Implementation of Freeman-Wimley prediction algorithm in a web-based application for in silico identification of beta-barrel membrane proteins

Abstract

Beta-barrel type proteins play an important role in both, human and veterinary medicine. In particular, their localization on the bacterial surface, and their involvement in virulence mechanisms of pathogens, have turned them into an interesting target in studies to search for vaccine candidates. Recently, Freeman and Wimley developed a prediction algorithm based on the physicochemical properties of transmembrane beta-barrels proteins (TMBBs). Based on that algorithm, and using Grails, a web-based application was implemented. This system, named Beta Predictor, is capable of processing from one protein sequence to complete predicted proteomes up to 10000 proteins with a runtime of about 0.019 seconds per 500-residue protein, and it allows graphical analyses for each protein. The application was evaluated with a validation set of 535 non-redundant proteins, 102 TMBBs and 433 non-TMBBs. The sensitivity, specificity, Matthews correlation coefficient, positive predictive value and accuracy were calculated, being 85.29%, 95.15%, 78.72%, 80.56% and 93.27%, respectively. The performance of this system was compared with TMBBs predictors, BOMP and TMBHunt, using the same validation set. Taking into account the order mentioned above, the following results were obtained: 76.47%, 99.31%, 83.05%, 96.30% and 94.95% for BOMP, and 78.43%, 92.38%, 67.90%, 70.17% and 89.78% for TMBHunt. Beta Predictor was outperformed by BOMP but the latter showed better behavior than TMBHunt

Keywords

• membrane proteins
• beta-barrels
• in silico
• vaccine