ResBiGAAT: Residual Bi-GRU with attention for protein-ligand binding affinity prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Protein-ligand interactions are vital for drug discovery and repurposing.
• Current computational methods often require 3D protein structures, which are not always available.
• Existing models can be overly complex, leading to inefficient computations.

Purpose of the Study:

• To develop a novel deep learning model, ResBiGAAT, for predicting protein-ligand binding affinity.
• To leverage protein and ligand sequence-level features and physicochemical properties.
• To overcome the limitations of 3D structure dependency in existing prediction models.

Main Methods:

• ResBiGAAT combines a deep Residual Bidirectional Gated Recurrent Unit with two-sided self-attention mechanisms.
• The model utilizes protein and ligand sequence data and their physicochemical properties.
• Performance was rigorously evaluated using 5-fold cross-validation and an external dataset.

Main Results:

• ResBiGAAT demonstrates efficient prediction of protein-ligand binding affinity using sequence information.
• The model shows competitive performance and generalizability on an external dataset.
• A publicly available web interface (resbigaat.streamlit.app) has been developed for user accessibility.

Conclusions:

• ResBiGAAT offers an efficient and accurate alternative for predicting protein-ligand binding affinity, particularly when 3D structures are unavailable.
• The model's sequence-based approach simplifies the prediction process and reduces computational complexity.
• The developed web interface facilitates the practical application of ResBiGAAT in drug discovery research.