Drug research and discovery can be substantially accelerated by computational methods for determining the protein-ligand binding affinity. Many deep learning-based models are currently put out to forecast the protein-ligand binding affinity and increase performance noticeably. Prediction of protein-ligand binding affinity, however, still faces significant obstacles. One difficulty is the difficulty of capturing the reciprocal information between proteins and ligands. Finding and highlighting the significant ligand and protein residue atoms is another difficult task.Drug research and discovery can be substantially accelerated by computational methods for determining the protein-ligand binding affinity. Many deep learning-based models are currently put out to forecast the protein-ligand binding affinity and increase performance noticeably. Prediction of protein-ligand binding affinity, however, still faces significant obstacles.
For the first time, the authors combine a graph neural network, a bitransport information mechanism, and physics-based distance terms in a novel graph neural network strategy called GraphscoreDTA for predicting protein-ligand binding affinity to overcome these limitations. In contrast to other approaches, GraphscoreDTA can both efficiently capture the mutual information between protein-ligand couples and highlight key ligand and protein residue atoms. On numerous test sets, the results demonstrate that GraphscoreDTA performs noticeably better than the state-of-the-art approaches. Additionally, investigations of drug-target selectivity on the homologous protein and cyclin-dependent kinase families show that GraphscoreDTA is a trustworthy tool for predicting protein-ligand binding affinity.
The source code is available at https://github.com/CSUBioGroup/GraphscoreDTA.
Reference:
Kaili Wang et. al.(2023)GraphscoreDTA: optimized graph neural network for protein–ligand binding affinity prediction 39(6):btad340