When the distance between two adjacent CA atoms (i.e. Ca_i and Ca_i+1) is shorter than 3.6 Angstrom, the two residues may take part in clashes. These are severely unrealistic geometry, however, cis-proline may be an exception usually.
When we use Modeller to build model, the default model routine (automodel class) seems only to check the improper chain breaks (the distance between two adjacent CA atoms is larger than 8 Angstrom) of templates, but not check the CA clashes of generated models. And the models sometimes have the CA clashes problem. How could Modeller avoid such improper geometry? Add some distance restraints by hand?
A typical example: T0345, a target just released of CASP7. Two templates with very high similarities (id% > 65%) are found, 2F8A and 1GP1. Here is my alignment file of 1GP1_A and T0345 -------------- >P1;1GP1A structure:1gp1:10:A:192:A:GLUTATHIONE PEROXIDASE:NA:2.00:0.171 ---RTVYAFSARPLAGGEPFNLSSLRGKVLLIENVASL-GTTVRDYTQMNDLQRRLGPRGLVV LGFPCNQFGHQENAKNEEILNCLKYVRPGGGFEPNFMLFEKCEVNGEKAHPLFAFLREVLPTP SDDATALMTDPKFITWSPVCRNDVSWNFEKFLVGPDGVPVRRYSRRFLTIDIEPDIETLL-*
>P1;T0345 sequence:T0345:1: :185: :T0345: :2.00:-1.00 MIAKSFYDLSAINL-DGEKVDFNTFRGRAVLIENVASLCGTTTRDFTQLNELQCRF-PRRLVV LGFPCNQFGHQENCQNEEILNSLKYVRPGGGYQPTFTLVQKCEVNGQNEHPVFAYLKDKLPYP YDDPFSLMTDPKLIIWSPVRRSDVAWNFEKFLIGPEGEPFRRYSRTFPTINIEPDIKRLLK* --------------- There are two cis-prolines (PRO:97, PRO:150) in each chain of 1GP1, correspondingly, there are two prolines on the same sites of T0345 (PRO:89, PRO:142), and residues around these prolines are very conserved, so I think the prolines in T0345 should be also cis-proline. After building models from this alignment, I have checked the CA clashes of models, and found that there are two clashes: -- pair distance 88 ARG: === 89 PRO: 2.798 A 141 SER: === 142 PRO: 2.803 A -- However there are NO CA clashes occurred in 1GP1. How do the two clashes happen? Are there some means to avoid these clashes?
Best regards,
Yun 2006/06/30
Yun He wrote: > When we use Modeller to build model, the default model routine (automodel > class) seems only to check the improper chain breaks (the distance between > two adjacent CA atoms is larger than 8 Angstrom) of templates, but not check > the CA clashes of generated models. And the models sometimes have the CA > clashes problem. How could Modeller avoid such improper geometry? Add some > distance restraints by hand?
The test you describe is alignment.check(), and is a very simple test, so can be run before modeling. If you want to detect bad models (e.g. CA clashes) after the modeling is complete, then that's the realm of model assessment such as GA341 or DOPE. Generally speaking, CA clashes should be prevented by the stereochemical restraints (CHARMM force field) although your template distance restraints may clash with these. You may need to weaken the homology-derived restraints in these regions, or remove them entirely by deliberately misaligning your templates to the target.
Ben Webb, Modeller Caretaker
-
Modeller Caretaker -
Yun He