BiasNet: A Model to Predict Ligand Bias Toward GPCR Signaling

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Discovery

Background:

• Signaling bias in G protein-coupled receptor (GPCR) ligands has significant clinical implications.
• Therapeutic advantage of G protein or β-arrestin bias depends on the specific signaling context.
• Understanding the structural basis of bias is crucial for rational drug design.

Purpose of the Study:

• To identify molecular scaffolds and pharmacophores associated with biased signaling in GPCR ligands.
• To gain insights into the structural determinants of G protein versus β-arrestin bias.
• To develop predictive models for GPCR ligand bias.

Main Methods:

• Analysis of GPCR biased ligands from the BiasDB database.
• Training five machine learning models using 15 feature sets on ligands exhibiting G protein or β-arrestin bias.
• Identification of key molecular fragments and scaffolds using random forest models and t-SNE clustering.
• Development of a web-based tool for bias prediction.

Main Results:

• Specific molecular fragments, including secondary and aromatic amines, were found to be more prevalent in β-arrestin biased ligands.
• Five distinct scaffolds were identified that exhibit either G protein or β-arrestin bias.
• Machine learning models demonstrated predictive power, correlating with unsupervised clustering methods.
• A web implementation (BiasNet) was created for predicting ligand bias.

Conclusions:

• Structural features significantly influence the G protein or β-arrestin bias of GPCR ligands.
• Machine learning approaches effectively identify bias-driving structural motifs and scaffolds.
• The developed BiasNet tool enhances the applicability of these findings for drug discovery efforts.