A comprehensive review of computational methods for Protein-DNA binding site prediction

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Chemistry

Background:

• Identifying protein-DNA binding sites is vital for understanding biological processes and advancing drug discovery.
• Experimental methods for locating DNA-binding sites are accurate but costly and time-consuming.
• Efficient and accurate computational methods for DNA-binding site prediction are urgently needed.

Purpose of the Study:

• To review and categorize major computational approaches for predicting protein-DNA binding sites.
• To benchmark state-of-the-art DNA-binding site prediction models.
• To evaluate the performance of different computational methods, including deep learning.

Main Methods:

• Categorization of computational methods into template detection, statistical machine learning, and deep learning.
• Benchmarking of 14 state-of-the-art prediction models on 136 non-redundant proteins.
• Comparative analysis of model performance based on prediction accuracy and efficiency.

Main Results:

• Deep learning-based methods, especially those utilizing pre-trained large language models, demonstrate superior performance.
• These advanced methods outperform traditional template detection and statistical machine learning approaches.
• The study provides insights into the applications of various DNA-binding site prediction techniques.

Conclusions:

• Deep learning, particularly large language models, represents the most effective computational strategy for DNA-binding site prediction.
• These findings highlight the potential of AI in accelerating biological research and drug development.
• The reviewed methods offer valuable tools for researchers investigating protein-DNA interactions.