Structure with high clashscore

Hello all, I'm happy to join the forum! I have a problem with clashscores and would be grateful for support. I have diffraction data for a 3HP-CoA Synthetase from N. maritimus resolved to 2.7 Angstroms. For MD, I used an alphafold structure (as I understand commonly results in high clashscore values) and after numerous rounds of refinement, the clashscore remains high at 11. I recently tried using gromacs to energy minimize and got values of 5 and 3 (according to molprobity), but after refinement these values immediately jumped back up to ~30. I wondered if you could describe the best way to reduce these? Attached is my latest round of refinement should you wish to take a look! Best, Jerome

Dear Jerome,
some ideas to check:
- MD: you probably mean MR (molecular replacement)
- before starting refinement, if you found a solution by MR, did you look at the structure, crystal packing etc., no big clashes between molecules?
- related with that, did you validate the space group? Run several MR's with different space groups (which make sense from indexing) and compare the results
- refinement: run simulated annealing (initially with rather high temperatures, later in refinements at lower temperature), this helps to get out of local energy minima; energy minimization is done as part of that also
- include hydrogens in the refinement, this reduces clashes throughout the structure (they are added as riding hydrogens, i.e. predicted from geometry, they don't need to be resolved in the map for that)
- check your atomic model and do model building, e.g. in Coot, use Ramachandran restraints etc., go through the structure and refine manually, reinject that model into Phenix for further refinement; iterate; check whether your map improves
- run composite omit or polder maps if you have doubts about some regions, e.g. to figure out a loop conformation etc.
- is the diffraction isotropic? i.e. does the crystal diffract equally well in all directions? anisotropic diffraction can make refinement more difficult, but it can be corrected for
- use the phenix.predict_and_build tool to improve your AlphaFold prediction and refine your atomic model, see https://phenix-online.org/documentation/reference/alphafold.html
Best,
Bruno
-----Original Message-----
From: JEROME JOHNSON via phenixbb

Hi, one more possibility - we find that if the structure is refined without hydrogens, but one or more ligands do have hydrogens (not intentionally), then Molprobity adds hydrogens to all residues and clashscore calculated by it (now including hydrogens) jumps a lot. The remedy would be to delete all hydrogens from the structure before the refinement (or refine with hydrogens added everywhere). Best Leonid JEROME JOHNSON wrote:
Hello all,
I'm happy to join the forum! I have a problem with clashscores and would be grateful for support.
I have diffraction data for a 3HP-CoA Synthetase from N. maritimus resolved to 2.7 Angstroms. For MD, I used an alphafold structure (as I understand commonly results in high clashscore values) and after numerous rounds of refinement, the clashscore remains high at 11. I recently tried using gromacs to energy minimize and got values of 5 and 3 (according to molprobity), but after refinement these values immediately jumped back up to ~30.
I wondered if you could describe the best way to reduce these?
Attached is my latest round of refinement should you wish to take a look!
Best, Jerome

Hi All,
Hi, one more possibility - we find that if the structure is refined without hydrogens, but one or more ligands do have hydrogens (not intentionally), then Molprobity adds hydrogens to all residues and clashscore calculated by it (now including hydrogens) jumps a lot. The remedy would be to delete all hydrogens from the structure before the refinement (or refine with hydrogens added everywhere).
this is an interesting observation. MolProbity always adds all H to compute clashscore -- one can't compute clashscore without H present. I suspect having H on ligands only triggers it to 'think' that H are already present, and it computes clashscore without adding H to everything else. Could you please send me the file off list (the one that has H only on the ligand)? I think we better investigate this..
I'm happy to join the forum! I have a problem with clashscores and would be grateful for support.
I have diffraction data for a 3HP-CoA Synthetase from N. maritimus resolved to 2.7 Angstroms. For MD, I used an alphafold structure (as I understand commonly results in high clashscore values) and after numerous rounds of refinement, the clashscore remains high at 11.
11 looks just fine. Note, clashscore=0 is not the goal because clashscore does not account for non-standard links and for example, can count as a severe clash atoms involved into nonstandard links, and similar. In Phenix Comprehensive Validation, go over the table of clashes, they are sorted, worst first, and check and try to address top few worst offenders. Pavel
participants (4)
-
Bruno KLAHOLZ
-
JEROME JOHNSON
-
Pavel Afonine
-
sazanov@ist.ac.at