TrGPCR: GPCR-Ligand Binding Affinity Prediction Based on Dynamic Deep Transfer Learning.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Predicting G protein-coupled receptor (GPCR)-ligand binding affinity is vital for drug development but faces challenges.
• Current data-driven methods often require 3D protein structures, which are frequently unavailable.
• Existing approaches may overlook crucial secondary structure information and suffer from insufficient data for deep learning.

Purpose of the Study:

• To develop a deep transfer learning model, TrGPCR, to accurately predict GPCR-ligand binding affinities.
• To address the challenge of limited GPCR data for training machine learning models.
• To incorporate protein secondary structures as predictive features.

Main Methods:

• Implemented a deep transfer learning approach using dynamic transfer learning.
• Utilized the Binding Database (BindingDB) as the source domain and the GLASS database as the target domain.
• Introduced protein secondary structures (pockets) as features for affinity prediction.

Main Results:

• The TrGPCR model demonstrated improved prediction accuracy compared to the DeepDTA model.
• Achieved a 5.2% improvement in root mean square error (RMSE).
• Achieved a 4.5% improvement in mean squared error (MAE).

Conclusions:

• TrGPCR effectively addresses the issue of insufficient GPCR data through transfer learning.
• The inclusion of protein secondary structures enhances binding affinity prediction.
• This method offers a more efficient and accurate approach for GPCR-ligand binding affinity prediction in drug discovery.