Diabetic wounds remain challenging due to complex nanoscale dysregulation under hyperglycemic conditions. A computational nanomedicine pipeline that coupled large language model (LLM)-powered literature mining for qualitative insights with multistage molecular simulations for quantitative validation was presented to investigate drug–protein nanointeractions. The pipeline first mapped 2989 existing drugs against 8739 diabetic wound-related protein targets and utilized an LLM-based analysis to qualitatively evaluate each drug–protein regulatory effect from literature evidence. A cheminformatic clustering with greedy coverage then distilled this vast search space down to 35 candidate drugs and 50 key proteins. These candidates were subsequently subjected to sequential molecular docking, molecular dynamics (MD), and quantum chemistry (QC) simulations to quantify their nanoscale binding interactions. By combining the AI-derived regulatory insights with physics-based binding metrics, the pipeline ranked all candidates using a composite (anti)therapeutic score. Folic acid emerged as the top candidate, consistent with pro-regenerative effects reported in the literature and exhibiting a strong interaction energy to fibroblast growth factor (ΔEinteraction= −78.1 kcal/mol) in simulations. In vitro scratch wound assays confirmed that folic acid accelerated wound closure to 134.90% of the untreated control (p < 0.001), in agreement with the in silico predictions. Overall, this integrated AI-guided nanoscale modeling approach shortened the literature-to-experiment cycle by over 70% compared to conventional methods and demonstrated a translational strategy that bridges nanoscale molecular interactions with therapeutic outcomes. These findings exemplify how combining AI-driven literature mining with quantitative nanoscale modeling can accelerate drug repurposing for diabetic wound care and other complex diseases.
Zhang et al. (Mon,) studied this question.