DNA protein binding recognition based on lifelong learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• DNA-binding protein prediction is crucial in bioinformatics.
• Existing statistical and machine learning methods show limitations in accuracy and speed.
• Advancements are needed for more efficient protein sequence analysis.

Purpose of the Study:

• To develop a novel dynamic deep network for DNA-binding protein classification.
• To enhance prediction accuracy and speed by fusing multiple evolutionary features.
• To explore the application of lifelong learning in bioinformatics classification tasks.

Main Methods:

• Feature extraction using Average Blocks, Discrete Cosine Transform, Discrete Wavelet Transform, Global encoding, Normalized Moreau-Broto Autocorrelation, and Pseudo position-specific scoring matrix.
• Development of a dynamic deep network with a lifelong learning architecture.
• Multi-feature fusion to improve the description of protein information.

Main Results:

• The proposed model demonstrated superior performance compared to other classification techniques.
• Achieved high single-sample specificity (81.0%, 83.0%) and sensitivity (82.4%, 90.7%).
• Attained high accuracy on benchmark datasets (88.4%, 80.0%, 76.6%).

Conclusions:

• The multi-feature fusion approach significantly enhances DNA-binding protein classification.
• The dynamic deep network based on lifelong learning offers a novel perspective for dichotomous classification problems.
• This research broadens the application of lifelong learning in bioinformatics.