PromGER: Promoter Prediction Based on Graph Embedding and Ensemble Learning for Eukaryotic Sequence.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Promoters are crucial DNA regulatory regions controlling gene transcription.
• Accurate promoter prediction is vital for understanding gene function and transcriptional activity.
• Existing machine and deep learning models have limitations in capturing complex promoter sequence relationships and biological information.

Purpose of the Study:

• To develop a novel prediction model, PromGER, for eukaryotic promoter sequences.
• To overcome limitations of current models by extracting deeper biological information and considering complex sequence relationships.
• To improve the accuracy and interpretability of promoter prediction.

Main Methods:

• PromGER employs four feature-encoding methods to extract local information from promoter sequences.
• Sequences are represented as graphs to capture potential relationships, with graph-embedding methods used for global feature extraction.
• A tree-based ensemble-learning framework integrates local and global features for prediction.

Main Results:

• PromGER demonstrated superior performance compared to seven existing methods.
• Significant improvements were observed in specificity (13%), accuracy (10%), Matthew's correlation coefficient (16%), precision (4%), F1 score (6%), and AUC (9%).
• Model interpretability was confirmed using t-distributed stochastic neighbor embedding (t-SNE) and SHAPley Additive exPlanations (SHAP) analysis.

Conclusions:

• PromGER offers a powerful and interpretable approach for eukaryotic promoter sequence prediction.
• The integration of local and global features within a graph-based framework enhances predictive accuracy.
• This study advances bioinformatics tools for gene regulation research.