Computational protocol: Combinatorial Pharmacophore-Based 3D-QSAR Analysis and Virtual Screening of FGFR1 Inhibitors

Similar protocols

Protocol publication

[…] Pharmacophore hypotheses were generated using the Phase 3.1 implemented in the Maestro 9.0 software package (Schrödinger, LLC). Phase translates the ligands into bit strings by applying a tree-based partitioning algorithm and distinguishes multiple binding modes using a bi-directional clustering approach []. Since pharmacophore modeling requires all-atom 3D structures to represent the active form of the inhibitor, it is crucial to consider a variety range of conformations so as to increase the possibility of finding the one close to the natural bound structure. Training ligands were firstly cleaned up to ensure the structures are in 3D, and the count ions and water molecules are excluded. Additionally, parameters were set to retain specified chiralities, generate a maximum of 32 stereoisomers and ionize at target pH 7.4. Once the clean-up ligand structures were generated, a conformational search was carried out using the ConfGen module of Maestro with default parameters, to generate a set of conformers for each structure. Potential energy calculation was carried out using the OPLS_2005 force field. An RMSD cutoff of 1.00 Å was used for eliminating redundant conformations. [...] The combinatorial QSAR model was firstly validated by an external test set with known activities. To further evaluate our model for virtual screening, namely, the ability of the combinatorial QSAR model in assigning high ranks to the known actives, a decoy dataset composed of 7665 compounds was collected from the SPECS database (http://www.specs.net) in a proportion of 36:1 to the test set, using DecoyFinder []. Of note here is that DecoyFinder may introduce potential biases for the ligand-based virtual screening, and it should be compared with other structure-based approaches with caution [,]. Compounds in the decoy dataset were also divided into 8 groups based on the similarity comparison described above. Each structure was subjected to ligand preparation using Ligprep and conformations generation using the ConfGen. The corresponding QSAR model was used to predict the activity using the “find matches to hypothesis” option, which finds matches from a database to a selected hypothesis and calculates activity if the hypothesis has a QSAR model. We then calculated enrichment factors (EF) of the decoy set to evaluate the ability of identifying actives from inactives. EF is calculated by Equation (1) []: (1)EF=HitssNsHitstNt where Hitss is the number of actives in the selected front fraction of the ranked list, Hitst is the quantities of actives in database, Ns is the total number of compounds in the selected fraction of the database and Nt is the quantities of compounds in database. […]

Pipeline specifications

Software tools ConfGen, Decoyfinder, LigPrep
Applications Drug design, Protein structure analysis
Diseases Neoplasms, Tooth Migration