Fig. 5: Workflow of the hierarchal template-based complex structure modeling approach.
From: Augmented BindingNet dataset for enhanced ligand binding pose predictions using deep learning
The MCS coverage was calculated between candidate compounds from ChEMBL and template ligands from PDB. If the largest MCS coverage exceeds or equals to 0.6, we assigned the atom coordinates of the MCS as those in the template ligand. Otherwise, the compound was fragmentated and aligned to templates with SHAFTS. The rest part of the compound was sampled, clustered, filtered, and scored, with 20 highest-scored conformations for MM-GB/SA minimization. Hybrid scores were determined again for minimized complex structures and templates, and the highest-scored complex was selected as the final model. Images of 3D structures are rendered with CHIMERA47 and image of 2D structure is generated with RDKit51.
