Section 8-6: Protein Tools

[ Previous chapter ][ This chapter ][ Next chapter ] There are various tools available which allow you to analyse single protein sequences.

• Prediction from Scratch

  In order to predict the structures of peptides or even proteins with yet unknown homology to known proteins, it is required to use methods which assign parameters to amino acids which will allow to get an estimation for a possible secondary structure. The programs in use frequently use three-state (or four-state) predictions:

• ALPHA HELIX

• BETA SHEET

• COIL (no structure at all)

• TURN (only used by a few methods)

  All methods work similar to the following assumption:

  Based on an analysis of known protein structures, amino acids are classified into "classes" which will most frequently occur in one of the three (or four) states as described above. Statistical methods are applied to evaluate the variance of these occurrences in all positions of all proteins, and numerical values are assigned to each amino acid and its probability to be found in one of the three (or four) states. A window is applied (see the section on window analysis earlier in this chapter) in order to calculate a value which is significant for the given position and amino acid in the peptide which shall be predicted. Plotting the curves of the three (or four) states it is possible to derive a prediction for the whole protein. Note that, due to this way of calculation, the states are not mutually exclusive and therefore a considerable uncertainty is implied.

  Chou and Fassmann pioneered this approach already many years ago and used the tertiary and secondary structures of proteins available at that time. The applicability of their method, therefore, is constrained to the set of proteins available at that time ( globular, soluble proteins). The precision is, on average, estimated to be 60-65% if the Chou and Fassman method is used. Robson and Garnier have improved precision to close to 65% on average by using a matrix of neighbour values rather than a single window approach. Eisenberg has used additional parameters such as "hydrophaty".

  Summarising, the use of secondary structure prediction from scratch is highly speculative and should not be the only method to reach conclusions. Averages on aligned sequences as described in later chapters of the BioCompanion will give the best results with an expectation to be close to 70% accurate.

• Molecular Modelling

  Based on a sequence homology with an existing, structurally known protein fragment, it is possible to use advanced computer graphics displays to build the peptide fragment covering the homology according to the known structure of the protein found in the database. The minimum sequence homology required to do this kind of model building severely depends on the individual peptide but should definitively be better than 30% - which means that more than 10 amino acids must be identical in a sequence of 30 amino acids. Building the model will be achieved in three steps:

  Initially, the sequence to be modelled is aligned to the sequence with the known structure coordinates. This procedure is described later and shall make you familiar with the kind of replacements which will need to be done in the model. Be careful to avoid the "impossible" - glycine residues frequently are required to adopt an unusual configuration, proline residues are either required in turn structures, or their introduction will possibly break secondary structure elements, and disulphide bridges are extremely important in protein structure.

  Secondly, an advanced computer graphics program is used. Famous representatives include, but are not limited to, the Insight II program package from Biosym Technologies, the Cerius program suite from Molecular Simulations, and the 'Whatif' program as created and used at EMBL, Heidelberg. The first two examples are high-end commercial software packages (since fall 1995, available from a single vendor) and require significant investment, whereas the 'Whatif' package is available for negotiable price to academic researches. All of these packages usually require a Silicon Graphics computer system with high-end graphics and a large monitor. If possible, a stereo viewing capability should be available. Be aware of the complexity of these programs - this is not rocket science any longer. Replacing amino acid side chains might be an easy procedure on the display but requires considerable thinking to get a reasonable biologically relevant model.

  Last, once the initial model is completed, mathematical procedures will need to be applied to refine this model. This will require that so-called potentials are assigned to the atoms of your peptide - which means that the groups and residues are classified by chemical topology to be of a certain class of arrangement. E.g., a peptide group will be semi-planar due to the hybridisation of the participating atoms. Subsequently, a semi-empirical force field is applied, and the structure is fit to "ideal" angles, atom distances and bond geometry with techniques known as molecular dynamics and structure refinement.

