miRe2e

miRe2e predicts precursor microRNAs (pre-miRNAs) from raw genome-wide sequence data using an end-to-end Transformer-based deep learning model.

Key Features:

• Transformer architecture: Implements an end-to-end deep learning model based on the Transformer architecture.
• Attention mechanisms: Employs the Transformer's attention mechanisms to infer global dependencies between inputs and outputs.
• Raw input processing: Processes raw genome-wide sequences without requiring secondary structure prediction, pre-processing, or handcrafted feature engineering.
• Automatic feature extraction: Learns relevant sequence representations automatically via deep learning instead of handcrafted feature extraction.
• Training data: Trained on datasets comprising pre-miRNAs, hairpin, and non-hairpin sequences.
• Genome-wide prediction: Performs genome-wide identification of pre-miRNA sequences.
• Validation and benchmarking: Validated on the human genome and benchmarked against state-of-the-art algorithms, reporting an order-of-magnitude (tenfold) performance improvement in comparative analyses.

Scientific Applications:

• De novo pre-miRNA prediction: Identification of novel pre-miRNAs directly from genomic sequence data.
• Genome-wide scanning: Genome-wide detection of pre-miRNA loci, including analyses on the human genome.
• miRNA research and disease studies: Support for studies of miRNA involvement in gene expression regulation and associations with complex human diseases.
• Benchmarking of prediction algorithms: Comparative evaluation and benchmarking against existing state-of-the-art pre-miRNA prediction methods.

Methodology:

Uses an end-to-end Transformer-based deep learning model with attention mechanisms, trained on pre-miRNA, hairpin, and non-hairpin sequence datasets to operate on raw genome-wide sequences without secondary-structure prediction or handcrafted features; validated on the human genome and compared to state-of-the-art algorithms.