Predicting Ca²⁺ and Mg²⁺ ligand binding sites by deep neural network algorithm.

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Alkaline earth metal ions, such as Mg²⁺ and Ca²⁺, are vital protein binding ligands in the human body.
• Accurate prediction of their binding residues is crucial for understanding biological processes.

Purpose of the Study:

• To develop and validate an efficient computational method for predicting Mg²⁺ and Ca²⁺ ligand binding sites in proteins.

Main Methods:

• Utilized deep neural network algorithms incorporating protein sequence characteristics, amino acid properties, and predicted structural information.
• Optimized deep learning hyper-parameters and employed a fivefold cross-validation strategy.
• Applied an undersampling data processing technique for independent testing.

Main Results:

• The developed deep learning model demonstrated superior prediction accuracy compared to the Ionseq method.
• Performance on independent test sets surpassed that of support vector machine algorithms.
• Optimized model achieved better prediction results through rigorous validation.

Conclusions:

• An efficient and accurate method for predicting Mg²⁺ and Ca²⁺ ligand binding sites has been successfully developed.
• The findings contribute to a better understanding of protein-ligand interactions involving alkaline earth metal ions.