High accuracy prediction of β‐turns and their types using propensities and multiple alignments

Abstract

We have developed a method that predicts both the presence and the type of beta-turns, using a straightforward approach based on propensities and multiple alignments. The propensities were calculated classically, but the way to use them for prediction was completely new: starting from a tetrapeptide sequence on which one wants to evaluate the presence of a beta-turn, the propensity for a given residue is modified by taking into account all the residues present in the multiple alignment at this position. The evaluation of a score is then done by weighting these propensities by the use of Position-specific score matrices generated by PSI-BLAST. The introduction of secondary structure information predicted by PSIPRED or SSPRO2 as well as taking into account the flanking residues around the tetrapeptide improved the accuracy greatly. This latter evaluated on a database of 426 reference proteins (previously used on other studies) by a sevenfold crossvalidation gave very good results with a Matthews Correlation Coefficient (MCC) of 0.42 and an overall prediction accuracy of 74.8%; this places our method among the best ones. A jackknife test was also done, which gave results within the same range. This shows that it is possible to reach neural networks accuracy with considerably less computional cost and complexity. Furthermore, propensities remain excellent descriptors of amino acid tendencies to belong to beta-turns, which can be useful for peptide or protein engineering and design. For beta-turn type prediction, we reached the best accuracy ever published in terms of MCC (except for the irregular type IV) in the range of 0.25-0.30 for types I, II, and I' and 0.13-0.15 for types VIII, II', and IV. To our knowledge, our method is the only one available on the Web that predicts types I' and II'. The accuracy evaluated on two larger databases of 547 and 823 proteins was not improved significantly. All of this was implemented into a Web server called COUDES (French acronym for: Chercher Ou Une Deviation Existe Surement), which is available at the following URL: http://bioserv.rpbs.jussieu.fr/Coudes/index.html within the new bioinformatics platform RPBS.

Role of the Extensin Superfamily in Primary Cell Wall Architecture

“…More recent work using mutants atprp2 and atprp4 show that AtPRP2 and AtPRP4 are required for stomatal guard cell function (Carter and Tierney, 2010). Significantly, predictions of secondary structure using COUDES software (Fuchs and Alix, 2005) indicate that the repetitive motifs of AtPRP2 and AtPRP4 form a coil of b-turns. This suggests that an elastic protein contributes to the elasticity of guard cell walls.…”

Predicting Βeta-Turns and Βeta-Turn Types Using a Novel Over-Sampling Approach

“…It has been used in many researches [3] [6]- [10] [12] [13] [15] as the standard dataset for the comparison. In addition, two more other datasets, BT547and BT823, that were constructed for training and testing COUDES [17], were also used in our study. These datasets contain 426,547 and 823 protein sequences, respectively.…”

“…As MCC, Q total , Q observed , Q predicted are often used to measure the quality of β-turn prediction methods [17], they are used to evaluate the performance of our method and are defined as below:…”

“…In the realm of β-turn types prediction, most methods are based on ANN [6] [16], probabilities with multiple sequence alignments as COUDES [17], or SVMs [9] [18] [19]. However, the quality of β-turns and β-turn types prediction is still inadequate.…”

Application of HMM to the Study of Three-Dimensional Protein Structure

“…The prediction of turns is also of special interest in protein study (Fuchs & Alix, 2005). The four-SL word HBDS can be linked to β-turns: the corresponding fragment conformations are shown in Figure 10.…”