Prompt-CBP: A Novel Prompt Learning-Based Model for Predicting Cross-Species Promoters

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Promoter sequences show both species-specific and conserved features.
• Species-specific sequences reduce cross-species prediction performance.
• Conserved motifs (TATA box, Inr, BRE, DPE) are vital for cross-species prediction.

Purpose of the Study:

• To develop a novel prompt learning-based model, Prompt-CBP, for enhanced cross-species promoter prediction.
• To leverage conserved promoter features while minimizing species-specific influences.
• To improve the accuracy and efficiency of predicting promoter sequences across different species.

Main Methods:

• A prompt generator creates targeted prompts based on conserved promoter features.
• DNABERT-2 encodes promoter sequences and generated prompts.
• Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) extract local and global sequence features.
• A fully connected layer performs binary classification for promoter prediction.

Main Results:

• Prompt-CBP significantly improved average accuracy (ACC) by 0.3213 and average area under the curve (AUC) by 0.4137 compared to methods without prompt learning.
• The model demonstrated comparable prediction performance to models trained on extensive data, even with smaller sample sizes.
• Prompt learning effectively guided the model to focus on conserved cross-species promoter features.

Conclusions:

• Prompt-CBP offers a powerful approach for accurate cross-species promoter prediction.
• The prompt learning strategy effectively addresses the challenge of species-specific variations in promoter sequences.
• This method enhances prediction performance and efficiency, particularly beneficial for limited datasets.