ReCGBM

ReCGBM predicts human Dicer cleavage sites in pre-miRNAs to support analysis of microRNA biogenesis and gene regulation.

Key Features:

• Prediction target: Predicts cleavage sites of human Dicer on precursor microRNAs (pre-miRNAs).
• Algorithm: Implements gradient boosting using the lightGBM algorithm.
• Feature types: Incorporates relational and class features derived from input sequences.
• Relational feature design: Uses specifically designed features that capture interactions between different sequences.
• Interpretability: Provides feature importance analysis to identify influential sequence characteristics.
• Central-region emphasis: Highlights the significant contribution of features located near the center of pre-miRNA molecules.
• Benchmarking: Demonstrates improved prediction accuracy relative to PHDCleav and LBSizeCleav in computational experiments.

Scientific Applications:

• RNA processing studies: Supports investigation of Dicer cleavage specificity and mechanisms of pre-miRNA processing.
• Gene regulation research: Aids studies of microRNA biogenesis and its effects on post-transcriptional gene regulation.
• Model development: Guides development of predictive models by identifying informative sequence features, especially near pre-miRNA centers.

Methodology:

Uses gradient boosting with lightGBM, engineered relational and class features that capture inter-sequence interactions, feature importance analysis for interpretability, and computational benchmarking against PHDCleav and LBSizeCleav.