DeepSE

DeepSE identifies super-enhancers (SEs)—clusters of active typical enhancers (TEs) enriched for the Mediator complex, master transcription factors, and chromatin regulators—using dna2vec sequence embeddings and a deep convolutional neural network to discriminate SEs from TEs based solely on DNA sequence.

Key Features:

• Deep convolutional neural network: Employs a deep convolutional neural network to learn discriminative features from sequence embeddings.
• dna2vec feature embeddings: Represents DNA k-mer sequences as continuous dna2vec vectors to capture sequence features.
• Sequence-only input: Operates using only DNA sequence information without requiring transcription factor ChIP-seq or chromatin mark datasets.
• SE versus TE classification: Specifically classifies super-enhancers (SEs) compared to typical enhancers (TEs).
• Training on DNA sequences: Trains the model on DNA sequence data where dna2vec embeddings encode input features.
• Cross-cell-line generalization: Demonstrates generalization across multiple cell lines, indicating conserved SE sequence patterns.
• Performance: Reports performance that surpasses existing state-of-the-art methods for SE identification from sequence.

Scientific Applications:

• Super-enhancer identification: Identify and annotate super-enhancers from genomic DNA sequences.
• Gene regulation and cell identity studies: Investigate sequence determinants of gene regulation and cell identity mediated by SEs.
• Cross-cell-line comparative analyses: Compare SE sequence patterns across different cell lines.
• Disease mechanism research: Support analyses linking SEs to disease mechanisms and regulatory dysregulation.

Methodology:

DeepSE trains a deep convolutional neural network on dna2vec-encoded DNA sequences to learn features that distinguish super-enhancers from typical enhancers.