SG-DCNN: A Deep Learning Method Integrating Self-Attention Mechanism and Generative Adversarial Network for...

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein functions depend on ion ligand binding.
• Accurate prediction of protein-ion ligand binding residues is vital.
• Small sample sizes and class imbalance challenge accurate prediction.

Purpose of the Study:

• To develop a novel algorithm for enhancing the prediction accuracy of protein-ion ligand binding residues.
• To address the challenges posed by small sample sizes and class imbalance in ion ligand prediction.

Main Methods:

• Introduced the SG-DCNN algorithm, integrating Generative Adversarial Network (GAN) and Self-Attention within a Deep Convolutional Neural Network (DCNN).
• Conducted theoretical analysis to assess the algorithm's effectiveness.
• Performed experimental validation using eight small sample ion ligands.

Main Results:

• The SG-DCNN algorithm demonstrated enhanced prediction accuracy.
• Achieved an independent testing accuracy of 78.3% for Sn.
• Obtained a Matthews correlation coefficient (Mcc) of 0.23.
• Outperformed previous prediction methods in accuracy.

Conclusions:

• The SG-DCNN algorithm effectively enhances the prediction of binding residues for small sample ion ligands.
• This method offers a promising approach for improving our understanding of protein functions reliant on ion binding.