MuStARD

MuStARD applies convolutional neural networks to detect and classify small RNA genomic loci, enabling discovery of novel small RNA genes (including pre-miRNAs and snoRNAs) and cross-species prediction.

Key Features:

• Pattern Recognition: Utilizes convolutional neural networks (CNNs) to detect characteristic patterns associated with small RNA genes and identify novel genomic loci similar to user-defined regions.
• Automated Feature Selection: Incorporates automated processes for feature and background selection, eliminating the need for manual domain-specific selection.
• Inter-species Prediction: Predicts functional elements in one species using models trained on another, for example predicting mouse pre-miRNAs and snoRNAs with models trained on the human genome.
• Dataset Filtering: Filters small RNA-Seq datasets to identify novel small RNA loci within and across species, demonstrated on human, mouse, and fly pre-miRNA prediction.

Scientific Applications:

• Small RNA locus identification: Discovery and classification of small RNA loci, including novel pre-miRNAs and snoRNAs.
• Comparative genomics: Cross-species model transfer for comparative genomics and evolutionary analysis of small RNA genes.
• Gene regulation studies: Exploration of genomic regions encoding small RNAs to investigate their roles in gene regulation.

Methodology:

Employs a multi-branch convolutional neural network architecture that learns from user-defined genomic regions, scans large genomic areas to identify novel regions with similar characteristics, and uses automated feature and background selection.