Predicting Promoters in Multiple Prokaryotes with Prompt.

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology

Background:

• Accurate prediction of prokaryotic promoters is vital for understanding gene regulation and function.
• Existing methods often lack cross-species applicability and interpretability due to limited focus on sequence motifs.

Purpose of the Study:

• To develop a generalized method, Prompt, for predicting promoters in multiple prokaryotic species.
• To enhance the interpretability of promoter predictions by incorporating sequence motif identification.

Main Methods:

• Prompt integrates Regression based on Selected k-mer (RSK), Contrastive Learning (CL), and Multilayer Perception (MLP).
• A voting strategy categorizes predictions into high-confidence and low-confidence sets.
• The method was evaluated on promoter prediction tasks across 16 prokaryotic species.

Main Results:

• Prompt achieved over 80% accuracy (Accuracy, Matthews correlation coefficient) on high-confidence datasets across 16 prokaryotes.
• Accuracy exceeded 90% for 12 of the 16 prokaryotes.
• Prompt demonstrated superior performance compared to existing prediction methods.

Conclusions:

• Prompt offers a generalized and interpretable approach for prokaryotic promoter prediction.
• The method's high accuracy and ability to identify motifs will aid in elucidating gene functions and regulatory mechanisms.
• Prompt is publicly available and poised to advance prokaryotic research.