NCResNet

NCResNet predicts noncoding RNAs from RNA sequences using a deep learning approach that integrates 57 hybrid features to improve identification of ncRNAs across short- and long-ORF transcripts from next-generation sequencing data.

Key Features:

• Hybrid feature set: Uses 57 hybrid features grouped into four categories: sequence features, protein-related features, RNA structure characteristics, and RNA physicochemical properties.
• Feature enhancement: Applies feature enhancement techniques to improve the discriminative power of input features.
• Deep feature learning: Employs deep feature learning policies within a neural network model to learn representations for ncRNA prediction.
• Benchmark evaluation: Evaluated on benchmark datasets from eight species and reported higher accuracy and Matthews correlation coefficient (MCC) than existing state-of-the-art methods.
• Short-ORF specialization: Demonstrates significant improvements for short-ORF RNAs in species such as mouse, Saccharomyces cerevisiae, zebrafish, and cow, with accuracy gains exceeding 10% and MCC gains exceeding 15%.
• Versatility across ORF lengths: Maintains competitive predictive performance on long-ORF RNA sequence datasets.

Scientific Applications:

• ncRNA discovery: Identification of noncoding RNAs from unannotated RNA sequences derived from next-generation sequencing.
• Transcript characterization: Distinguishing ncRNAs across short- and long-ORF transcripts to support transcriptome annotation.
• Cross-species annotation: Improving ncRNA annotation and comparative analyses in species including mouse, Saccharomyces cerevisiae, zebrafish, and cow.
• Functional and disease studies: Facilitating discovery of ncRNAs implicated in biological processes, regulatory mechanisms, diseases, and cancers.

Methodology:

Implements a deep learning–based neural network that combines feature enhancement and deep feature learning applied to 57 hybrid features grouped into sequence, protein-related, structural, and physicochemical categories, with performance evaluated on benchmark datasets from eight species.