VacPred: Sequence-based prediction of plant vacuole proteins using machine-learning techniques

ABSTRACT

Subcellular localization prediction of the proteome is one of major goals of large-scale genome or proteomesequencing projects to define the gene functions that could be possible with the help of computationalmodeling techniques. Previously, different methods have been developed for this purpose using multi-labelclassification system and achieved a high level of accuracy. However, during the validation of our blind datasetof plant vacuole proteins, we observed that they have poor performance with accuracy value range from*1.3% to 48.5%. The results showed that the previously developed methods are not very accurate for the plantvacuole protein prediction and thus emphasize the need to develop a more accurate and reliable algorithm. Inthis study, we have developed various compositions as well as PSSM-based models and achieved a highaccuracy than previously developed methods. We have shown that our best model achieved *63% accuracyon blind dataset, which is far better than currently available tools. Furthermore, we have implemented our bestmodels in the form of GUI-based free software called ‘VacPred’ which is compatible with both Linux andWindow platform. This software is freely available for download at www.deepaklab.com/vacpred.