miRWoods

miRWoods detects and predicts microRNAs and their precursor spans using a duplex-focused precursor detection method combined with stacked random forests to improve microRNA annotation and discovery across species.

Key Features:

• Duplex-focused precursor detection method: Employs a duplex-focused precursor detection method to locate hairpin precursors of microRNAs.
• Stacked random forests: Uses stacked random forests as the core classification framework for microRNA prediction.
• Specialized detection layers: Incorporates specialized layers specifically for detecting both mature and precursor microRNAs.
• Optimization metric: Tuned to optimize the harmonic mean of precision and recall to balance sensitivity and specificity.
• Training and evaluation datasets: Developed and tuned on well-annotated human genome datasets and evaluated on mouse data.
• Improved recall and low-read detection: Recalls an average of 10% more annotated microRNAs compared to other approaches and predicts microRNAs with minimal read counts.
• Application to under-annotated genomes: Applied to Felis catus and Bos taurus small RNA sequencing datasets, revealing hundreds of novel microRNAs.
• Sample diversity: Demonstrated on muscle and skin samples from cats, ten different tissues from cows, and human and mouse cells.
• Specific discoveries: Identified a microRNA within an intron of TYK2 in both cat and cow genomes and detected instances of mirtrons in the human genome.
• miRNA family expansion: Supports an expanded miR-2284 family in bovine, a larger mir-548 family in humans, and an enlarged let-7 family in felines.

Scientific Applications:

• MicroRNA annotation and discovery: Improves annotation of known microRNAs and discovers novel microRNAs in under-annotated genomes.
• Cross-species miRNA comparison: Facilitates comparative analysis of microRNA repertoires across human, mouse, Felis catus, and Bos taurus.
• Low-abundance miRNA detection: Detects and predicts microRNAs present at minimal read counts in small RNA sequencing datasets.
• Intronic miRNA and mirtron identification: Identifies intronic microRNAs such as the TYK2 intronic microRNA and human mirtrons.
• miRNA family expansion studies: Supports expansion and characterization of miRNA families including miR-2284, mir-548, and let-7 across species.

Methodology:

miRWoods applies a duplex-focused precursor detection method combined with stacked random forests and specialized layers for mature and precursor detection, tuned to optimize the harmonic mean of precision and recall; development used well-annotated human genome datasets and evaluation used mouse data.