Designing novel linagliptin analogs through combined generative artificial intelligence, molecular docking, molecular dynamics simulation, and binding energy estimation for targeting fibroblast activation protein.

Fibroblast activation protein (FAP) is a highly specific biomarker overexpressed in cancer-associated fibroblasts, making it a promising diagnostic target. Developing novel high-affinity molecular probes for FAP-targeted diagnostics remains an active area of research. In this study, an integrated computational pipeline combining generative artificial intelligence, molecular docking, molecular dynamics (MD) simulation, binding energy estimation, and pharmacokinetic predictions was used to design and evaluate novel linagliptin-based compounds as potential FAP-binding analogs. From a preliminary library of 1,016 analogs generated using DeepLigBuilder, eight unique candidates with nine binding models were obtained based on binding affinity predicted using DeepLigBuilder and binding energy based on molecular docking, and four candidates were identified based on dynamic stability assessments and binding energy estimation. Two novel linagliptin-based candidates, Ligand90 and Ligand150, were identified as putative FAP-binding analogs based on ADMET profiling and Lipinski's Rule of Five compliance. This integrated computational strategy provides a valuable framework to rationally design computationally prioritized candidates for targeting FAP.