You are here

How to select docked models

2 posts / 0 new
Last post
How to select docked models
#1

Dear all, 

I have generated 500 docked complex model with docking_protocol.linuxgccrelease using a command like this:

docking_protocol.linuxgccrelease\
  -s start_prepacked.pdb\
  -partners A_B\
  -dock_pert 3 8\
  -ensemble1 prot1_ensemble.list\
  -ensemble2 prot2_ensemble.list\
  -out:path:all docking\
  -nstruct 500 | tee docking.log

I wish to select 10 out of the 500 models for next step such as local refining. However, I am confused about the selection criteria. Accroding to the docking protocol available on net (https://www.rosettacommons.org/docs/latest/application_documentation/docking/docking-protocol), two scores can be referred to: the "total" (column 2) and "I_sc" (column 6). And the "I_sc" should be more relevent. So I run this command:

    sort -n -k6 -k2 score.sc | awk 'NR<=10{print $2, $6, $40}'

and get: 

  total         I_sc              model  

  -512.410    -19.416    start_prepacked_0314
   935.903    -17.147    start_prepacked_0176
   368.028    -16.874    start_prepacked_0490
  -469.329    -16.648    start_prepacked_0256
  -712.922    -16.225    start_prepacked_0013
  -312.844    -15.919    start_prepacked_0263
  -453.569    -15.693    start_prepacked_0389
  500.628     -15.370    start_prepacked_0096
  -626.392    -15.345    start_prepacked_0068
  -641.953    -15.313    start_prepacked_0280

If only considering the I_sc, it seems that I should pick up these 10 models, but some models in these list, such as the second and the third models, have very high positive total_scores. Should I just rule them out or keep them for local refining? Would local refinement lower the total score? 

With many thanks
 

Category: 
Post Situation: 
Mon, 2020-06-22 00:50
Sunyp_IM

Hello,

The recommended way to select models is to sort them based on I_sc and pick the top 10 lowest scoring models. Total score is not a good discriminator for docking. If you have sampled enough, you might find the top 10 models are very similar. In that case, you could use the clustering application to cluster the top 200 by I_sc and then pick 10 clusters with either the lowest I_sc or most members.

Best,

Shourya

Mon, 2020-06-22 13:12
ssrb