GNNSeq: A Sequence-Based Graph Neural Network for Predicting Protein-Ligand Binding Affinity.

Area of Science:

• Computational chemistry and cheminformatics
• Machine learning in drug discovery
• Bioinformatics and structural biology

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for effective drug discovery.
• Existing sequence-based models often lack accuracy and robustness, limiting their generalizability.
• Need for models that do not require pre-docked complexes or structural data.

Purpose of the Study:

• To develop GNNSeq, a novel hybrid machine learning model for predicting protein-ligand binding affinity.
• To overcome limitations of existing models by leveraging sequence features exclusively.
• To enhance accuracy, robustness, and generalizability in binding affinity predictions.

Main Methods:

• GNNSeq integrates a Graph Neural Network (GNN) with Random Forest (RF) and XGBoost.
• The model extracts molecular characteristics and sequence patterns from protein and ligand sequences.
• A unique kernel-based context-switching design optimizes efficiency and dynamically adjusts feature weighting.

Main Results:

• GNNSeq achieved a Pearson correlation coefficient (PCC) of 0.784 on the PDBbind v.2020 refined set and 0.84 on the PDBbind v.2016 core set.
• External validation on DUDE-Z dataset showed an average AUC of 0.74.
• Hybrid models incorporating GNNSeq reached a PCC of 0.97, with efficient training times (5000+ complexes in ~1.5 hours).

Conclusions:

• GNNSeq offers an efficient and scalable solution for binding affinity prediction.
• The model demonstrates improved accuracy and generalizability, facilitating large-scale virtual screening.
• GNNSeq is publicly available via a server-based GUI for cost-effective hit identification.