I am using Gold to dock the Maybridge fragment library into the PDE4b target. I use Chemscore for docking and ASP for rescoring (as suggested by the program template for phosphodiesterases). I have decided to begin with a fast screening. My question is how can I use the scores of the initial screening to filter out some compounds and redock the selected ones in a slower more accurate way?
Dear Anastasia Patsiarika
You asked a good question
There are many ways to reduce the number of compounds. One of the most important ways is to:
If you have a lot of compounds, it is better to first filter the compounds based on Lipinski's rule to reduce the number of compounds.
You can use Open Babel or DataWarrior or Discovery studio or schrodinger for this purpose
Lipinski's rule are:
▪ there are more than 5 H-bond donors (nHDon)
▪ there are more than 10 H-bond acceptors (N + O)
▪ molecular weight (MW) is over 500
▪ Moriguchi's logP (MLogP) is over 4.15
http://www.openmolecules.org/datawarrior/download.html
http://openbabel.org/wiki/Main_Page
Milad's suggestion is great. However, goldmine which is part of the Gold suite is designed for analysing, post-processing Gold docking results. It helps to do all sort of filtering too.
Of course, you can go with the drug-likeness (Lipinski's rule of five and Veber rule) and ADMET filtering. Moreover, you can set a threshold for the docking score based on the docking score of the known drugs or inhibitors against the particular target that you have selected.
Thank you for your interesting answers! I have decided to combine your suggestions. Also I found a free online tool for library preprocessing and filtering, it is called FAFDrugs4 in case someone is interested
I would not suggest filtering your dataset based on Lipinski's rule of five. This rule has been prevalent for two decades now, and recent research/review has shown that a significant percentage of FDA-approved drugs in fact do not satisfy it. (https://www.nature.com/articles/nrd.2018.197). By using these criteria to filter your dataset early on in the virtual screening pipeline, you run the risk of losing out on a lot of potentially good molecules. It would be better to filter out your dataset based on reactive or toxic functional groups, and after an initial less accurate fast screening stage, decide on a cutoff docking score to take the moelcules above it forward. You could look up previous studies to see what kind of cutoffs have been used in general. I haven't personally used GOLD so wouldn't like to comment on that, but for Autodock or Glide I would go for ligands scoring above -7 Kcal/mol or depending on the size of original dataset and the output of the first stage, a fixed percentage like 10, 25, 50, 70% etc of the top scoring hits to move to the next stage. You also have to take into account the computational power and time at your disposal.
- Badges
- Science topic