Artificial intelligence in metalloprotein binding site prediction: A systematic review bridging bioinformatics and...

Metalloproteins, accounting for nearly half of all known proteins, are fundamental to biological processes and applications, including catalysis, electron transport and pharmaceutical engineering. Accurately predicting metal-binding sites is critical for elucidating molecular interactions and advancing innovations in the biotechnology and pharmaceutical industries. Traditional sequence- and structure-based prediction methods often struggle with the complexity and diversity of metalloprotein datasets. In contrast, machine learning (ML) and deep learning (DL) techniques have emerged as transformative tools that offer superior performance in handling nonlinear biological data. This systematic review follows Preferred Reporting Items for Systematic Reviews (PRISMA) and Meta-Analyses guidelines. The analysis revealed that Random Forest models achieved up to 99 % accuracy in sequence-based predictions, whereas convolutional neural networks (CNNs) reached up to 96 % accuracy in structural data. Despite these advancements, challenges such as data imbalance, underrepresentation of metal ions, and structural heterogeneity limit the generalizability of the models. To address these issues, we propose a structured decision-making framework to guide researchers in selecting appropriate models based on dataset characteristics, research goals, and performance trade-offs. This review serves as a practical resource for bioinformaticians and bioengineers, facilitating AI-driven methods in metalloprotein research, and supporting design of novel proteins for pharmaceutical industrial purposes.