<![CDATA[This study evaluates the efficacy of AdaBoost and XGBoost in classifying Cellular Essential Genes (CEG) and Organismal Essential Genes (OEG) of Drosophila melanogaster using a hybrid feature set of DNA sequences, protein sequences, and network topology 185 features comprising Tri-Nucleotide Composition (TNC, from DNA) and Fourier Transform (FT, from DNA only), Amino Acid Composition (AAC, from protein sequences), and Protein-Protein Interaction (PPI) degree (from network topology) retrieved from the CLEARER database, with Random Forest Gini feature selection and SMOTETomek balancing nested within a leakage-free stratified 5×10-fold cross-validation pipeline, demonstrating that XGBoost consistently outperforms AdaBoost by achieving 96.88% accuracy, 0.864 F1-score, and 0.845 MCC on the CEG hold-out test set, while sequence-derived features (TNC and AAC) emerge as the dominant predictors. Sequence-based features (TNC and AAC) dominated the selected feature set, with FT features accounting for 18 of the 45 selected features, confirming the value of genomic spectral signal processing as a complement to compositional representation. Overall, this study demonstrates the value of integrating genomic signal processing with boosting-based learning and provides a reproducible, leakage-controlled framework for essential gene classification that can inform future cross-organism prediction studies.]]>
Copyrights © 2026
