Annotation of Allosteric Compounds to Enhance Bioactivity Modeling for Class A GPCRs.

Area of Science:

• Pharmacology and Cheminformatics
• Computational Drug Discovery

Background:

• Proteins possess orthosteric and allosteric binding sites, crucial for drug discovery.
• Allosteric modulators are increasingly important but distinguishing binding data is challenging.
• Existing chemical databases often mix allosteric and orthosteric binding information, hindering analysis.

Purpose of the Study:

• To improve the retrieval of allosteric and orthosteric binding data from scientific literature.
• To develop predictive models for classifying compound binding types.
• To assess the impact of accurate binding type information on drug-target interaction predictions.

Main Methods:

• Utilized an enhanced text mining approach to extract binding data from ChEMBL release 22.
• Developed convolutional deep neural networks for predicting binding types in class A G protein-coupled receptors (GPCRs).
• Employed temporal split validation to evaluate model performance.

Main Results:

• Achieved a Matthews correlation coefficient (MCC) of 0.54 for predicting binding types.
• Demonstrated high sensitivity for orthosteric (0.94) and allosteric (0.54) predictions.
• Incorporating accurate binding types as descriptors improved binding prediction performance (MCC from 0.27 to 0.34).

Conclusions:

• Accurate classification of allosteric and orthosteric binding data is feasible using advanced text mining and machine learning.
• The inclusion of binding type information significantly enhances the accuracy of predictive models in drug discovery.
• This work provides a valuable dataset for researchers focusing on allosteric modulators and GPCRs.