Prevalence of atrial fibrillation (AF) has surged by over 30% in the past two decades, prompting a shift toward personalized medicine to improve diagnostic and prognostic accuracy. Computational approaches are expected to enhance the precision of AF diagnosis and treatment by integrating valuable biomarkers improving individualized therapies. However, accurately capturing inter-patient variability, including both anatomical and electrophysiological characteristics, is critical for the effectiveness of these models. In this study, we aim to simulate arrhythmic behaviors using virtual patient cohorts with bi-atrial geometries, focusing on the dependencies between anatomical and electrophysiological markers and their propensity to induce AF. We performed 800 simulations across 20 virtual patients, including variations in electrical and structural remodeling. Relationships between anatomical or electrophysiological parameters have been associated with arrhythmic vulnerability rates. We found that virtual patients with increased atrial lateral extent, longer Bachmann’s bundle length, and higher total activation time biomarkers are up to 20% more likely for AF to be induced. We also identified that virtual patients with larger atria are more vulnerable to AF progression, as increased electrical remodeling provided +22% more arrhythmic vulnerability. This work aids in identifying biomarkers for patient stratification, enabling more precise and personalized AF therapies.