Classification of peptides using ensembles by applying different strategies that deal with imbalanced data and combination rules

The discovery and synthesis of peptides with antimicrobial properties is a promising alternative to fight against multi-resistant bacteria. There are multiple studies that deal with the classification of peptides according with their probability to possess antimicrobial activity. One of the challenges in these classification processes is related with the amount of available data. For the case of antibacterial peptides classifiers, the size of the positive class is much bigger than the negative class. In this work, we propose two strategies to deal with the imbalance situation of the data by using ensembles. The first one is based on algorithm modifications and the second one with data management. For each strategy we used five combination rules. The performance of the ensembles was calculated using the area under the ROC curve (AUC). Our results suggest that care must be taken with ensembles and that individual classifiers must be studied in-depth.