MLDSPP: Bacterial Promoter Prediction Tool Using DNA Structural Properties with Machine Learning and Explainable AI

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Bacterial gene expression relies on promoter regions, but their identification is challenging due to genomic diversity.
• Accurate promoter identification is essential for understanding bacterial gene regulation.

Purpose of the Study:

• To develop MLDSPP, a novel machine learning tool for predicting bacterial promoter regions.
• To leverage DNA structural properties and advanced ML strategies for improved promoter prediction accuracy.

Main Methods:

• Utilized DNA duplex stability and base stacking properties with machine learning models (SVM, Random Forest, XGBoost).
• Employed one-hot encoding for nucleotide sequences and Shapley values for model interpretability.
• Validated MLDSPP against current tools (Sigma70pred, iPromoter2L) across 12 bacterial genomes.

Main Results:

• XGBoost model demonstrated superior performance, achieving F1-scores >95% in most cases.
• MLDSPP outperformed existing state-of-the-art promoter prediction tools.
• Explainable AI (Shapley values) enhanced model interpretability.

Conclusions:

• MLDSPP is a novel and effective tool for predicting bacterial promoter regions.
• The integration of DNA structural features and ML enhances predictive power.
• This tool can significantly advance bacterial genomics and gene regulation studies.