Structure-Aware Heterogeneous Information Fusion Framework for Protein-Ligand Binding Affinity Prediction.

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Accurate prediction of protein-ligand binding affinities (PLAs) is vital for efficient drug discovery and development.
• Current methods often overlook the benefits of heterogeneous graph augmentation and multimodal data integration (sequence and structure).

Purpose of the Study:

• To develop a novel multimodal data fusion approach, GIF-PLA, for enhanced protein-ligand binding affinity prediction.
• To improve the generalization and robustness of predictive models by incorporating complementary information from different data modalities.

Main Methods:

• Representing protein-ligand complexes as heterogeneous graphs with meta-paths.
• Utilizing cascaded deep neural networks to process graph, protein sequence, and ligand SMILES string data in parallel.
• Employing a late fusion module to integrate multilevel information for final prediction.

Main Results:

• GIF-PLA achieved a Pearson's correlation coefficient (Rp) of 0.784 and a root-mean-square error (RMSE) of 1.157 on benchmark datasets.
• Demonstrated superior performance compared to state-of-the-art methods.
• Ablation studies confirmed the significant contributions of meta-paths and multimodal fusion.

Conclusions:

• GIF-PLA effectively captures both structure-oriented and sequence-oriented information for accurate PLA prediction.
• The proposed method shows significant promise for enhancing the reliability of protein-ligand interaction predictions.
• Multimodal data fusion and meta-path augmentation are critical for advancing PLA prediction accuracy.