Image Title 3

BetaTPred3 Algorithm

In random forest model we use binary profile, evolutionary information in form of PSSM profile and PSIPRED predicted secondary structure of tetra peptide.

More »
Image Title 1

Beta Turn in proteins

Beta-turns are the most common type of non-repetitive structures, and constitute on average 25% of the residues in all protein chains. In a beta turn, a tight loop is formed when the carbonyl oxygen of one residue forms a hydrogen bond with the amide proton of an amino acid three residues down the chain. This hydrogen bond stabilizes the beta bend structure. A beta turn can reverse the direction of its peptide chain.

Image Title 2

Prediction of Beta Turn

In the past numerous methods were developed to predict the beta turns. But all of this method were trained to predict residue level prediction instead of four residue level. Since, a beta turn is composed of four consecutive amno acids. Using this simple approach we have achieved good prediction accuracy and realistic prediction of beta turns. To predict beta turns in your protein click »Submit »

Image Title 2

Propensity based Beta Turn prediction

In the past statistical methods were developed to predict the beta turns based upon propensity score of beta turn. The propensity score was calculated using few hundered PDBs. We have calculate new propensity score using ~18000 PDBs. Users can predict beta turns based upon various position based propensity score. Click here to »Submit »

Image Title 2

Designing of Beta Turn

For the first time, we have developed a module thats helps user in understanding the positional preference of pairs of amino acids. First, user sequence is mapped and various propensity score are shown for all possible tetrapeptide. Second, the module performs all possible mutation in a tetrapeptide, either to increase or decrease its beta turn formation probability. Click here to »Submit »

Image Title 2

Prediction of Beta Turn Type

In the past numerous methods were developed to predict the beta turn types. Using the turn level approach we have significantly improved the prediction accuracy of beta turn types. To predict beta turns types in your protein click »Submit »

Image Title 4

Download dataset

To Download previously created dataset and the lastest dataset. We have created two dataset, one is 30% non-redunant and other is 90% non-redunant dataset.

More »

Algorithmic Details
Here we describe the steps involved in the backend of BetaTPred3 prediction method. The main concept in this study is the use of turn level prediction rather than residue level prediction which is employed in previous methods. The turn level prediction approach has advantage to the residue based approach in the sense that the prediction is more realistic and the models developed are simple and not complex.
    Turn level Beta-turn prediction
    Turn level Beta-turn Types prediction
    Propensity based Beta-turn prediction
    Designing (inducing/breaking) Betaturns in a protein



Turn level Beta-turn prediction

First, using input sequence we generate PSSM profile and predicted secondary structure using psipred inherent in HHSuite. Next, we generate pattern of four consecutive residues. If the pattern is making beta turn, we call it postive dataset otherwise we call it negative dataset. Next we combined the PSSM of four residues and predicted secondary strucutre to generate a window length of 92. With the inclusion of tetra-peptide propensity score, the final input vector becomes 93. In the end, we developed RandomForest model to predict beta turn in proteins. Figure 1 (given below) gives a graphical representation of the Turn level prediction approach of BetaTpred 3.0

img
Figure 1. Graphical representation of Turn level prediction approach of BetaTPred 3.0

Turn level Beta-turn Types prediction

There are 9 subtypes of Beta-turns which are Type I, Type I', Type II, Type II', Type IV, Type VIa1, Type VIa2, Type VIb and Type VIII. The turn level approach is also applied in the prediction of Beta-turn Types. For each turn type, random forest based model is developed resulting in total of 9 models. Finally the turn type with probability greater than the respective thresholds is given as output. (Figure 2)

img
Figure 2. Graphical representation of prediction of Betaturn Types using BetaTPred 3.0

Propensity based Beta-turn prediction

In this approach, input sequence is splitted in overlapping windows of length 4. Different propensity scores of each window is calculated like Residue Position wise propensity (P1, P2, P3 and P4); Residue Pair wise propensity (P1-2, P1-3, P1-4, P2-3, P2-4, P3-4); Tri-peptide propensity (P1-2-3, P2-3-4); Tetra-peptide propensity (P1-2-3-4). The respective scores for each propensity based method is averaged. A threshold (T) is applied on the average score. If average score is above the threshold, the pattern is classified as Beta-turn, otherwise the pattern is classified as non-turn. Figure 3 (given below) displays the propensity based Betaturn prediction.

img
Figure 3. Graphical representation of prediction of Betaturns based on different propensity based methods.

Designing (inducing/breaking) Betaturns in a protein

Protein Engineers are more interested in mutating some residue(s) in their protein of interest by which they can induce or break betaturn in the protein. In this module, the Propensity based approach is used to classify each pattern of the protein as Beta-turn or non-turn. In the next step, each pattern can be further mutated which will change its overall propensity towards formation of Betaturn. In this way, a user can identify specific pattern and mutation which can induce or break Beta-turns.

img
Figure 4. Graphical representation of the working of Design module of BetaTPred 3.0