PROJECT SUMMARY / ABSTRACT More than 1 million Americans are hospitalized with sepsis each year, and nearly one in five don’t survive. Most efforts to reduce sepsis deaths begin with the premise that patients are largely similar, and that ether moving treatment earlier or targeting therapeutics to a single mechanism will improve outcomes. In prior work funded by a NIGMS R35 award, we derived sepsis endotypes using a suite of machine learning methods inside the electronic health records (EHR) in a large integrated health system. These endotypes differed in biology, outcomes, and treatment response, and were reproduced in thousands of patients. But how will they lead to precision care? In this Renewal, we will leverage our clinical translational laboratory and remnant blood collection to better understand the biology of sepsis endotypes and explore new domains related to pathogen, microbiome, and molecular mechanisms. We will use Bayesian causal networks and reinforcement learning to optimize treatment policies over endotypes in more than 10 million EHR encounters. Finally, we will move learning online and embed endotypes inside the EHR at the point-of-care. These steps will take the science of sepsis endotypes and inform clinical decisions made under time pressure and uncertainty. By testing endotype treatment policies at the “live-edge”, we will strengthen causal inference, mechanistic insight, and learn while doing. My program will be supervised by external advisory boards with expertise in machine learning, inflammation, immunology, computational and systems biology, causal methods, artificial intelligence, and health information technology. This work will further develop my clinical-translational laboratory and cross-cutting mentorship of junior scientists.