Deep-loop

Deep-loop predicts CTCF-mediated chromatin loops from DNA sequence using deep learning to study three-dimensional genome architecture relevant to transcription regulation and DNA replication.

Key Features:

• Convolutional Neural Network (CNN) architecture: A sequence-based deep learning CNN integrates multiple genomic feature encodings to predict chromatin loops.
• k-tuple Nucleotide Frequency Component: Captures frequency patterns of nucleotide k-tuples from DNA sequence input.
• Nucleotide Pair Spectrum Encoding: Encodes information about nucleotide pair relationships within sequences.
• Position Conservation and Position Scoring Function: Assesses evolutionary conservation and positional importance in the genomic sequence.
• Natural Vector Features: Provides additional contextual numerical descriptors derived from sequence data.
• Performance evaluation: Model performance was assessed using rigorous cross-validation demonstrating accurate identification of CTCF-mediated loops across cell types.

Scientific Applications:

• Chromatin loop prediction: Predicts CTCF-mediated chromatin loops from genomic sequence data.
• 3D genome architecture studies: Enables analysis of three-dimensional chromosomal organization relevant to gene regulation.
• Transcription regulation and DNA replication research: Supports investigation of loop-mediated impacts on transcription regulation and DNA replication.
• Cross-cell-type analysis: Facilitates large-scale prediction of chromatin loops across diverse cell types.

Methodology:

Deep-loop employs a sequence-based convolutional neural network trained on inputs encoded by k-tuple nucleotide frequency, nucleotide pair spectrum, position conservation and scoring functions, and natural vector features, with performance assessed by cross-validation.