An integrative computational framework based on a two-step random forest algorithm improves prediction of...

Area of Science:

• Biochemistry
• Bioinformatics
• Structural Biology

Background:

• Zinc-binding proteins are crucial metalloproteins with diverse functional roles.
• Accurate prediction of zinc-binding sites aids in inferring protein function and predicting 3D structure.

Purpose of the Study:

• To develop an integrative framework for improved prediction of zinc-binding sites.
• To investigate sequence, structure, and network features relevant to zinc-binding site prediction.

Main Methods:

• Combined multiple sequence, structural properties, and graph-theoretic network features.
• Employed a two-step feature selection using random forest for efficiency and importance quantification.
• Benchmarked against existing methods (SitePredict, zincfinder) on a dataset of 1,103 protein chains.

Main Results:

• Achieved >80% recall at 75% precision for key residues (Cys, His, Glu, Asp), outperforming other methods by 10%-28%.
• Independent tests showed superior recall at both residue (0.790) and protein (0.759) levels.
• Demonstrated better Area Under the Curve (AUC) and Area Under the Recall-Precision Curve (AURPC) compared to existing tools.

Conclusions:

• The developed method offers a robust approach for large-scale zinc-binding site identification.
• Provides valuable insights into features characterizing zinc-binding sites across different levels.
• The framework enhances protein function and structural prediction capabilities.