Better checkpoint blockade biomarker discovery with Bayesian survival modeling

Identification: 4028


Better checkpoint blockade biomarker discovery with Bayesian survival modeling

Tavi Nathanson1, Jacqueline Buros Novik1, Bulent Arman Aksoy1, Alexander Rubinsteyn1, Eliza Chang1, Arun Ahuja1, Jeff Hammerbacher1, 2Alex Snyder, 2Matt Hellmann

1Icahn School of Medicine at Mount Sinai, 2MSK

Our lab has been working for several years to discover biomarkers to predict the therapeutic or adverse effect of checkpoint blockade. We've analyzed or re-analyzed molecular profiling data from hundreds of patients and several tumor types and have made several improvements on the probabilistic models used to predict patient response. These improvements allow us to jointly model progression and mortality as separate outcomes, evaluate interactions of biomarkers, allow the effect of biomarkers to vary over time, and much more. Collectively, these improvements support the development of richer models to describe the response to checkpoint blockade. We illustrate the benefits of this approach in the context of simulated data and examples from our recent analyses of trial data. The software implementing these methods has been released as an open source Python library. The models themselves are written in Stan, a probabilistic programming language released under a permissive open source license, which has interfaces to R, Python, Matlab, Stata and Julia.


Credits: None available.

You must be logged in and own this product in order to post comments.