PHDcleav

PHDcleav predicts Dicer cleavage sites within precursor microRNAs (pre-miRNAs) to identify precise RNase III processing positions critical for RNA interference and gene regulation.

Key Features:

• Algorithm: Uses a Support Vector Machine (SVM) trained on experimentally validated human miRNA hairpins from miRBase.
• Sequence window: Analyzes fourteen nucleotides surrounding cleavage sites, with the highest baseline accuracy obtained using a binary profile of the 5p arm (66%).
• Structural integration: Incorporates secondary structure information, increasing accuracy from 66% to 86%.
• Training and validation: Models were trained and tested on 555 experimentally validated cleavage sites using 5-fold cross-validation.
• Independent test performance: Independent testing achieved approximately 82% accuracy.
• Biological focus: Targets Dicer (RNase III) processing positions in pre-miRNAs to assess effects on miRNA seed regions and target repertoires.

Scientific Applications:

• SNP and variant impact analysis: Assess functional consequences of genetic variations or SNPs that alter Dicer cleavage sites and downstream gene silencing.
• Novel miRNA discovery: Aid discovery of novel miRNAs in the human genome by predicting likely Dicer cleavage positions.
• Pre-miRNA design for gene silencing: Support design of Dicer-specific pre-miRNAs for gene silencing by predicting processing positions.

Methodology:

Employs a Support Vector Machine trained on experimentally validated human miRNA hairpins from miRBase, analyzing a 14-nucleotide window around cleavage sites with a binary profile of the 5p arm, integrating secondary structure, and trained/tested on 555 sites with 5-fold cross-validation (independent test ≈82% accuracy).