Dear Modeller users,
I beg your kind help or explanations for I feel quite at loss with calculation of structure-structure alignments and optimal RMSD superpositions with modeller.
At first, I found it quite problematic to produce homology models properly superimposed on templates. I started with seemingly simple idea using automodel class, namely adding:
a.final_malign3d = True
to the model-default.py script from modeller examples. The result was that the final superposition identified just 8 equivalent CA-CA pairs, for which the RMSD CA was evaluated as 2.79A. However, when I superimposed the very same pair of structures (model 1fdx on 5fd1 template) and calculate RMSD CA in another program using all 54 CA-CA pairs according to alignment.ali used by model-defult.py, I got 0.5A. As for the automodel class, intended for simple use, it surprised me that quite basic task as RMSD CA superposition is not done correctly. By "correct" or at least the most basic way I understand the superposition giving the minimum of RMSD using all CA atoms equivalent according to the alignment. Maybe I used wrong functionality but I have not found an alternative in the documentation for the automodel class.
Finding the automated final multiple structure alignment unreliable, I started looking into proper ways of superimposing structures. I have read the documentation on alignment.malign3d and finding it obsolete, on alignment.salign. Now I am at least able to calculate alignments based on sequence similarity only and use them to superimpose structures. However, I am still not able to calculate/improve the alignments based on structure-structure similarity, and again this seems due to insufficient number of equivalent CA-CA pairs identified by modeller.
This is the code I tested, using the structures from modeller examples. First, sequence-sequence alignment (which works):
from modeller import * env = environ() env.io.atom_files_directory = ['.', 'atom_files'] mdl = model(env) aln = alignment(env) for code in '1fas', '2ctx': mdl.read(file=code) aln.append_model(mdl, align_codes=code, atom_files=code) aln.salign(rr_file='${LIB}/blosum62.sim.mat', feature_weights=(1,0,0,0,0,0), improve_alignment=True, similarity_flag=True, # The score matrix is not rescaled rms_cutoff=300, current_directory=True, write_fit=True, fit=True, fit_atoms="CA", output='ALIGNMENT QUALITY') aln.write(file='test1.ali')
The above part produces sensible sequence alignment in test1.ali and it reports 61 equivalent CA-CA pairs, but structures saved in the files 1fas_fit.pdb and 2ctx_fit.pdb are NOT superimposed. I can, however, use the aln object from above to correctly superimpose structures with model and selection classes:
mdl.read(file='1fas') sel = selection(mdl).only_atom_types('CA') mdl2 = model(env, file='2ctx') sel.superpose(mdl2, aln) mdl.write(file='1fas_fit2.pdb') mdl2.write(file='2ctx_fit2.pdb')
Now the structures written out to 1fas_fit2.pdb and 2ctx_fit2.pdb ARE superimposed. They are however different enough that only 5 equivalent CA-CA pairs within 3.5A cutoff are reported. I am able to confirm that with sufficiently large cutoff, there are 61 CA-CA pairs, consistently with the alignment aln (checked by examination of the test1.ali file). Such confirmation is produced with:
aln.compare_structures(rms_cutoffs=[999]*11)
Finally, what does not work for me at all is when I try to improve the alignment based on structure-structure similarity. The command below - as a continuation of the previous code snippets - fails with the exception that the number of equivalent positions is 0:
aln.salign(fit=True, fit_atoms="CA", feature_weights=(0,1,0,0,0,0), rms_cutoff=1000.0, improve_alignment=True, current_directory=True, write_fit=True, output='ALIGNMENT QUALITY')
Questions: 1. Why it fails, if the previous checks find at least 5 equivalent positions even with the default cutoff (3.5), and with the rms_cutoff=1000.0 and fit_atoms="CA" there should be 61 equivalent positions according to the alignment?
2. Why the first salign invocation has saved structures which were not superimposed, incoherently with the options fit=True, write_fit=True?
3. How one shoud use the salign function to reliably calculate optimal structure-structure alignment, or optimize some initial one as I tried to do?
Thanks in advance, Paweł Kędzierski