Autor: |
Gumbis, Gediminas, Houston, Douglas R |
Jazyk: |
angličtina |
Rok vydání: |
2021 |
Zdroj: |
Gumbis, G & Houston, D R 2021, ' Using semiempirical quantum mechanical methods to accelerate virtual screening and medicine discovery ', University of Edinburgh . |
Popis: |
In order to identify lead compounds for drug targets, virtual screening offers a costefficient alternative to experimental high throughput screening. Molecular docking isan important first step of the structure-based virtual screening process, wherecompounds are attempted to be positioned in the protein binding site in exactly thecorrect position and orientation that they would adopt naturally, to enable accurateprediction of binding affinity. The charges on ligand atoms and surrounding proteinatoms are believed to play a key role in a successful docking as electrostatic energygrids identify positions where ligand and protein charges would be complementary.Semiempirical quantum mechanical (SEQM) charge calculation methods capture thepolarisation present in a protein-ligand system and have been shown to give moreaccurate molecular docking than the default Gasteiger charges. However previoussample sizes were small and students of my supervisor saw no improvement withmedium sized datasets. This project looked to test whether SEQM calculatedcharges using the PM6-D3H4 Hamiltonian gave more accurate docking thanGasteiger charges when docking with the program DOCK6, after an initial dockingwith GWO Vina, on a large dataset of 1874 complexes.Surprisingly it was found that Gasteiger charges performed slightly but significantlybetter than SEQMs and that SEQMs led to a small decrease in docking accuracycompared to the primary Vina pose. 1170 Vina poses were docked within 2 Å of thecrystallographic ligand pose, compared to 1047 Gasteiger-DOCK6 poses and 962SEQM-DOCK6 poses. The SEQM approach works well for some complexes butdecision-tree based models using 1753 BINANA, ECIF and ligand descriptors wereunable to predict this differential performance with any reliability. A macroaveragedf1 score of only 0.51 was achieved in a binary classifier predicting whether SEQMsor Gasteiger charges gave more accurate docking. |
Databáze: |
OpenAIRE |
Externí odkaz: |
|