ΔΔG Calculation of a Mutation with PyRosetta

As for the ddG prediction being closer to the experimentally determined value with -beta_nov16, I think you answered your own question here…

Using the -beta_nov16 scoring function with Cartesian refinement, we observed a slighlty closer ΔΔG prediction to experimental value (-4.29 kcal/mol) compared with the reported value of -4.95 kcal/mol with only the first run of the Python code (i.e. without the -ddg:mut_file flag), so would this function be a better predictor of ΔΔG of mutation for other proteins as well?

Also, with the optimized parameters of the -beta_nov16_cart function, wouldn’t it be more efficient in sampling the global minimum as compared with the -beta_nov15 scoring function with Cartesian refinement utilized by Alford et al. (https://www.biorxiv.org/content/biorxiv/suppl/2017/02/07/106054.DC1/106054-1.pdf)?

You are asking a question here that we don’t know the answer to.  Perhaps try it with a few other proteins?  Benchmark.  You could even contact Rebecca to collaborate perhaps.  

Again, the beta score function has not been published yet. It is unclear if it is quicker or better.  Without data how do we know?  Perhaps the creator of the scorefunction could tell you – but being a developer of Rosetta, I have yet to hear much about it. Perhaps I missed a paper?  I don’t know, but personally I’ve just been using REF2015 as it was heavily benchmarked for many different tasks in Rosetta.     

Hi,

May I ask which platform did you use? PyRosetta or Rosetta? I have some confusion while dealing with the ddG calculation in PyRosetta

Best

Me? PyRosetta, but for a different ball game: speed.

My benchmarking was for assessing ∆∆G calculations on unminimised models with a strong time constraint (> 60 s) for my app, venus.cmd.ox.ac.uk: for this a neighbourhood of 8–15Å was FastRelax minimised for 1-3 cycles before and after the change. Cartesian was too slow and the improvement negligible, so I am using vanilla ref2015 scaled by some factor.

My problems with the ∆∆G calculations were destabilising outliers that caused backbone distortions, especially on a different segment of the protein so I resorted to using median instead of mean for the “mean absolute error” metric and all the data made sense —if a mutation is destabilising, I don’t care if it’s +10 kcal/mol or +100 kcal/mol, but it throws off the mean. Machine learning ∆∆G calculators don’t have hard-to-calculate outlier problems hence why papers comparing methods seem to disagree and handwave it away on different subsets of the ProTherm database.

Hi matteoferla,

Thank you so much for your reply! I tried your app VENUS and it is so efficient and quick to give feedback on ∆∆G for a specific mutation, so cool!

I am new to pyrosetta and so sorry that I have no idea about your problem. BTW, may I ask you a question related to the PyRosetta ∆∆G calculation? 

Following the cartesian_∆∆G Protocol in Rosetta, which is claimed to have a Pearson correlation of 0.74, I tried to transform the calculation into the PyRosetta version. My logic is stated in the following:

1.  Prepare a cleaned PDB file and apply cartesian fast_relax with the function of ref2015_cart, record the score in function of ref2015 and denote it as S0
2.  Mutate the pose with the function mutate_residue, then apply MinMover and Monte Carlo trial mover to minimize the energy, denote the score at this stage as S1
3.  Calculate the ∆∆G as S1-S0.

   
   
   from pyrosetta import *
   
   from pyrosetta.rosetta import *
   
   from pyrosetta.toolbox import *
   
   from pyrosetta.teaching import *
   
   init("-ignore_unrecognized_res 1 -ex1 -ex2 -flip_HNQ -relax:cartesian -nstruct 200 -crystal_refine -optimization:default_max_cycles 200")
   
   
   
   
   
   #Parameter set up#
   
   testPose= Pose()
   
   testPose = pose_from_pdb("1BNI.clean.pdb")
   
   scorefxnDDG=get_fa_scorefxn()
   
   
   
   #Firstly relax the structure for later modificiation#
   
   from pyrosetta.rosetta.protocols.relax import FastRelax
   
   scorefxnRelax = pyrosetta.create_score_function("ref2015_cart")
   
   relax = pyrosetta.rosetta.protocols.relax.FastRelax()
   
   relax.constrain_relax_to_start_coords(True)
   
   relax.coord_constrain_sidechains(True)
   
   relax.ramp_down_constraints(False)
   
   relax.cartesian(True)
   
   relax.min_type("dfpmin_armijo_nonmonotone")
   
   relax.min_type("lbfgs_armijo_nonmonotone")#for non-Cartesian scorefunctions use'dfpmin_armijo_nonmonotone'
   
   relax.set_scorefxn(scorefxnRelax)
   
   
   
   
   
   
   
   relax.apply(testPose)
   
   s0=scorefxnDDG(testPose) #Record the energy score after cartesian_relax
   
   
   
   #Energy minimization
   
   min_mover = MinMover() #define a Mover in type of MinMover
   
   mm=MoveMap()
   
   mm.set_bb_true_range(28,36)
   
   min_mover.movemap(mm)
   
   min_mover.score_function(scorefxnDDG)
   
   #min_mover.min_type("dfpmin")
   
   min_mover.tolerance(0.01)
   
   print(min_mover)
   
   ddG=[]
   
   def minimize_Energy(pose):
   
   #Minimization#
   
   min_mover.apply(pose)
   
   
   
   #Trial_mover define#
   
   kT=1
   
   mc=MonteCarlo(pose,scorefxnDDG,kT)
   
   mc.boltzmann(pose)
   
   mc.recover_low(pose)
   
   
   
   trial_mover = TrialMover(min_mover,mc)
   
   #Monte Carlo#
   
   for i in range (100):
   
   trial_mover.apply(pose)
   
   
   
   return
   
   
   
   #Point mutation#
   
   AA=
   
   mp=Pose()
   
   for i in AA:
   
   mp.assign(testPose)
   
   mutate_residue(mp,52,i)
   
   relax.apply(mp)#relax after minimization
   
   minimize_Energy(mp)
   
   
   
   s1=scorefxnDDG(mp)
   
   dg=s1-s0
   
   ddG.append(dg)
   
   
   
   
   
   
   
   
   
   #Output#
   
   
   
   print("The energy after relax: ",s0)
   
   print("The ddG: ")
   
   print(ddG)
   
   
   
   

    

However, I found that according to this logic, the ∆∆G I got are always a positive value (I fixed the residue position and perform all 21 types of mutation). Do you know which steps in my protocol are wrong, or may you give me some suggestions to calculate the ∆∆G in PyRosetta?

Thank you so much for your help! 

Yours sincerely