SubMito-XGBoost

SubMito-XGBoost predicts protein submitochondrial localization to support studies of mitochondrial function and disease-related protein targeting.

Key Features:

• Feature Extraction: Extracts sequence features using g-gap dipeptide composition (g-gap DC), pseudo-amino acid composition (PseAAC), auto-correlation function (ACF), and bi-gram position-specific scoring matrix (Bi-gram PSSM).
• Data Balancing: Balances class distributions using the Synthetic Minority Oversampling Technique (SMOTE).
• Feature Selection: Selects informative features with the ReliefF algorithm.
• Prediction Model: Uses eXtreme Gradient Boosting (XGBoost) for classification of submitochondrial localization.
• Validation and Performance: Evaluates by leave-one-out cross-validation (LOOCV) with accuracies of 97.7% on M317 and 98.9% on M983, and 94.8% on independent test set M495, representing improvements of 2.8–12.5% (M317) and 3.8–9.9% (M983) over other methods.
• Cross-species Robustness: Demonstrates robust performance across both plant and non-plant protein datasets.

Scientific Applications:

• Mitochondrial function studies: Supports analysis of protein roles within mitochondrial compartments to investigate mitochondrial biology.
• Disease mechanism research: Aids investigation of diseases associated with mitochondrial dysfunction, including Parkinson's disease, multifactor disorder, and Type-II diabetes.
• Drug design: Assists identification of mitochondrial-targeted proteins relevant to therapeutic development.
• Comparative localization analysis: Enables study of submitochondrial protein targeting across plant and non-plant species.

Methodology:

Extracts features (g-gap DC, PseAAC, ACF, Bi-gram PSSM) from protein sequences, applies SMOTE for oversampling, uses ReliefF for feature selection, trains an XGBoost classifier, and evaluates performance with LOOCV on M317 and M983 and independent testing on M495.