Computer-aided drug design (CADD) is a critical initiating step of drug development, but a single model capable
of covering all designing aspects remains to be elucidated. Hence, we developed a drug design modeling framework
that integrates multiple approaches, including machine learning based quantitative structure-activity relationship (QSAR)
analysis, 3D-QSAR, Bayesian network, pharmacophore modeling, and structure-based docking algorithm. Restrictions for
each model were defined for improved individual and overall accuracy. An integration method was applied to join the results
from each model to minimize bias and errors. In addition, the integrated model adopts both static and dynamic analysis
to validate the intermolecular stabilities of the receptor-ligand conformation. The proposed protocol was applied to
identifying HER2 inhibitors from traditional Chinese medicine (TCM) as an example for validating our new protocol.
Eight potent leads were identified from six TCM sources. A joint validation system comprised of comparative molecular
field analysis, comparative molecular similarity indices analysis, and molecular dynamics simulation further characterized
the candidates into three potential binding conformations and validated the binding stability of each protein-ligand complex.
The ligand pathway was also performed to predict the ligand “in” and “exit” from the binding site. In summary, we
propose a novel systematic CADD methodology for the identification, analysis, and characterization of drug-like candidates.