Abstract
The clinical translation of metal-organic frameworks (MOFs) - a promising class of porous materials for nanomedicine - is hindered by a poor understanding of their complex interactions with the immune system and in vivo immunotoxicity. To address this gap, a hierarchical "Safety-by-Design" pipeline is established and validated, integrating machine learning (ML) with ex vivo human blood studies and targeted in vivo models. This multi-stage workflow enables the systematic profiling of MOF immunotoxicity, de-risking their development. The power of this approach is demonstrated using four clinically relevant MOFs - NU-901, PCN-222, UiO-66, and ZIF-8 - revealing distinct, framework-dependent immune fingerprints. The initial in silico screening correctly flagged NU-901 and ZIF-8 as potential hazards to human health. These predictions are subsequently validated ex vivo, where NU-901 is confirmed to be selectively cytotoxic to CD14(+) monocytes, and ZIF-8 is identified as a specific pro-inflammatory agent via IL-6 induction. In contrast, candidates predicted to be safe - UiO-66 and PCN-222 - demonstrated high biocompatibility ex vivo and advanced to in vivo studies, where they caused only minimal and transient immune activation. This study provides a validated, resource-efficient roadmap for preclinical immunotoxicity assessment, establishing a rational paradigm to accelerate the safe clinical translation of MOFs and other advanced nanomedicines.