Skip to content

How useful are the modeled protein structures?

by Kundan Ingale on January 17th, 2012

Wish you all “A Very Happy and Prosperous New Year!!!”

Since my last post, discussion on individual threads was required, in this post I will discuss on modeled protein structures.

Why model protein structures?

To understanding the biological function of proteins including their interactions with other proteins, ligands, substrates and inhibitors, it is necessary to determine their 3D structure. Classical methods of structure determination take massive time and money.

Thus, computational methods for predicting the protein structure form the sequence, like Ab-initio, threading, comparative modeling, become increasingly important.

How accurate are modeled protein structures?

For reasons like huge number of conformations of protein structures and incomplete understanding about physical stability of protein structures; the chances of errors in modeled structures are high. The extent of errors inherited in the model may be different in different regions, like the non-conserved surface loops are expected to be the least reliable parts of a protein structure may deviate markedly from experimentally determined control structure.

Modeled helices and strands are ideally straight, while bends, due to many factors steric interaction between side chains or interaction with solvent molecules, are often found in the secondary structures of proteins. On the contrary, such a tricky assignment of secondary structure could produce large bends in the protein structure, which can be analyzed by visual inspection.

Optimizing local atomic geometry of the structure, deviation from global topology is expected and when a small number of discrete states are used to represent a protein structure, the structure is necessarily distorted. Although small in nature the distortion effects can be cumulative, small distortions in a larger target can lead to substantial structural changes.

Use of modeled protein structures?

Inclusion of predicted protein structures with local structural distortions in the screening process may yield much lower enrichment of known actives, especially for structural distortions present near the binding site, resulting in significant drop-off in the ability to recognize ligands. Interpretation of the in-vivo expression profiles of proteins on the basis of such a model may be misleading.

Identifying the sources and magnitude of errors/variations in predicting biological profiles for small molecules could prove critical in cases where modeled proteins are the only solution.

Kundan Ingale

I am Medicinal Chemist exploring applications of CADD. Working as Application Scientist, at VLife Sciences Technologies Pvt. Ltd. I am closely associated with technology development for computer aided drug discovery and its application.

More Posts - Website

  1. Chandrasekharan Ramakrishnan permalink

    I agree with the opinion of the author that modelling of protein structure from sequence can sometimes contain errors. But my doubt is how to find that the model is incorrect, to say the least. While it is true that modelling is a good starting point for determining the structure, more deterministic criteria are needed for effective checking. The final model definitely depends on the type of input, both qualitative and quantitative. I have two basic questions. 1) Can the structure determined by X-ray crystallography be taken as “THE STRUCTURE” or it is only one of the possible structure. 2) Even assuming that a structure determined by X-ray method is the right one, is there any confirmed medthod which can be used to obtain the structure from sequence, not by just using the available data on already determined structures judiciously or in otherwords modelling?

  2. Paul Hawkins permalink

    This article may be of use

    The authors show that, in some cases, homology models give “better” virtual screening by docking results than experimental structures. Now this may be due entirely to the well known deficiencies in the scoring functions used in the docking engines tested, but it is a provocative observation nonetheless.


    • Kamalakar Jadhav permalink

      Yes, I agree with Paul on this. However I would like to add few more points to bring in Prof. Ramakrishnan’s attention.
      I believe, the success of computational model protein structure primarily rests on
      (i) An existence of a homologue with known structure,
      (ii) Modeler’s ability to detect this homologue and
      (iii) The quality of the model building process (method/algorithm)once the homologue is detected.

      The quality of an in silico or computational model is more subjective and ultimately defined through the usefulness of the model

Leave a Reply

Note: XHTML is allowed. Your email address will never be published.

Subscribe to this comment feed via RSS

Follow Me