Tree based models for classification of membrane and secreted proteins in heart

Computational differentiation of membrane and secreted proteins is one of the challenging and interesting topics in bioinformatics. It is a laborious as well as time-consuming task to experimentally differentiate between membrane and secreted proteins. In this study, we used tree-based classifiers such as decision trees, random forest, light gradient boosting machine, gradient boosting decision tree and extreme gradient boosting trees using sequence-based descriptors, viz. amino acid composition, dipeptide composition, conjoint triads, composition/transition/distribution and pseudo amino acid composition in the prediction scheme to enhance the predictive power of algorithms. RF on CTD was able to better discriminate the classes in the classification problems secreted versus non-secreted and secreted versus membrane proteins while in membrane versus non-membrane. Feature selection using mutual information considerably increased the prediction accuracy of the models. Multiclass models to distinguish membrane protein and secreted proteins from other proteins in the heart were enhanced by the addition of protein interaction network-based features, with highest accuracy being displayed by XgBoost. It is expected that the models developed using tree-based algorithms will be useful for classification and annotation proteins with known sequences.