Enhanced Prediction of mRNA Subcellular Localization Using a Novel Ensemble Learning and Hybrid Approach

Unraveling the subcellular localization of mRNA is an imperative aspect in the realm of biotechnology. This resolution can illuminate the inner workings of genetic regulatory mechanisms, gene expression modalities, and the evolution of cellular physiological and developmental processes. However, the experimental delineation of mRNA subcellular localization imposes significant temporal and resource commitments. Despite the development of multiple algorithms and predictive models for mRNA subcellular localization, their performance indexes have not been markedly high. In this paper, we introduce a novel hybrid approach to categorize mRNA into five distinct subcellular locales, including the cytoplasm, endoplasmic reticulum, extracellular region, mitochondria, and nucleus. Our model exploits the strengths of ensemble learning with a hybrid methodology, incorporating multiple biologically pertinent features extracted from the input sequencing data. Additionally, the model dynamically adjusts the weightages of functions and the minority class, through the modulation of the weight ratio of disparate models during their contribution to the principal model. Overall, our model delivers promising results, with an average accuracy of 0.89 in an independent dataset for the classification of mRNA subcellular localizations into five subclasses. This displays a significant performance elevation in contrast to preceding algorithms, particularly in instances where the classes are adequately sampled.