Unexpected Results with kstar.py

[amin-sagar]

Dear Osprey developers and users,
I am trying my first calculations with Osprey and I am having some problems.
I thought of doing a simple experiment. I took a structure of a protein-peptide complex where the tyrosine at position 10 in the peptide makes 2 hydrogen bonds with the protein and mutated it to Alanine. Then I used the kstar.py script that is the part of osprey 3.2.304 release and wanted to see if I can recover Tyrosine at this position. The only change I made to the script is in the following section

for res in ['A10']:
    protein.flexibility[res].setLibraryRotamers(osprey.WILD_TYPE, 'VAL','TYR','TRP','GLY').addWildTypeRotamers().setContinuous()

and

# configure K*
kstar = osprey.KStar(
    proteinConfSpace,
    ligandConfSpace,
    complexConfSpace,
    epsilon=0.1, # lowered to 0.1
    writeSequencesToConsole=True,
    writeSequencesToFile='kstar.results.tsv'

The output is as follows

Seq ID	Sequence	K* Lower Bound	K* Upper Bound	K* Lower (log10)	K* Upper (log10)	Total # Confs.	Complex Pfunc Lower	Complex Pfunc Upper	Complex Free Energy Lower	Complex Free Energy Upper	Complex Epsilon	Complex # Confs.	Protein Pfunc Lower	Protein Pfunc Upper	Protein Free Energy Lower	Protein Free Energy Upper	Protein Epsilon	Protein # Confs.	Ligand Pfund Lower	Ligand Pfund Upper	Ligand Free Energy Lower	Ligand Free Energy Upper	Ligand Epsilon	Ligand # Confs.	Time (sec)
0	A10=ala	19873270	20062400	7.29826940334862	7.30238279496535	21	1.053005E+050	1.058498E+050	-68.3111386962339	-68.3080535054792	0.00518873424026019	10	87.96891	87.96891	-2.65507579520591	-2.65507579520591	0	2	5.997605E+040	6.023265E+040	-55.686837097797	-55.6843052529257	0.0042600917046854	9	18
1	A10=VAL	5.030577E-90	5.05604E-90	-89.2983821690412	-89.2961895289745	21	3.809074E-47	3.812045E-47	63.3871786491476	63.3876409605632	0.000779244813620255	9	125.71	125.71	-2.86679102592102	-2.86679102592102	0	3	5.997605E+040	6.023265E+040	-55.686837097797	-55.6843052529257	0.0042600917046854	9	22
2	A10=TYR	3.32008	3.679395	0.521148520882051	0.565776403017213	413	1.583778E+044	1.747705E+044	-60.4152346301624	-60.3568249388948	0.0937955212204782	396	791.9794	791.9794	-3.95833427611541	-3.95833427611541	0	8	5.997605E+040	6.023265E+040	-55.686837097797	-55.6843052529257	0.0042600917046854	9	153
3	A10=TRP	2279.949	2408.665	3.35792504673286	3.38177632960835	55	3.244213E+047	3.412766E+047	-64.9087608488039	-64.8787226108531	0.049389069792382	39	2362.394	2362.394	-4.60647610537634	-4.60647610537634	0	7	5.997605E+040	6.023265E+040	-55.686837097797	-55.6843052529257	0.0042600917046854	9	166
4	A10=GLY	1338317	1344700	6.12655884527996	6.12862537584528	19	2.927157E+048	2.928589E+048	-66.1835711936372	-66.1832810908582	0.000489051100186772	9	36.31242	36.31242	-2.13033093964937	-2.13033093964937	0	1	5.997605E+040	6.023265E+040	-55.686837097797	-55.6843052529257	0.0042600917046854	9	170

I have tried both with all the protein residues fixed and continuous flexibility for four residues of the protein (defined as the ligand).
I have two problems.

1. In neither case, I get the best K* score for Tyrosine.
2. The output PDBs for all but the last mutation are empty, irrespective of the score.
   They all look like this.

EMARK   3
REMARK   3 Generated by OSPREY 3.2.304
REMARK   3
REMARK   3 COMMENT : Top 10 conformations for sequence A10=VAL
REMARK   3
END

I have attached the results and the python script as well(as a text file).
I would be really grateful for any suggestions about what modifications can be done to recover the expected residue at this position and solve the problem with only the last tried mutation being saved as pdb.
Best,
Amin.

kstar_results.ods
kstar_txt.txt

[amin-sagar]

Update.
I installed the new 4.0.2 version.
I tried using the exact script on the original structure which has Tyrosine at position 10.
Now, Tyr has a better score than Alanine.

sequence: A10=tyr
	K* score: 18.710770 in [18.710287,18.718765] (log10)
Residues           A9    A10   A11   A12   B143  B146  B212  B213 complex Sequence   tyr  
Ensemble of 0 conformations:

result:
	sequence: A10=ALA
	K* score: 13.386076 in [13.386059,13.387858] (log10)
Residues           A9    A10   A11   A12   B143  B146  B212  B213 complex Sequence   ALA  
Ensemble of 0 conformations:

However, If I first mutate to Alanine, I get the following result.

sequence: A10=ala
	K* score: 13.321488 in [13.321471,13.326351] (log10)
Residues           A9    A10   A11   A12   B143  B146  B212  B213 complex Sequence   ala  
Ensemble of 0 conformations:

result:
	sequence: A10=TYR
	K* score: 6.536769  in [6.536392 , 6.580573] (log10)
Residues           A9    A10   A11   A12   B143  B146  B212  B213 complex Sequence   TYR  
Ensemble of 0 conformations:

This seems to suggest that the scoring function does recognize Tyr to be better. But the sampling doesn't find the right best scoring conformation.
I would be really grateful for any advice regarding this issue.
Best,
Amin.

[gusennan]

Hi @amin-sagar,

thanks for the question and including your script. Could you please attach your structure?

To your question:

I would be really grateful for any suggestions about what modifications can be done to recover the expected residue at this position and solve the problem with only the last tried mutation being saved as pdb.

Failure to save the low-energy ensemble is an issue we are working on, see #188

[aminsagar]

Hi @gusennan
Thanks for your response.
Unfortunately, I can't share the structure I am currently working on as it belongs to someone else.
However, I will try my best to reproduce the issue on another structure that I can share.
If there is something you would like me to check with the structure I am currently working on, I can do that and report the results.
Best,
Amin.

[amin-sagar]

I tried to reproduce the error with another structure but I couldn't. Osprey was able to find the critical residue. I also confirmed by looking at the structure saved by Osprey that the critical interactions are formed. I had to make sure that this was the last mutation in the list in order to do that.
So, I did a more extensive minimization and equilibration of my first structure, where I had seen the error. Now, Osprey is able to find the correct residue. Therefore, I think the problem was some clashes in my structure which were not resolved by minimization.
The only problem now is saving the structures of the tried mutations, but I think this issue can be closed.