Integration of LBDD and SBDD Studies on Drug Design: A Fatty Acid Amide Hydrolase (FAAH) Case Study

The inhibition of the fatty acid amide hydrolase (FAAH), an endocannabinoid system component, emerged as a potentially new therapeutic target for a range of clinical disorders such as acute and chronic pain. Some α-ketoheterocycle derivatives demonstrated interesting analgesic and anti-inflammatory activities in vitro. Ligand-Based Drug Design techniques such as knowledge graph convolutional networks (kGCN) and hologram quantitative structure-activity relationship (HQSAR) using α-ketoheterocycle derivatives from five different datasets were generated to discover the relation between the chemical structures and the inhibition activity. Meanwhile, structure-based drug design simulations as interaction fields (MIF), molecular docking, and ligand sites studies (LSI) from FAAH were performed using Autogrid software and FTmap/FTsite servers. The results of both studies were merged to propose predictive models. The resulting kGCN model demonstrated adequate accuracy area under the curve by receiver operating characteristic (AUC-ROC 0.7922). From contribution maps of the Ligand-Based Drug Design (LBDD) models and the generated probes using MIF and LSI, it was observed that the oxazole ring, the ketone group, and the apolar chain present in the structures of the inhibitors are important, besides the evidence of the Cys269 and Val270 residues importance for the potential interaction, confirmed by carried docking studies. These fragments and structural information can be used to carry out new FAAH potential inhibitors studies and report kGCN as an accurate classification technique.

Keywords:
FAAH; KGCN; graph-based classification models; neural networks; HQSAR; molecular