ML-RBF: Predict protein subcellular locations in a multi-label system using evolutionary features

Abstract Machine learning assisted sub-cellular protein localization has been an emerging research area from the last two decades since it provides fast, low cost and more precise tagging of different sub-cellular proteins. However, most of the research is focused on single-plex proteins which always resides on a particular location in all the cells at all times. Later studies reveal that many proteins can reside at more than one location and are therefore called multi-plex. These proteins were previously noted mistakenly as moved away from their location due to some cellular malfunction caused by some disease. The dynamic behavior of these proteins depicts their importance in cellular functionality and thus they are more worthy of being studied by the researchers. The current study proposes a novel model for the classification of multi-label proteins using evolutionary feature extraction via Position Specific Scoring Matrix. Two benchmark multi-label datasets (of bacteria and viruses) are employed to draw realistic comparisons. The study utilizes three state-of-the-art classifiers to draw a working comparison and the results are discussed rigorously utilizing various statistical performance metrics specifically proposed for multi-label classification. The proposed model yielded 93% and 94% average precisions for the two datasets respectively. It demonstrates its reliability and capability for utilization in further similar studies.