A neural network method for prediction of β-turn types in proteins using evolutionary information

Abstract

A web server for prediction of beta-turn types I, II, IV and VIII based on above approach is available at http://www.imtech.res.in/raghava/betaturns/ and http://bioinformatics.uams.edu/mirror/betaturns/ (mirror site).

Sequence-Only Based Prediction of β -Turn Location and Type Using Collocation of Amino Acid Pairs

“…As a result, prediction of the location of beta-turn types, in contrast to a binary prediction that would identify location of beta-turns, would provide additional, structural, information *Address correspondence to this author at the Department of Electrical and Computed Engineering, University of Alberta, Edmonton, Alberta, Canada T6G 2V4; E-mail: lkurgan@ece.ualberta.ca that concerns the and angles. A commonly used benchmark dataset of 426 non-homologous protein chains [10], which have been used to rank and test several methods for prediction of beta-turn types [11,12], reveals that some of these types are infrequent and thus they are commonly combined together [11]. To this end, we focus on prediction of beta-turn types I, II, IV and VIII, while the remaining types I', II', VIa1, VIa2 and VIb, which only make up 304, 165, 44, 17 and 70 respectively out of the total 7153 betaturns in the aforementioned dataset have been combined into one set referred to as non-specific (NS), which is consistent with [11].…”

“…A commonly used benchmark dataset of 426 non-homologous protein chains [10], which have been used to rank and test several methods for prediction of beta-turn types [11,12], reveals that some of these types are infrequent and thus they are commonly combined together [11]. To this end, we focus on prediction of beta-turn types I, II, IV and VIII, while the remaining types I', II', VIa1, VIa2 and VIb, which only make up 304, 165, 44, 17 and 70 respectively out of the total 7153 betaturns in the aforementioned dataset have been combined into one set referred to as non-specific (NS), which is consistent with [11]. The challenging aspect of the beta-turn type prediction is that these turns are not isolated in a chain.…”

“…The machine learning methods include BETATURNS [11] and BTPRED [12], which are based on artificial neural networks (ANN), and a hybrid multinomial logistic regression and ANN model [18]. BTPRED encodes the sequence using a large window of 11 residues centered over the predicted residue together with secondary structure predicted with PHDsec [19] to perform predictions.…”

“…In comparison with the methods that predict beta-turn types for entire protein sequences outlined above, which include [11][12][13] and which use significant amount of auxiliary information such as PSSM and predicted secondary structure, it is clear that a method based solely on the protein sequence would be simpler to design and execute. However, this may lead to reduced quality as only limited information (sequence) relative to the competition would be used.…”

“…We also note that only two sequence-only based method (method that uses as the input only the protein sequence and no sequence derived information such as PSSM or predicted secondary structure) are available for prediction of beta-turn types [1,23]. The method in [23] addresses prediction of only type I and type II beta-turns, while the method in [1] predicts beta-turn types I, I', II, II', VI, and VIII, which are different than targets addressed by newer prediction methods [11]. This provides additional motivation for the development of the proposed method.…”