Dear Hakeem, 

I think you have two alternatives: (1) take the validated model prior to MD and minimize structure then perform docking and virtual screening. (2) take several frames of MD and perform docking and virtual screening on them directly.

good luck

Selection of frames is a tedious job (which one to select/discard). In my openion, the frame which leads to change in conformation must be included.

Docking against homology models increases the uncertainty of results and it seems the MD simulation revealed some problems in your structure. Pay special attention to the binding/active site where you will perform the docking, both during the construction of the model (identity, coverage) and its validation/refinement (is this site involved in the structural changes you observed in the MD?). If there are known active ligands for your system, you may validate your protocol first by docking them along with decoys and analyzing how they rank.

Take procheck for the last confirmation of your MD, if it is giving better results than the modelled one, You can conclude that refined one is better, further you can do PCA based RMSF of Calpha atoms to see further fluctuations. 

Dear Hakeem

I think you should cluster your 20ns trajectory with some RMSD cutoff. This is the best way to get a non-redundant structure(snap shot) from your trajectory. If you select the last frame or any other frame then you might miss the best stable conformation of your model protein. MD simulation just generate random snapshots (conformation) and you have to pick out the best conformation out of the refined structure. Cluster can give you the unique conformation and the selection will be justified, non-biased to carry on further steps.

 Thank you all for your prompt comments.

@ Juan the active site is not involved in the structural changes , that might support the docking.

Experimental studies have provided few inhibitors , based on which i have prepared the library for VS which can done prior to docking and validating the same. !!