Project Summary Atrial fibrillation (AF) is a major arrhythmia worldwide, causing palpitations, stroke and mortality, and affecting 2-5 million Americans. Unfortunately, therapy to eliminate AF has had limited success. In our last funding cycle, we focused on localized drivers as potential AF mechanisms. Mapping of drivers has now been validated by concurrent optical mapping of human AF, and their features and have been validated by several other methods in patients. Nevertheless, ablation results for these and other proposed mechanisms for AF outside the pulmonary veins are mixed. It is unclear if this reflects difficulties of AF mapping, or different mechanisms between patients. The project will develop a novel mechanistic framework for AF that simplifies existing indices by building on scientific consensus that organized AF is easier to treat, and disorganized AF has worse prognosis. This concept spans many existing indices and may help to reconcile them. We have 3 specific aims: (1) To define if the impact of ablation depends on the extent of organizing surrounding the ablation site; (2) To establish candidate mechanisms for organized and disorganized AF zones in individual patients with specific profiles, using machine learning applied to known cases with and without ablation success in our large registry. This comprises detailed AF maps during ablation and after Maze surgery, clinical data and outcomes. (3) To use novel clinical tools to predict whether patients will respond to PVI, other ablation or Maze surgery based on whether targeted regions control larger atrial areas and their locations. This study will deliver immediate translational and clinical impact, and directly enable personalized medicine for AF ablation. We use detailed clinical mapping in patients, signal processing and computer modeling to develop a novel mechanistic framework and widely applicable clinical tools. We will use tools including machine learning and statistics to classify mechanisms based upon outcomes from ablation in individual patients. We will make our data and code available online. Our team is experienced in electrophysiology, computer science, machine learning, biological physics and statistics. The proposal is thus highly feasible.