Identifying GPCR-drug interaction based on wordbook learning from sequences.

Area of Science:

• Pharmacology and Cheminformatics
• Computational Biology and Bioinformatics

Background:

• G protein-coupled receptors (GPCRs) are crucial drug targets involved in numerous physiological processes and diseases.
• Experimental screening of GPCR-drug interactions is costly and time-consuming, necessitating computational approaches.
• Accurate prediction of GPCR-drug interactions is vital for developing targeted therapies.

Purpose of the Study:

• To develop a novel, sequence-based computational method for predicting GPCR-drug interactions.
• To overcome the limitations of experimental methods in large-scale GPCR-drug interaction identification.
• To establish a foundation for efficient drug design targeting GPCRs.

Main Methods:

• Utilized a novel bag-of-words (BoW) model for extracting sequence features from GPCRs.
• Employed discrete Fourier transform (DFT) for higher-order pattern extraction from drug molecular fingerprints.
• Integrated GPCR and drug features into a distance-weighted K-nearest-neighbor (DWKNN) prediction engine, with potential for ensemble learning.

Main Results:

• The proposed sequence-based method demonstrates superior generalization ability compared to existing methods on GPCR-drug interaction datasets.
• An enhanced post-processing procedure (PPP) further improved prediction performance.
• The novel feature extraction techniques proved effective for predicting GPCR-drug interactions.

Conclusions:

• The developed methods are effective for predicting GPCR-drug interactions and show potential for other target-drug or protein-protein interaction predictions.
• The modified BoW model for GPCR sequence feature extraction may benefit protein classification and attribute prediction tasks.
• The source code is available for academic research, promoting further development in the field.