ES-ARCNN

ES-ARCNN predicts enhancer strength by using data augmentation and a residual convolutional neural network to distinguish strong and weak enhancers.

Key Features:

• Data Augmentation: Employs reverse complement and shift augmentation strategies to expand the enhancer training dataset.
• Residual Convolutional Neural Network (CNN): Uses a residual CNN architecture to learn sequence patterns and features relevant to enhancer strength prediction.
• Performance: Achieved 66.17% accuracy in 10-fold cross-validation and 65.5% accuracy on an independent dataset, exceeding competing methods by over 4%.
• Mechanistic Insights: Supports predictions with transcription factor binding sites (TFBSs) enrichment analysis, indicating a correlation between enhancer strength and TFBS density in specific tissues.

Scientific Applications:

• Gene regulation and tissue-specific expression: Predicts strong versus weak enhancers to support studies of transcriptional regulation and tissue-specific expression patterns.
• Cellular differentiation and development: Aids investigation of regulatory mechanisms underlying cellular differentiation and development.
• Therapeutic target identification and personalized medicine: Assists identification of potential regulatory targets for therapeutic intervention and personalized medicine.

Methodology:

Computational steps explicitly include reverse complement and shift data augmentation, training a residual CNN, evaluation by 10-fold cross-validation and independent dataset testing, and TFBS enrichment analysis.