but patient-care uses require substantial validation.

6. 12:33

   Human data can clarify drug response. Abraham argues that a key drug-development challenge is learning whether a treatment works in people: simulations, lab tests and animal studies do not fully represent human biology. He says models trained on clinical data could help explain why some patient groups respond and others do not, and point to pathways or targets for further study.

7. 15:07

   Abraham urges focus on nearer-term patient gains. Abraham says frontier labs are moving into biology, but he cannot know exactly what they are doing and the public work he can see remains limited in scope. He argues that drug discovery may take a long time, while diagnostic and treatment-response tools could affect patients sooner.

8. 19:35

   Open-source projects offer a way in. Abraham recommends joining Sophont’s open-source community, MedArc. He says the company releases much of its research and its core models, and offers projects, a challenge and a leaderboard for work on understanding tissue slides.

9. 21:27

   Trials could provide evidence within a year. Abraham says a year is a tight timeline, but expects some tools could reach retrospective clinical trials and potentially prospective trials, producing evidence of their usefulness. He says simple treatment-response tools might begin helping patients within a couple of years if the field commits more resources—not that cancer will be cured in a year.

Published Oct 21, 2026.