A deep learning framework for enhancer prediction using word embedding and sequence generation.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Enhancers are non-coding DNA fragments regulating gene transcription.
• Accurate enhancer identification is vital for understanding gene expression, human diseases, and cancer development.
• Identifying enhancers is challenging due to sequence variability and dispersion.

Purpose of the Study:

• To propose a novel deep learning framework for accurate enhancer identification and classification.
• To address the challenge of limited benchmark dataset size for enhancer identification.
• To leverage natural language processing techniques for DNA sequence analysis.

Main Methods:

• Utilized RankGAN for dataset augmentation while preserving data characteristics.
• Applied FastText word embedding to represent DNA sequences as numerical matrices.
• Developed a Long Short-Term Memory Convolutional Neural Network (LSTM-CNN) for feature extraction and identification.

Main Results:

• Achieved an accuracy of 0.7525 and MCC of 0.5051 for enhancer prediction on an independent test set.
• Obtained an accuracy of 0.6972 and MCC of 0.3954 for enhancer type prediction.
• Demonstrated superior performance compared to existing methods in enhancer prediction and classification.

Conclusions:

• The proposed deep learning framework effectively identifies and classifies DNA enhancers.
• The study highlights the potential of natural language processing in advancing bioinformatics.
• This approach offers a promising direction for improving the understanding and diagnosis of gene-related diseases.