From mushroom metabolome to a prioritized anti-diabetic metabolite candidates: LC-MS/MS and AIDDISON™-assisted screening of Pleurotus citrinopileatus

The decline in pharmaceutical R&D productivity necessitates innovative strategies to explore nature's vast chemical diversity. Traditional natural product discovery is hampered by slow, resource-intensive methods where novel therapeutics may reside. This study introduces a high-throughput workflow integrating untargeted metabolomics with an artificial intelligence (AI) drug discovery platform to rapidly identify anti-diabetic candidates from the mushroom Pleurotus citrinopileatus . High-resolution LC-MS/MS analysis identified 1025 distinct molecular features, which were annotated with high confidence using sub-ppm mass accuracy and MS 2 fragmentation validation. These metabolites were subjected to in silico ADMET profiling and pharmacokinetic clustering, which prioritized a subset of 27 compounds (Cluster 14) with favorable pharmacokinetic profiles including predicted active transport permeability. Subsequent multi-target molecular docking of this cluster against α-amylase, α-glucosidase, and DPP4 prioritized 5S-hydroxynorvaline-S-Ile as a high-confidence multi-target hit. This dipeptide-like compound exhibited favorable multi-target binding profiles across all three enzymes (dG = −6.43, −6.26 and − 4.75 kcal/mol for DPP4, α-amylase, and α-glucosidase, respectively) by forming critical hydrogen bonds with their respective catalytic residues. Docking simulation via AutoDock further validated 5S-hydroxynorvaline-S-Ile as the top-ranked binder among the 27 compounds against the three targets. This work successfully demonstrates a novel, AI-driven paradigm that efficiently navigates complex metabolomic data to discover and prioritize a putative multi-target hit requiring experimental validation, demonstrating this integrated approach as an efficient prioritization framework for future natural product drug discovery.