Homology Model Validation

Question

[Zak]

Dear all, i have modelled a protein structure using Modeller and performed loop refinement,but i am having problem validating my structure through SAVES, Procheck's

Ramachandran plot shows warning sign (plot shows that 95% residues lie in core region,4.7 in allowed,.02 in gener and 0 in disall.) Chi1-chi2 plots: 5 labelled residues (out of 292)

residue property shows Error (Residue properties: Max.deviation: 3.4 Bad contacts: 5 * Bond len/angle: 5.0 Morris et al class: 1 1 2 * 4 cis-peptides G-factors Dihedrals: 0.03 Covalent: -0.11 Overall: -0.02 )

Rest of the procheck fields are (Green)ok.

Verify3D shows 76.10% of the residues had an averaged 3D-1D score > 0.2 (Warning) Errat Overall quality factor 79.087.

I tried energy minimization of the structure using Spdb,but the results got even worse and the quality of structure decreased.

I am faced with 2 problems 1. how to go about correcting these errors and warning. 2. when can i be sure that my model has been validated by saves.

I'd be really thankful if some could help me in this regard,this is part of my project work and i can't move ahead until im sure that my protein has been modeled correctly. Thanks in advance....

Answer 1

[dimkal]

I would run a short MD simulation (1-3ns) of your protein with restraints on the backbone atoms for the residues in secondary structure elements. This will relax your system and hopefully unstrain the modeled loops. Look into Desmond MD package, it's pretty straight forward building the system and running the simulations.

Answer 2

[João Rodrigues]

First and foremost, ask yourself the question if you really need the model to be "perfect". Loops are loops, they are supposed to be flexible, so whatever fancy algorithm/method you apply it only going to give you a snapshot of what's happening. Depending what you need your model for, you might be looking at something irrelevant.

Nevertheless, I would first look at your alignments and templates. The best way for improving your model lies there. If these are already optimal, you can try to indeed run minimization but since MODELLER already does it you should get almost nothing from it.

Second, MD is likely going to make your structure drift away from what you have, giving you an ensemble of states. Although some might fit better to the stereochemistry criteria, the structure might drift away from the templates, and if you based it on them, it was for a reason.

Finally, don't trust or rely too much on the validation software. Run a crystal structure through it and it will complain loudly..