CORENup

CORENup identifies nucleosome positions in eukaryotic genomes from DNA sequences using a parallel deep learning architecture.

Key Features:

• Input representation: DNA sequences are encoded using a one-hot representation as model input.
• Parallel architecture: A parallel architecture integrates a fully convolutional neural network (CNN) and a recurrent layer.
• Feature capture: The model captures both non-periodic and periodic sequence features relevant to nucleosome positioning.
• CNN component: The fully convolutional CNN detects local patterns and sequence motifs.
• Recurrent component: The recurrent layer captures long-range dependencies and sequential information.
• Integration and classification: A dense layer combines outputs from the parallel components to produce a final nucleosome-positioning classification.
• Benchmarking and performance: Empirical evaluations on public datasets from various organisms and comparisons using two groups of datasets reported superior classification metrics and computational efficiency versus leading methods.

Scientific Applications:

• Genome-wide nucleosome mapping: Identification of nucleosome positions at genomic scale from DNA sequence data.
• Regulation and chromatin studies: Investigation of nucleosome dynamics and their regulatory roles in transcription and chromatin organization.
• Method benchmarking: Comparative evaluation and benchmarking of nucleosome-positioning prediction methods using public datasets.

Methodology:

DNA sequences are one-hot encoded and processed by a parallel architecture comprising a fully convolutional neural network and a recurrent layer, whose outputs are combined by a dense layer for final classification; performance was evaluated on public datasets and benchmarked using two groups of datasets.