TransBind allows precise detection of DNA-binding proteins and residues using language models and deep learning

Area of Science:

• Computational biology
• Bioinformatics
• Molecular biology

Background:

• Identifying DNA-binding proteins and residues is crucial for understanding biological processes.
• Experimental methods are slow and costly.
• Current machine learning methods struggle with accuracy, data imbalance, and reliance on multiple sequence alignments (MSAs).

Purpose of the Study:

• To develop an accurate and efficient alignment-free deep learning framework for predicting DNA-binding proteins and residues.
• To overcome the limitations of existing methods, particularly for orphan or rapidly evolving proteins.

Main Methods:

• Developed TransBind, an alignment-free deep learning framework.
• Utilized features from pre-trained protein language models.
• Predicted DNA-binding proteins and residues directly from single primary sequences.

Main Results:

• TransBind significantly outperforms state-of-the-art methods in accuracy and computational efficiency.
• The framework effectively handles data imbalance issues.
• Demonstrated superior performance through extensive evaluations and case studies.

Conclusions:

• TransBind offers a powerful and efficient solution for identifying DNA-binding proteins and residues.
• The alignment-free approach makes it suitable for a wider range of proteins, including orphan and rapidly evolving ones.
• TransBind is accessible via a web server for broader research application.