previous section previous page next page next page

Online Lectures on Bioinformatics


Protein Structure

Comparative modelling

If the target sequence of amino acids one wishes to obtain a structural model of is very similar to a template protein of known structure, it's possible to obtain the structural model by copying the backbone elements of the template and adding loops and side chains. This is known as comparative protein modelling, knowledge based modelling or homology modelling.

The modelling process coarsely divides into the following steps:

  • choose suitable template sequence(s)
  • align template and target
  • build backbone
  • construct loop and generate side chains

Fully automated programs performing comparative modelling exist which obtain reasonable results only if template and target share at least 50 % identical residues in an alignment.

exercise 3
exercise 3

Detection of suitable templates ususally is done by performing a BLAST-search against the sequences in the PDB (Brookhaven Protein Data Bank), the repository for protein-structure obtained by X-ray crystallography and NMR. All sequences in the PDB with an E-value of Blast below a certain treshold are considered as candidates for the template.

The next step consists of aligning target and template sequence. At this point the modeller has to decide, if the detected template is suited for the modelling process. Of course, template and target should share regions of high similarity. Taking 100 positions of the alignment into account, a rule of thumb is, that the alignment at least should contain 30 % identities. Most important is of course the existence of conserved regions. A multiple sequence alignment with family members of the template or several template sequences may be constructed.
Note that the alignments produced by a dynamic programming algorithm reflect evolutionary relatedness of sequences. But the goal is to obtain a structural alignment, such that the backbone may be overtaken to the model to be generated. Therefore a manual correction of the alignment may be very important in order to obtain a reliable model. A structural correction of the alignment after the modelling-process implies a repetition of previous modelling-steps.

Several methods are used for loop building and side chain reconstruction. For example Ramachandran plots compute suited torsion angles in polypeptides, whereat atoms which are supposed to be hard spheres with van der Waals radii are not allowed to collide [9]. Futhermore force field scores are used [10][11].

Comments are very welcome.