Protein coding regions prediction by fusing DNA shape features.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Protein-coding regions are challenging to predict due to varying exon and intron lengths.
• Current deep learning methods often overlook DNA shape's role in gene expression.
• Structural complexity in genomes hinders accurate prediction of coding regions.

Purpose of the Study:

• To develop an improved method for predicting protein-coding regions.
• To integrate DNA shape features into deep learning models for enhanced accuracy.
• To address limitations of existing methods in complex genomes.

Main Methods:

• Developed a Convolutional Neural Network-Bidirectional Recurrent Neural Network (CNNS-BRNN) model.
• Incorporated DNA shape features alongside traditional sequence features using a fusion coding technique.
• Evaluated model performance against baseline methods.

Main Results:

• The proposed CNNS-BRNN model significantly improved the prediction of protein-coding regions.
• The fusion coding technique combining DNA shape and sequence features enhanced model performance.
• Achieved AUC and F1 score improvements of 2.3% and 5.3%, respectively, over the baseline.

Conclusions:

• Integrating DNA shape features substantially boosts the performance of protein-coding region prediction models.
• The CNNS-BRNN model offers a more robust approach for analyzing complex biological genomes.
• This method advances the field of computational genomics and gene prediction.