  Molecular dynamics will initiate movements of atoms by introducing a "temperature" which allows to "shake" the molecule into a more ideal configuration. To do this step reliably, it is essential to know of constraints which limit some of the three-dimensional properties of the protein or peptide. These constraints will be data from X-ray crystallography or NMR structure analysis. Without these constraints, model building is highly speculative. You should not desire to start a study involving molecular dynamics refinement unless you have at least some constraints such as disulphides, maximal cross-section or other data. A very serious and hence unsolved problem is to compensate for interactions of your model peptide with the environment. This environment is frequently "vacuum" - at least this is computationally the easiest approach. Be aware that this is not a very satisfying approximation. Water or membrane molecules are much better suited as interaction partners. However, supercomputing performance will be needed to run this analysis, and the results will be only of limited relevance if you did not use any constraints.

  Summarising, it must be stated that the molecular modelling approach for secondary structure prediction may be very useful and suggestive if small changes of a peptide sequence to an already known peptide structure are to be applied. The larger the deviation, i.e., the lower the similarity to a known structure, the less relevant will be the results. Experimental structure evaluation with X-ray and/or NMR techniques might be required for satisfactory models.

Remember that the prediction of secondary structure without a reasonable homology to three-dimensional data is rather unsafe. Programs which employ three-dimensional modelling techniques require special hardware (powerful computers) and dedicated software, hence, are beyond the scope of the BioCompanion .

The programs available to you in the desktop environment wil typically be restricted to secondary structure prediction from scratch. In order to display the secondary structure plots, you need to have a computer screen which is capable of displaying graphics. It is recommended that you have access to a colour graphics device if you want to run these programs. Remember to have set the graphics environment correctly with setplot if you work with GCG locally. X-Windows setups must have set the DISPLAY environment correctly.

To display several measures of secondary structure, use

% pepplot

To generate a table of several measures (with a comparison of Garnier-Robson and Chou-Fassman predictions), use

% peptidestructure

The generated output file can be plotted "two-dimensionally", but for serious inspection the one-dimensional plotting is recommended (use the corresponding menu option):

% plotstructure

Given the assumption that the protein fragment adopts a helical structure, the program helicalwheel can be used.

The program moments plots a three-dimensional map which displays moments of hydrophathy in dependence of the sequence and the rotational angle of the peptide bond (90 - 110 degrees is OK for helices, 0 or 180 degrees is indicating chances for beta sheet).

The programs peptidemap and peptidesort work like the DNA counterparts .

The program peptidesort can also be used advantageously to determine the composition of a peptide. If the <SPACE BAR> is hit on the question "Which Enzymes?", no fragmentation is calculated. Rather, the composition is detailed to a much larger extent than the composition program will provide.

The isoelectric point of the denatured protein can be determined from the titration curve plotted by the program isoelectric .

Frequently, you might want to know where "acidic" or other regions of your protein sequence are located. As ambiguity symbols in the single-letter peptide alphabet are not defined, you might rewrite your sequence and use the window program in order to plot the result with statplot . The data for the simplify program are located in a file which you can get from the GCG program database with the command

% fetch simplify.txt

This file has a self-describing format, and basically will replace each amino acid listed in the second column with an amino acid listed in the first column:

 

  
 D   DEQN
  

  

================================= Begin Exercise 7

Summary of single-sequence tools: Translate the sequence GENEMBL:M19311 in the determined reading frame, perform a secondary structure prediction from scratch, and plot the acidic amino acids as function of the sequence.

To use amino acid sequences, the computer needs a defined reading frame in the DNA sequence which allows the translation into a peptide sequence. The translated amino acids are written into a peptide sequence. The purpose of this exercise is to create the sequence M19311.pep and predict its secondary structure. Proceed as follows:

• Refer to previous exercises to get the required reading frame. Use either map and extractpeptide or the translate program to generate the peptide sequence.

• Use the pepplot , moments and peptidestructure / plotstructure programs to predict the secondary structure of your peptide. ( CAUTION: many * symbols will indicate an erroneous sequence!)

• Use fetch to get a simplification file, review it and plot the occurrence of acidic amino acids in your peptide after simplify with window and statplot .

================================= End Exercise 7