The Wilcoxon Mann Whitney test (two samples), is a non-parametric test used to compare if the distributions of two populations are shifted , i.e. say where k is the shift between the two distributions, thus if k=0 then the two populations are actually the same one. This test is based in the rank of the observations of the two samples, which means that it won’t take into account how big the differences between the values of the two samples are, e.g. if performing a WMW test comparing S1=(1,2) and S2=(100,300) it wouldn’t differ of comparing S1=(1,2) and S2=(4,5). Therefore when having a small sample size this is a great loss of information.

Now, what happens when you perform a WMW test on samples of size 2 and 2 and they are as different as they can be (to what the test concerns), lest say S1=(1,2) and S2=(4,5). Then the p-value of this test would be 0.333, which means that the smallest p-value you can obtain from a WMW test when comparing two samples of size 2 and 2 is 0.3333. Hence you would only be able to detect differences between the two samples when using a level of significance greater than 0.333 .

Finally you must understand that having a sample of two is usually not enough for a statistical test. The following table shows the smallest  p-value for different small sample sizes when the alternative hypothesis is two sided. (Values in the table are rounded).

One of the still open questions in bioinformatics is that of the flexibility of proteins, and it is one in which I am quite interested. Our main source of structural information is X-ray diffraction experiments, in which the crystalline protein is intrinsically rigid. While there is an ever increasing body of NMR data, and Molecular Dynamics simulations are becoming faster and cheaper, complete information about the dynamics of every PDB structure is a long way off. All atom MD simulations for a complete protein still take either days, or a powerful computer. So, naturally, any papers that talk about ways to get flexibility information catch my attention.

The paper that caught my attention was CABS-flex: Server for fast simulation of protein structure fluctuations.. There also is a connected paper – Consistent View of Protein Fluctuations from All-Atom Molecular Dynamics and Coarse-Grained Dynamics with Knowledge-Based Force-Field – which benchmarks their method against MD simulations.

The CABS model pops up in a number of other places – there is a CABS-fold server, and the method was first described in the paper Consistent View of Protein Fluctuations from All-Atom Molecular Dynamics and Coarse-Grained Dynamics with Knowledge-Based Force-Field from 2004.

What does the webserver do?

You give it a coordinate file, wait some hours, and it gives you back a series of (clustered) structures, a C&alpha trajectory, a residue flexibility profile, and a nice little video. Which is nice. It is, however, a little limited; you can only do a single chain (so no modelling of small molecule or peptide interactions), there is no way of fixing part of the model so it does not flex, and it can be picky about your PDB files – no chain gaps, no strange amino acids. You can tell it to analyse a PDB chain by just entering the code, but these are frequently rejected for having unacceptable coordinate files.

How does it work?

The CABS model makes a reduced representation of the protein, and explores its conformational space by making moves – which are accepted or rejected based on a (comparatively complex) force field. Many moves are attempted, and over a large number of iterations, this builds up an ensemble of structures, which show the possible confirmations of the protein. So, its a bit like MD, but rather than moving atoms based on calculated forces, you make random moves, and accept the sensible ones.

The model is shown above. The protein structure is reduced to 4 atoms per residue. There are 5 types of move, and in each step, a random move of each type is tried ~n times (n being the number of amino acids in the protein). These are accepted or rejected based on their force field, and the process is repeated several hundred times. The force field used to accept/reject moves is quite complex – there are short range (sequence dependent and sequence independent), hydrogen bond, and long range (again, sequence dependent and sequence independent) interactions. There are sub-categories within these, and the relative contributions of the interactions can be changed. There is no solvent, so the long range interactions have to provide the forces to keep the protein together, and the hydrogen bond interactions act to maintain the secondary structure elements. More detail can be found in the 2004 paper.

Is it true?

The authors justify the method by comparing it to MD simulations taken from a database, and conclude that CABS-flex gives answers as similar to MD simulations as MD simulations using different force fields are to each other. Simply put, the same residues move in their simulations as move in MD simulations. They move more, which backs up their claim that they demonstrate flexibility over a longer time-frame than MD simulations. They do admit that they get the best agreement when they heavily restrict their secondary structure – raising the question of how much of the agreement between all of the methods is down to all the simulations having the same secondary structure.

To conclude, this could be a useful method, particularly in that it can give long time-frame flexibility – providing you treat it as any other measure of flexibility, with a pinch of salt. It is a bit of a shame that the interface is (perhaps frustratingly) simple, and you need to pre-process your coordinate files, as many PDB coordinate files will not be accepted.

The unidirectional process of translation converts one dimensional mRNA sequences into three dimensional protein structures. The reactant, the mRNA, is built from 64 codon variants, while the product, the protein, is constructed from the 20 biologically relevant amino acids. This reduction of 64 variants to a mere 20 is indicative of the degeneracy contained within the genetic code. Multiple codons encode for the same amino acid, and hence any given protein can be encoded by a multitude of synonymous mRNA sequences.  Current structural biology research often assumes that these synonymous mRNA sequences are equivalent; the ability of certain proteins to unfold and refold without detrimental effects leading to the assumption that the information pertaining to the folded structure is contained exclusively within the protein sequence. In actuality, experimental evidence is mounting that shows synonymous sequences can in fact produce proteins with different properties; synonymous codon switches having been observed to cause a wide range of detrimental effects, such as decreases in enzyme activity, reductions in solubility, and even causing misfolding.

