You are here

Score function in protein-protein docking with constrains

3 posts / 0 new
Last post
Score function in protein-protein docking with constrains

Hi, everyone!

I am performing protein-protein docking with AtomPairConstaint (the experiment shows the distance between two ligands) and I have some questions:

a. Whether the ref2015  or ref2015_cst function considers the ligand (with params file)?

b. If I use ref2015_cst in the docking_protocol, does the score function consider the constraint weight in both low- and high-resolution stages?

c. Is the ref2015 function default weights suitable for scoring the docking structure as the scoring tutorial use docking weights?


Post Situation: 
Mon, 2023-02-06 03:31
Zehui Zhou

Both ref2015 and ref2015_cst should properly consider the ligand -- the Lennard Jones, electrostatic, solvation, hydrogen bond terms, etc. should be active so long as the ligand is properly understood from the params file.

ref2015_cst is specifically a high-resolution ("fullatom") scorefunction. Low-resolution protein-protein docking typically happens in "centroid mode", which requires a scorefunction which is properly set up for centroid mode, which typically means a completely reparameterized scorefunction. You should be able to add the atom pair constraint term to this scorefunction as well.

Note that you have to be a little bit careful about constraints in centroid mode, as not all atoms are present. (In particular, sidechain atoms are replaced with a single "centroid" atom.) This often means altering the constraint form slightly to refer to backbone heavyatoms for proteins. For non-protein ligands, the approach is to represent them as a "united atom" model which contains all the heavy atoms, but without the apolar hydrogens. If you have constraints to ligand heavy atoms, you're probably okay. (Note that you may or may not need to add an additional centroid-mode params file with -extra_res_cen ... there's some facility to auto-convert internally between the fullatom representation to the centroid one.

The different weights files are parameterized to optimize certain things differently. The docking weights file was specifically parameterized to function well in protein-protein docking, though it might not be as useful in other situations. In contrast, the (more recent) ref2015 scorefunction is parameterized to do well on a range of tasks. The concept is that the more general ref2015 scorefunction is more generalizable because it isn't specifically optimized for the subset of protein-protein complexes which the docking scorefunction has been parameterized to. I believe the general recommendation at this point would be to use ref2015 for protein-protein docking. However, if you're particularly interested, I would encourage you to run a benchmark set (positive controls) for proteins/situations with known results that are similar to your experimental conditions of interest. You can compare the different scorefunctions and see if the docking scorefunction does better in this particular sub-task. (That said, if that's a bit much, I wouldn't feel bad about just using ref2015.)

Mon, 2023-02-06 14:27

Thanks for your detailed answer and valuable insights.

As you mentioned, the scorefunction_cst is used in both low- and high-resolution modes. I'm wondering if adding AtomicDistance to the MOVERS would be a better way to achieve the desired distance between proteins and ligands. This constraint would modify the overall energy surface, which could lead to a biased docking result. For example, if the distance falls outside the range of x0 - tol to x0 + tol during the Initial Perturbation, the second partner will be pulled back into range by the constraint, rather than by the "biophysical theory of an encounter complex formation". This could increase the number of conformations that fall on the edge of the range.


Wed, 2023-02-15 05:40
Zehui Zhou