DeePromoter: Robust Promoter Predictor Using Deep Learning.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Promoter regions are crucial for gene transcription initiation, regulating RNA polymerase binding.
• Accurate promoter recognition is vital in bioinformatics, but existing prediction tools require improvement.

Purpose of the Study:

• To develop a robust deep learning model, DeePromoter, for accurate recognition of human and mouse eukaryotic promoter sequences.
• To enhance the reliability and reduce false positives in promoter prediction.

Main Methods:

• Developed DeePromoter, a deep learning model combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM).
• Introduced a novel method for constructing a challenging negative set derived from promoter sequences, improving discrimination.
• Applied the model to analyze short eukaryotic promoter sequences.

Main Results:

• DeePromoter demonstrated superior performance in recognizing human and mouse promoter sequences compared to existing tools.
• The proposed negative set reconstruction method significantly reduced false positive predictions.
• The model achieved high accuracy in analyzing promoter characteristics.

Conclusions:

• DeePromoter offers a significant advancement in promoter prediction accuracy and reliability.
• The novel negative set strategy is effective in improving model discrimination.
• A web server is available for public use, facilitating promoter prediction research.