Overview

The Blood-Brain Barrier (BBB) is a highly specialized vascular structure that plays a crucial role in maintaining Central Nervous System (CNS) homeostasis. This selective barrier shields the brain from neurotoxins and foreign substances, including many drugs. However, its restrictive nature poses challenges in the development of drugs targeting CNS-related diseases.

Enter SMILES Strings

 Example of input file (1, 2, 3)
Predict Clear

What is graphB³?

graphB³ is a Graph Convolutional Network (GCN)-based model designed to predict the Blood-Brain Barrier (BBB) permeability of drug candidates using atomic-level molecular features. By leveraging deep learning and graph-based representations, it provides highly accurate predictions while offering insights into the molecular substructures that influence BBB permeability.

Importance of BBB Permeability

BBB permeability is a critical factor in determining the potential efficacy of drug candidates for CNS disorders. Drugs must cross the BBB to exert therapeutic effects on the brain. Assessing BBB permeability early in the drug discovery pipeline can save time, labor, and costs, making it a pivotal step in CNS drug development.

The Role of Deep Learning in Drug Discovery

The Role of Deep Learning in Drug Discovery Deep learning has transformed drug discovery by enabling data-driven predictions that significantly reduce the cost, time, and effort required for identifying potential therapeutics. Traditional approaches rely on molecular-level physiochemical properties and rule-based models, which, while effective, often fail to capture the complex structural and atomic interactions governing Blood-Brain Barrier (BBB) permeability. Recent advancements in Graph Convolutional Networks (GCNs) have provided a more robust framework by leveraging the inherent graph-like nature of molecular structures. Unlike conventional methods that depend on pre-defined descriptors, GCNs extract atomic-level features and structural relationships directly from molecular graphs, improving predictive accuracy. graphB³ builds upon this approach, incorporating an Explainable AI (XAI) framework to not only predict BBB permeability but also identify critical substructures influencing drug transport across the barrier. By integrating deep learning with domain-specific insights, graphB³ enhances the rational design of CNS-targeting drugs, paving the way for more efficient and interpretable drug discovery pipelines.

graphB³: A GCN-Based BBB Prediction Model

To overcome these limitations, we present graphB³, a cutting-edge Graph Convolutional Network (GCN)-based model. This model leverages atomic-level features to predict the BBB permeability of drug candidates with remarkable accuracy and robustness.

Key Features and Achievements of graphB³

• High Performance: Average Accuracy: 0.88 , AUC-ROC: 0.94 , F1 Score: 0.91 (on an independent test set).
• Explainable AI (XAI): graphB³ incorporates an XAI component to identify the most significant atomic-level features and molecular substructures contributing to BBB permeation.
• The development of potential drugs for CNS diseases with deeper insights into the molecular determinants for permeability.

GitHub Repository: https://github.com/dhanjal-lab/graphB³

Key Features of the GitHub Version:

• Process more than 10 SMILES at a time.
• Step-by-step tutorial for setup and execution.
• Customizable for different datasets and research needs.