iProL: identifying DNA promoters from sequence information based on Longformer pre-trained model

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Promoters are crucial DNA sequences regulating gene transcription.
• Efficient computational identification of promoters is vital for molecular biology and genetics.
• Existing promoter identification methods face challenges with imbalanced sample recognition and require improvement.

Purpose of the Study:

• To develop an advanced computational model for accurate promoter identification.
• To address the limitations of existing methods in recognizing both positive and negative samples.
• To enhance the understanding and prediction of E. coli promoters.

Main Methods:

• Utilized the Longformer pre-trained model from natural language processing.
• Integrated one-dimensional convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM).
• Treated DNA sequences as plain text, requiring no prior biological knowledge.

Main Results:

• The iProL model demonstrated superior and more balanced performance compared to existing methods.
• Experimental results confirmed iProL's effectiveness on a novel, independent test set.
• The model successfully extracted both local and global features from DNA sequences.

Conclusions:

• iProL represents a significant advancement in computational promoter identification.
• The model offers a more accurate and balanced approach to recognizing promoter sequences.
• iProL's performance suggests its potential for broader applications in genomic research.