PLPMpro: Enhancing promoter sequence prediction with prompt-learning based pre-trained language model.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Promoter regions are crucial for gene transcription initiation, making their accurate identification a key bioinformatics challenge.
• While pre-trained language models (PLMs) show promise for promoter prediction, their full potential is yet to be realized.

Purpose of the Study:

• To introduce PLPMpro, a novel model that leverages prompt learning with PLMs to improve promoter sequence prediction.
• To demonstrate the effectiveness of prompt learning in enhancing PLM capabilities for biological sequence analysis.

Main Methods:

• Developed PLPMpro, a model integrating prompt learning with pre-trained language models.
• Conducted experiments to evaluate PLPMpro's performance against existing promoter prediction methods.
• Performed detailed analyses of prompt learning settings and soft module configurations.
• Utilized interpretation experiments to assess the biological relevance captured by the pre-trained model.

Main Results:

• PLPMpro significantly improved the prediction accuracy of promoter sequences.
• The model demonstrated superior performance compared to standard PLM-based and deep learning methods for promoter prediction.
• Prompt learning was shown to be effective in enhancing the predictive power of PLMs.
• Interpretation experiments confirmed that the PLM captures underlying biological semantics.

Conclusions:

• Prompt learning offers a powerful approach to maximize the utility of PLMs in bioinformatics.
• PLPMpro represents a significant advancement in promoter prediction, offering a novel perspective on applying PLMs to biological problems.