ES-ARCNN: Predicting enhancer strength by using data augmentation and residual convolutional neural network.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Enhancers are non-coding DNA sequences regulating gene transcription, vital for cell and tissue development.
• Experimental validation of enhancers is resource-intensive, driving the need for computational prediction methods.
• Existing computational methods often struggle with accurate enhancer strength prediction.

Purpose of the Study:

• To develop an accurate computational method for predicting enhancer strength (strong vs. weak).
• To improve upon the performance of current state-of-the-art enhancer prediction tools.
• To provide a user-friendly web application for enhancer strength prediction.

Main Methods:

• Utilized data augmentation techniques (reverse complement and shift) to expand the enhancer dataset.
• Employed a Residual Convolutional Neural Network (ES-ARCNN) model for enhancer strength prediction.
• Trained and validated the model using 10-fold cross-validation and an independent test dataset.

Main Results:

• ES-ARCNN achieved the highest accuracy (66.17%) in 10-fold cross-validation compared to other methods.
• The method demonstrated improved performance on an independent dataset with 65.5% accuracy, exceeding existing methods by over 4%.
• Transcription factor binding site (TFBS) enrichment analysis revealed a correlation between enhancer strength and TFBS density.

Conclusions:

• ES-ARCNN effectively predicts enhancer strength, outperforming current state-of-the-art computational methods.
• Data augmentation significantly enhances prediction performance.
• The study provides mechanistic insights into enhancer strength and offers a practical web tool for researchers.