Everything You Want to Know About Pharmacophore Model Construction
SummaryMolecular docking technology remains the most popular structure-based drug design method, which takes full advantage of protein-ligand interaction information. However, in virtual screening, compared with molecular docking, pharmacophore-based methods have obvious advantages in terms of computational cost and accuracy, and docking-based virtual screening methods show a higher false positive rate.
- Author Name: Janet Smith
- Author Website: https://www.computabio.com/
Molecular docking technology remains the most popular structure-based drug design method, which takes full advantage of protein-ligand interaction information. However, in virtual screening, compared with molecular docking, pharmacophore-based methods have obvious advantages in terms of computational cost and accuracy, and docking-based virtual screening methods show a higher false positive rate. Therefore, combining complementary ligand-based pharmacophore model methods and receptor-based methods helps to improve the reliability of the method. Studies have shown that in virtual screening, receptor-based pharmacophore models are very successful in the discovery of new active molecules. In addition, the pharmacophore is also used in the molecular docking procedure to better distinguish the wrong conformation from the correct conformation, thereby improving the success rate of molecular docking.
The following takes the virtual screening process of ABL tyrosine kinase inhibitors as an example to briefly describe the construction process of the receptor-based pharmacophore model and its application in virtual screening. STI-571 is a classic tyrosine kinase inhibitor for the treatment of chronic myeloid leukemia. Among different tyrosine kinases (ABL, c-KIT, SYK), the compound exhibits different binding conformations.
- Pharmacophore model construction
Wolber et al. created multiple pharmacophore models based on the protein crystal structures of STI-571 and its homologs (1iep, 1fpu, 1opj). The pharmacophore models were then combined using LigandScout software. The finally obtained pharmacophore model contains 4 hydrophobic aromatic groups, 2 aliphatic hydrophobic groups, 2 hydrogen bond donor groups, 1 hydrogen bond acceptor group, and 8 repulsive volume groups.
- Validation of the pharmacophore model
The PDB database was screened based on the pharmacophore model. The database consists of 2765 drug-like compounds. Screening of the PDB database showed that this pharmacophore was able to find all STI-571 structures without false positive results.
- Pharmacophore-based virtual screening
By screening the Maybridge database, which consists of 590 million compounds. Finally, seven compounds with ABL tyrosine kinase activity were obtained, which can be used as lead compounds for new ABL tyrosine kinase inhibitors.
Structure-based pharmacophore models can be constructed for proteins with existing crystal structures. For proteins with unknown crystal structures, a pseudo-receptor can be constructed based on the ligand pharmacophore model. Höltje et al. successfully used virtual receptor technology to construct multiple proteins without protein crystal structures. For example, the pharmacophore of 5-HT2A was constructed using 20 5-HT2A inhibitors with different backbones. Subsequently, based on the low-resolution bacteriorhodopsin protein (5-HT2A homologous membrane protein), amino acids were placed around the ligand, and finally a virtual receptor model of 5-HT2A was obtained. A predictive QSAR model was finally obtained through this virtual receptor and used to guide the synthesis of new inhibitors.
The commonly used software for constructing virtual receptor models are Yak and PrGen. In these two software, the process of constructing virtual receptors is:
- Calculate the pharmacophore element of the ligand
- Place amino acid conformations around ligands;
- Minimize the energy of the virtual receptor while ensuring that the position, orientation and conformation of the ligand do not change;
- The correlation-coupled minimization is performed by introducing a coupling constant to ensure a good correlation between the experimental affinity and the calculated interaction energy;
The conformation of the receptor is then fixed and the ligand is optimized for conformation, which is repeated multiple times until a highly correlated virtual receptor model is obtained. To verify the reliability of the virtual receptor model, a leave-one-out (LOO) or external test set of molecules was used to evaluate the predictive performance of the virtual receptor.
Another widely used virtual receptor generation software is Flo. The process of constructing virtual receptors in this method is as follows:
- Generate low-energy conformational ensembles of training set molecules;
- Optimize the conformation to achieve maximum matching of functionally similar groups in the molecule while reducing the internal energy of the compound;
- The compound is then surrounded with groups that mimic the properties of protein pockets, such as guanidine groups to simulate real groups that hydrogen bond with acidic groups in ligand molecules, and propane molecules to simulate hydrophobic interaction pockets in the binding pocket;
- Equilibrate virtual receptors using a kinetic simulation approach.
Virtual receptor models were used to analyze key ligand-protein interaction sites and to construct 3D-QSAR models to predict ligand biological activity. A large number of applied studies have pointed out that the virtual receptor-based 3D-QSAR model has higher accuracy in predicting the biological activity of compounds compared with the classical 3D-QSAR method (CoMFA).