Abstract Objective Post-surgical seizure freedom in drug-resistant epilepsy (DRE) patients varies from 30 to 80%, implying that in many cases the current approaches fail to fully map the epileptogenic zone (EZ). This suggests that the EZ entails a broader epileptogenic brain network (EpiNet) beyond the seizure-zone (SZ) that show seizure activity. Methods We first used computational modeling to identify putative complex-systems- and systems-neuroscience-driven mechanistic biomarkers for epileptogenicity. We then extracted these epileptogenicity biomarkers from stereo-EEG (SEEG) resting-state data from DRE patients and trained supervised classifiers to localize the SZ with these biomarkers against gold-standard clinical localization. To further explore the prevalence of these pathological biomarkers in an extended network outside of the clinically-identified SZ, we also used unsupervised classification. Results Supervised SZ-classification trained on individual features achieved accuracies of 0.6–0.7 areaunder-the-receiver-operating-characteristics curve (AUC). However, combining all criticality and synchrony features improved the AUC up to 0.85. Unsupervised classification uncovered an EpiNet-like cluster of brain regions with 51% of regions outside of SZ. Brain regions in this cluster engaged in inter-areal hypersynchrony and locally exhibited high amplitude bistability and excessive inhibition, which was strikingly similar to the high seizure-risk regime revealed by computational modeling. Significance The finding that combining biomarkers improves EZ localization shows that the different mechanistic biomarkers of epileptogenicity assessed here yield synergistic information. On the other hand, the discovery of SZ-like pathophysiological brain dynamics outside of the clinically-defined EZ provides experimental localization of an extended EpiNet. Key points We advanced novel complex-systems- and systems-neuroscience-driven biomarkers for epileptogenicity Increased bistability, inhibition, and power-low scaling exponents characterized our model operating in a high seizure-risk regime and SEEG oscillations in the seizure-zone (SZ) Combining all biomarkers yielded more accurate supervised SZ-classification than using any individual biomarker alone Unsupervised classification revealed more extended pathological brain networks including the SZ and many non-seizure-zone areas that were previously considered healthy