Evidence-informed dynamic modelling and intervention prioritisation in diabetic bone marrow

Aims. Fragmented understanding of how diabetes disrupts interconnected bone marrow niches has hindered integrated strategies to restore function. Building on a previous static network, we developed a dynamic model to simulate intervention responses. Methods. The investigator and ChatGPT dialogued to extract evidence from a frozen corpus and develop the model. A Python implementation extended Boolean logic to three ordinal levels (0, 1 and 2) representing biological states and, where applicable, intervention strength. Under high metabolic stress, 15 interventions were assessed at three strengths and four with binary settings; 171 pairs were tested at maximal strength. A domain balanced score (0 to100) integrated favourable and adverse changes. Staged simulations assessed withdrawal; regulatory constraints on recovery after metabolic stress reduction were examined separately. Results. Endothelial reactive oxygen species (ROS) suppression achieved the highest score among 19 single interventions (24.2), improving three functional domains. Endothelial protein kinase B (Akt) augmentation scored 15.2 (two domains); oncostatin M (OSM) blockade and haematopoietic 66 kDa SHC transforming protein 1 (p66Shc) loss each scored 13.6 (two domains). ROS suppression combined with either OSM blockade or p66Shc loss scored 37.9, (five domains). No pair restored all functions under high stress. Benefits of ROS suppression, Akt augmentation and ROS and OSM blockade disappeared after withdrawal. Reducing metabolic stress led to variable scores (range 4.5 to 100), increasing mainly with restored endothelial function, followed by progenitor recovery and interruption of inflammatory and adipocyte stromal signalling. Conclusions. These simulations offer a rule based framework for investigating diabetic marrow recovery. Experimental studies must establish whether it can guide targeted therapy.