XG-ac4C

XG-ac4C identifies N4-acetylcytidine (ac4C) sites in mRNA using machine learning to enable computational detection of this post-transcriptional modification and support study of its effects on mRNA stability and translation.

Key Features:

• XGBoost classifier: Implements eXtreme Gradient Boosting (XGBoost) as the core machine-learning algorithm for ac4C site prediction.
• Electron-ion interaction pseudopotentials: Employs electron-ion interaction pseudopotentials and their trinucleotide counterparts at ac4C sites as input features.
• Interpretability: Integrates Shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME) to elucidate feature contributions.
• Performance (AUPR): Reports improvements in area under the precision-recall curve of 9.4% in cross-validation and 9.6% in independent tests relative to existing methods.

Scientific Applications:

• ac4C site mapping: Identification of N4-acetylcytidine sites in mRNA for mapping modification landscapes.
• mRNA stability and translation studies: Enabling analyses of how ac4C correlates with mRNA stability and translational regulation.
• Genomics and transcriptomics: Supporting studies in genomics and transcriptomics that require precise identification of post-transcriptional modifications.

Methodology:

Uses the eXtreme Gradient Boosting (XGBoost) algorithm with electron-ion interaction pseudopotentials and trinucleotide counterparts as features, applies SHAP and LIME for feature importance analysis, and evaluates performance with cross-validation and independent tests reporting AUPR improvements of 9.4% and 9.6%.