Comparison of the ability to predict true linear B-cell epitopes by on-line available prediction programs

COSTA, JUAN GABRIEL; FACCENDINI, PABLO LUIS; SFERCO, SILVANO; LAGIER, CLAUDIA MARINA; MARCIPAR, IVÁN SERGIO

Paraná

Congreso; 3er congreso de la sociedad argentina de bioinformática y biología computacional; 2012

Sociedad Argentina de Bioinformática y Biología Computacional

Comparison of the ability to predict true linear B-cell epitopes by on-line available prediction programs J. Gabriel Costa 1, Pablo L. Faccendini 2, Silvano J. Sferco 3, Claudia M. Lagier 2, Iván S. Marcipar 1 1 Laboratorio de Tecnología Inmunológica, Facultad de Bioquímica y Ciencias Biológicas, Universidad Nacional del Litoral. Paraje El Pozo. Santa Fe, Argentina. 2 IQUIR, Depto. de Química Analítica, Facultad de Ciencias Bioquímicas y Farmacéuticas, Universidad Nacional de Rosario. Suipacha 531. Rosario, Argentina. 3 Departamento de Física, Facultad de Bioquímica y Ciencias Biológicas, Universidad Nacional del Litoral, Paraje El Pozo. Santa Fe, Argentina; and INTEC (CONICET-UNL), Güemes 3450, Santa Fe, Argentina. Background: Several experimental methods have been developed to identify B epitopes from infectious microorganism proteins. However, these methodologies are long term demanding and quite expensive. Our work deals with the use of prediction programs to identify useful B cell linear epitopes to develop immunoassays. Therefore, we have tested 5 free, on-line prediction methods (AAPPred, ABCpred, Bcepred, BepiPred and Antigenic), widely used for predicting linear epitopes, using the primary structure of protein as the only input. Each program uses a very different algorithm. Methods and Results: To compare the quality of the predictor methods we have used their positive predictive value (PPV), i.e. the proportion of the predicted epitopes which are true, experimentally confirmed epitopes, in relation to all of the epitopes predicted. Eleven proteins which had been whole mapped experimentally by highly reliable techniques to detect epitopes, were studied. Each program was run and predicted epitopes were compared with the 65 true epitopes dispayed in the proteins. In order to identify useful predicted linear epitopes, none supposed true negative set was used. The confidence intervals of PPV were calculated with at 90% level of significance for each different prediction procedures. The best PPV were obtained with AAPpred and ABCpred, 69.1% and 62.8% respectively. We also statistically evaluate the differences between theses PPV values when counting with paired data. This allowed us studying which program produced a PPV value different from that calculated for another program, stated with 90% certainty. Then, to monitor the programs prediction efficiency, we compared the epitope identifying positive prediction value with that obtained when randomly selecting regions of the molecule under study. Our results indicate that only 2 of the programs studied predicted epitopes with a statistically significant higher positive prediction value than a random procedure, these being AAPPred and ABCpred. Although, we analyzed if the epitopes predicted by the consensus of several programs were more efficient than those which had been predicted with each program alone or with partial consensus. But we observed that considering as true epitopes only the consensus regions to several programs, does not improve PPV value with respect to the results produced by each program individually. Conclusion: We conclude that AAPPred and ABCpred yield the best results, as compared with the other programs and with a random prediction procedure. We also ascertained that considering the consensual epitopes predicted by several programs does not improve the prediction positive predictive value.