Causality
Ataraxis is developing methods that infer the true biological effects of treatment, not just surface-level correlations in data. Our research aims to reveal why certain therapies work, for whom, and under what conditions.
The limits of population-level predictions
The existing prognostic tests do not estimate a personalized effect of specific treatment. A patient with high-risk, aggressive disease may see little benefit from chemotherapy, while a lower-risk patient may benefit substantially. Our methodology is designed to surface that variation, estimating treatment effect at the individual level rather than inferring it from group averages.
The challenge: real-world data is biased
In real-world oncology care, treatment is not assigned randomly. Patients with higher-risk disease are more likely to receive intensive therapy, while lower-risk patients may avoid it. This creates treatment assignment bias in observational data. Without correcting for this bias, AI models can draw the wrong conclusion, for example suggesting that treatment leads to worse outcomes simply because treated patients were sicker to begin with. Causality helps address this challenge, allowing us to modify our models such that they estimate the true effect of treatment.
Introducing Ataraxis™ Tau
Ataraxis Tau is our causal AI framework that enables us to provide personalized predictions of treatment benefit, moving beyond the prognostic predictions, ignoring the impact of treatment, by standard-of-care tests. Tau is trained using advanced causal inference methods that learn individualized treatment effects from real-world data, enabling us to train on larger, more diverse datasets without relying on scarce clinical trial data.
We use Ataraxis Tau to build highly actionable clinical-grade diagnostic tests and tools. We have already developed the first tests built on top of Tau, including Ataraxis Breast CTX, our individualized chemotherapy benefit prediction test for HR+/HER2- breast cancer patients. The same technology of Ataraxis Tau can be applied to other questions, including the benefit of radiotherapy, novel endocrine therapies, CDK4/6 inhibitors, and more, both within breast cancer and beyond.
Tau trains on large, multi-institutional real-world datasets and uses adjustment techniques to account for treatment-assignment bias, enabling learning from observational data to approximate learning from randomized clinical trial data. By learning across multiple cohorts, Tau learns to generalize beyond any single institution.
Chemotherapy benefit is the first application of Tau because it is common and clinically consequential. The same causal framework can extend to other oncology decisions, including endocrine therapies, targeted therapies, and immunotherapy strategies as data grows.
Ataraxis Tau in practice: Predicting chemotherapy benefit
Ataraxis Breast CTX is the first large-scale clinical application of the Ataraxis Tau platform. It is built to estimate the true, individualized benefit of chemotherapy using real-world data.
While many existing approaches rely on a single risk score and assume higher risk implies higher benefit, CTX evaluates outcomes under two treatment pathways, identifying high-risk patients unlikely to benefit from chemotherapy as well as lower-risk patients who may still derive meaningful benefit.