Ataraxis AI, a company building clinical AI tools to personalize cancer care, today announced scientific breakthroughs and validation results of Ataraxis’s AI models.

  • Ataraxis developed breakthrough causal AI methods that learn from large-scale, longitudinal data to predict patient outcomes in counterfactual treatment scenarios. These methods were encompassed in an AI world model called Ataraxis Tau.

  • Ataraxis Tau was applied to breast cancer, developing a model simulating patient outcomes with and without adjuvant chemotherapy.

  • In three clinical validation studies encompassing 18,426 patients, including analyses of Phase 3 randomized controlled trials TAILORx and UNIRAD, Ataraxis’s models were successfully confirmed to be prognostic and predictive of chemotherapy benefit.

  • Ataraxis Tau generalizes its prognostic and predictive capabilities zero-shot to multiple other solid tumors beyond breast cancer in an analysis of 6,692 patients across 16 other cancer types.

Instead of relying on population-level patient characteristics and biomarkers, Ataraxis developed causal AI methods that simulate individual patient outcomes in counterfactual treatment scenarios. These methods learn to recover treatment effects from routine, longitudinal clinical data as well as from randomized trials. These methods have been incorporated into Ataraxis’s broader AI world model, dubbed Ataraxis Tau.

The first version of Ataraxis Tau was trained primarily with breast cancer data to predict outcomes in patients with hormone receptor-positive, HER2-negative early-stage breast cancer. Specifically, Tau was originally trained to predict outcomes in patients treated with and without chemotherapy. This design allows Ataraxis, for the first time ever, to identify patients who–despite having high baseline risk of recurrence–might not benefit from adjuvant chemotherapy and who might be considered for other adjuvant therapy options.

Ataraxis’s capabilities in breast cancer were confirmed in three clinical validation studies released today: analyses of TAILORx and UNIRAD Phase 3 trials and a real-world multi-institutional study encompassing over 10,000 patients from 17 cohorts in 10 countries.

  1. In the analysis of the TAILORx trial, Ataraxis was found to be predictive of adjuvant chemotherapy benefit (interaction p = 0.016) within the intermediate-risk-score group. Today, these patients are de-escalated to endocrine therapy alone. In Ataraxis’s analysis, the AI model identified patients who benefit from adjuvant chemotherapy with improved 5-year disease-free survival (95.2% vs. 89.6%), while patients predicted to not benefit from chemotherapy had no significant difference in outcomes.

  2. In the analysis of the UNIRAD trial, Ataraxis identified patients within clinically high-risk, node-positive patients who have good long-term outcomes with standard of care therapy. Today, these patients are considered for escalation with drugs such as adjuvant CDK4/6 inhibitors, with no validated biomarkers to help in choices between adjuvant therapy options. Additionally, Ataraxis’s AI model was found to be predictive (p=0.01) of everolimus (mTOR inhibitor) benefit in an identified subgroup.

  3. In a study describing the development and validation of a causal AI model for chemotherapy benefit, Ataraxis included data encompassing 11,135 breast cancer patients from 17 cohorts in 10 countries. The AI model had near-perfect calibration across the full range of predicted risk and was predictive of chemotherapy benefit (p<0.001). Additionally, within a subgroup of 983 patients tested with Oncotype DX, a legacy molecular assay, Ataraxis’s AI model had more robust predictive performance (p=9.7 × 10⁻⁴ versus p=0.025).

Finally, prognostic and predictive capabilities of the Ataraxis Tau world model generalized to cancers it was never trained on. In the real-world study, Ataraxis Tau was applied zero-shot to 6,692 patients spanning 16 cancer types. Ataraxis Tau was prognostic in 11 tumor types, including gastrointestinal cancers, lung cancers, skin cancers, and others. Chemotherapy benefit prediction also transferred zero-shot to five of the seven cancers treated with platinum- or taxane-based regimens, which are shared with breast chemotherapy, showing that Tau learned a treatment-specific predictive signal.

“Today’s results mark a major milestone for Ataraxis. We have previously shown that our AI models can accurately predict outcomes and treatment response in breast cancer. Now, we have gold-standard evidence supporting those findings and the broader clinical use of our tools. These results also bring us closer to a new era of personalized treatment. Our AI world model, Ataraxis Tau, can predict recurrence and treatment benefit across cancer types. As we scale Tau, it will serve as the foundation for future tools designed to help personalize treatment for every cancer patient,” said co-founder and CEO of Ataraxis AI, Jan Witowski, MD, PhD.

This distinguishes the result from prior pan-cancer analyses of AI pathology models, which have reported performance only after fitting or fine-tuning a separate model for each cancer type. A causal model trained on breast cancer alone identified risk and chemosensitivity across most other cancers, suggesting it has learned determinants of treatment response that are shared across malignancies rather than specific to one disease.

Ataraxis is building on that foundation, extending individualized benefit prediction to additional therapies and tumor types and working toward a causal model of treatment response that spans oncology, so that any patient facing a treatment decision can be told not just how aggressive their cancer is, but what a given therapy is likely to do for them.

About Ataraxis AI

Ataraxis AI is a company building clinical AI tools to personalize cancer care. By integrating routine digital pathology with clinical data, Ataraxis aims to deliver individualized insights into prognosis and treatment benefit across multiple cancer types.

Ataraxis™ Breast is Ataraxis AI’s flagship multimodal platform for breast cancer. Delivered as a suite of AI-powered clinical tests, it combines routine digital pathology with clinical data to generate predictions of recurrence risk and therapy response, supporting decision-making across the full continuum of care. The Ataraxis Breast platform was developed using 30,000+ multimodal patient cases and is designed to fit into existing pathology workflows without requiring additional tissue or specialized assays. All Ataraxis Breast tests are offered as laboratory-developed tests through a CLIA-certified and CAP-accredited laboratory. To learn more, visit ataraxis.ai.

Media gallery

About The Author
Business Wire