DeepREAL

DeepREAL predicts genome-wide ligand-induced activities and function selectivity of G-protein coupled receptors (GPCRs) using deep learning to support drug discovery.

Key Features:

• End-to-End Deep Learning Framework: Implements an end-to-end deep learning architecture to predict receptor activities at genome-wide scale.
• Multi-Scale Modeling: Employs multi-scale modeling to capture the complexity of drug–target interactions across different biological levels.
• Self-Supervised Learning: Utilizes self-supervised learning on extensive protein sequence datasets to mitigate scarcity of labeled receptor activity data.
• Pre-trained Binary Interaction Classification: Incorporates pre-trained binary interaction classification models to address distribution shifts when predicting interactions for novel chemicals.
• Out-of-Distribution Performance: Demonstrates state-of-the-art performance in out-of-distribution benchmark settings as reported in validation studies.
• Extensibility Beyond GPCRs: The methodological framework can be adapted to other gene families beyond GPCRs.

Scientific Applications:

• Function Selectivity Prediction: Predicts ligand function selectivity such as agonist versus antagonist effects on GPCRs.
• Genome-wide Receptor Activity Mapping: Enables genome-wide prediction of receptor activities induced by novel chemicals to link drug–target interactions to functional outcomes.
• Drug Discovery and Therapeutic Optimization: Supports drug discovery and optimization of therapeutic strategies by providing receptor activity predictions informed by genomic data.

Methodology:

Integrates self-supervised learning trained on tens of millions of protein sequences with pre-trained binary interaction classification models within an end-to-end deep learning framework that leverages multi-scale modeling to address data scarcity and distribution shifts; training and validation data sources include Pfam, GLASS, and IUPHAR/BPS.