Machine‐Learning‐Guided DNAzyme Nanocages for Photoresponsive Immunomodulation and Spatiotemporal Control of Diabetic Bone Regeneration

ABSTRACT Inflammation poses a therapeutic paradox, as cytokine signals driving inflammatory disease can also provide transient repair cues. Effective intervention therefore requires selective suppression of pathogenic mediators and control over inhibition timing. RNA‐cleaving DNAzymes are well suited as transcript‐selective catalytic effectors, but their therapeutic utility depends on discovering potent sequences and controlling when intracellular catalysis begins. Here, machine learning identifies highly active DNAzymes against tumor necrosis factor‐α (TNF‐α) as a model inflammatory regulator and reveals design principles governed by RNA accessibility and RNA–DNA interaction energetics. The optimized DNAzyme is incorporated into an upconversion nanoparticle‐coupled DNA nanocage through photocleavable linkers, generating a photoresponsive construct for near‐infrared‐triggered release and spatiotemporally controlled gene silencing. The construct exhibits macrophage anti‐inflammatory activity and, after microneedle delivery, shows efficacy alongside etanercept in a TNF‐associated inflammatory skin disease model. Beyond this benchmark, the system is further applied to diabetic bone regeneration, where delayed activation reshapes inflammatory dynamics in a regenerative microenvironment. This temporal intervention preserves beneficial early inflammation while promoting subsequent resolution through enhanced efferocytosis, thereby accelerating bone repair and shifting DNAzyme therapy from indiscriminate inhibition toward phase‐specific immunoregulation. This work establishes a machine‐learning‐guided, photoactivatable DNAzyme nanocage strategy to enable targeted inflammatory control and diabetic bone regeneration.