NCYPred

NCYPred classifies short non-coding RNAs (sncRNAs), focusing on Y RNAs, to identify Y RNA transcripts and homologs across vertebrates, nematodes, insects, and bacteria using learned sequence representations from deep learning.

Key Features:

• Attention-based Bidirectional LSTM network: Employs a bidirectional Long Short-Term Memory (LSTM) network enhanced with an attention mechanism to capture complex dependencies in nucleotide sequences.
• Comprehensive dataset: Trained on 45,447 sncRNA sequences sourced from Rfam version 14.3.
• High classification accuracy across species and classes: Predicts Y RNA sequences and homologs across vertebrates, nematodes, insects, and bacteria and classifies 11 additional sncRNA classes with performance comparable to state-of-the-art methods.
• t-SNE on learned representations: Applies t-Distributed Stochastic Neighbor Embedding (t-SNE) to learned sequence representations for visualization and analysis of clustering and relationships among sncRNA classes.

Scientific Applications:

• Y RNA research: Characterizing Y RNAs implicated in DNA replication initiation and their potential role as tumor biomarkers.
• Genomic and transcriptomic annotation: Annotating and classifying sncRNAs in genomic and transcriptomic datasets across vertebrates, nematodes, insects, and bacteria.
• Biomarker discovery: Supporting identification of Y RNAs and other sncRNA classes in disease-focused studies.
• Comparative and evolutionary analysis: Enabling comparative analysis of Y RNA homologs and other sncRNA classes across multiple species.

Methodology:

Training a bidirectional LSTM network with an attention mechanism on 45,447 sncRNA sequences from Rfam v14.3 and applying t-Distributed Stochastic Neighbor Embedding (t-SNE) to the learned sequence representations.