The ribosome (yellow) passes along the mRNA, turning the sequence of codons into a protein. Under our model, the speed of the translation for each codon varies, in turn differing the time available to the nascent peptide chain to explore the fold space. Through this method codon choice becomes an additional source of structural information.

For my 10 week DTC project within the Deane group,  I was tasked with resurrecting the Coding Sequence and Structure database (CSandS; read: Sea Sands) and using it to test for the evolutionary conservation of codon choice over groups of proteins with similar structures. With experimental differences noted in protein product structure between synonymous sequences, we wished to investigate if codon choice is an evolutionary constraint on protein structure, and if so, to what degree, and in what manner. To test for evolutionary conservation we combined the theories of codon optimality and cotranslational folding, our hypothesis being that the choice of codon directly affects the translation efficiency of the ribosome; consequently different codons give the nascent polypeptide emerging from the ribosome varying amounts of time to explore the fold-space.  By assigning optimality scores to each codon, we can search for regions of fast translation and slow translation, then by looking for correlations within aligned sets of structural similar proteins we can identify sites where codon choice is of importance. While the 10 weeks project focussed mainly on methodology and implementation, my preliminary results indicate that a large percentage of proteins are under selective pressures with regards to codon choice. I found that in most sets of proteins, there is an greater amount of correlation than would be expected by chance, this crucially suggests that there is additional structural information contained within the mRNA that is lost once translation occurs.

For additional information on the background, methodology and implementation of this research, please feel free to view the presentation slides at:  http://www.slideshare.net/AlistairMartin/evolut-26981404

As regular readers of this blog will know (I know you’re out there somewhere!), one of the main focusses of OPIG at the moment is antibody structure. For the last ten weeks (as one of my short projects for the Systems Approaches to Biomedical Science program of the DTC) I have been working on predicting the structure of the CDR-H3 loop.

So, a quick reminder on antibody structure: antibodies, which have a characteristic shape reminiscent of the letter `Y’, consist of two identical halves, each containing a heavy and a light chain. Heavy chains are made up of four domains (three constant domains, CH1, CH2 and CH3; and one variable domain, VH), while light chains have two (one constant domain, CL; and one variable domain, VL). The variable domains of both the heavy and light chain together are known as the Fv region; most naturally occurring antibodies have two. At the ends of these Fv regions are six loops, known as the complementarity determining regions, or CDRs. There are three CDRs on each of the VH and VL domains; those located on the VL domain are labelled L1, L2 and L3, while those found on the VH domain are labelled H1, H2 and H3. It is these loops that form the most variable parts of the whole antibody structure, and so it is these CDRs that govern the binding properties of the antibody. Of the six CDRs, by far the most variable is the H3 loop, found in the centre of the antigen binding site. A huge range of H3 lengths have been observed, commonly between 3 and 25 residues but occasionally much longer. This creates a much larger structural diversity when compared to the other CDRs, each of which has at most 8 different lengths. It is the H3 loop that is thought to contribute the most to antigen binding properties. Being able to model this loop is therefore an important part of creating an accurate model, suitable for use in therapeutic antibody design.

Predicting the structure of the loop requires three steps: sampling, filtering and ranking. There are two types of loop modelling method, which differ in the way they perform the sampling step: knowledge-based methods, and ab initio methods. Knowledge-based methods, or database methods, rely upon databases of known loop structures that can be searched in order to find fragments that would form feasible structures when placed in the gap. Whilst predictions are made relatively quickly in this way, one disadvantage is that the database of fragments may not contain anything suitable, and in this situation no prediction would be made. Ab initio (or conformational searching) methods, on the other hand, do not rely upon a set of previously known loop structures – loop conformations are generated computationally, normally by sampling dihedral angles from distributions specific to each amino acid. The loops generated in this way, however, are not ‘closed’, i.e. the loop does not attach to both anchor regions, and therefore some sort of loop closure method must be implemented. The assumption is made that the native loop structure should represent the global minimum of the protein’s free energy. Ab initio methods are generally much slower than knowledge-based ones, and their accuracy is dependent on loop length (long loops are harder to predict using this method), however unlike the database methods, an answer will always be produced.

For my project, I have examined the performance of FREAD (a knowledge-based method) and MECHANO (an ab initio method) when predicting the structure of the H3 loop. At the moment, FREAD produces better results than MECHANO, however we hope to improve the predictions made by both. By optimising the performance of both methods, we hope to create a ‘hybrid’ loop modelling method, thereby exploiting the advantages of both approaches. Since I’ve decided that this is the project I want to continue with, this will be the aim of my DPhil!