EnhancerMatcher: comparing cell-type-specific enhancer activity of DNA sequences using deep convolutional neural...

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Transcriptional enhancers are crucial regulatory elements but challenging to identify computationally due to their variable genomic locations and orientations relative to target genes.
• The limited availability of experimentally validated enhancers hinders the development of accurate machine learning models for enhancer prediction.

Purpose of the Study:

• To develop a novel computational tool, EnhancerMatcher, for accurate and efficient identification of cell-type-specific transcriptional enhancers.
• To overcome the limitations of existing methods by enabling enhancer prediction across diverse cell types using minimal reference data.

Main Methods:

• Developed EnhancerMatcher, a convolutional neural network (CNN)-based tool utilizing two confirmed enhancers as references for sequence classification.
• Trained the model on putative enhancers from the CATlas Project and human genomic control sequences.
• Evaluated enhancer activity by classifying sequences in triplets: two known enhancers and a third test sequence.

Main Results:

• EnhancerMatcher achieved high performance on human test data, with 90% accuracy, 92% recall, and 87% specificity.
• Demonstrated strong cross-species generalization, successfully identifying mouse enhancers using a human-trained model.
• Showcased consistent performance across various cell types, irrespective of data representation size, and provided interpretability via class activation maps.

Conclusions:

• EnhancerMatcher is a powerful, versatile, and generalizable tool for enhancer discovery and regulatory sequence analysis.
• The CNN-based approach effectively leverages minimal reference enhancers for robust cell-type-specific enhancer identification.
• The tool's ability to perform cross-species prediction and provide interpretable results enhances its utility in genomic research.