TIMPRED-A WEBSERVER FOR THE PREDICTION OF TIM-BARREL PROTEIN FOLD

Abstract

TIM-barrel is highly conserved and an ubiquitously occurring protein fold in nature and is the most common protein fold in the Protein Data Bank. TIM barrel scaffold is a highly diverse fold with at least known 33 distinct enzyme super families catalyzing activity completely unrelated enzymatic reactions. Moreover, some proteins with TIM barrel architecture, particularly belonging to GH-k (Glycosyl hydrolases) clan have been reported without any enzymatic reactions. The pair wise amino acid sequence similarity of Tim-barrel folds from non-homologous enzyme families is generally below the detectable level. Tim-Barrel prediction server has been developed. A machine learning algorithm using SVM-based method have been used as the prediction method. The prediction model has been developed consisting of training and test set of about 7600 non redundant sequences. Three different features using amino acid count, dipeptide composition and position specific scoring matrix have been used to train the SVM. The overall accuracy and maximum MCC for all the features are 0.75 & 84.93% , 0.75.78 & 75.78%, 0.9434 & 94.34% has been achieved respectively. A five-fold cross validation was used to evaluate the performance of these methods. Currently there is no prediction server that is devoted for TIM-BARREL fold recognition; therefore TIMPRED will prove itself a very important tool in the area of protein fold prediction and identification.