Ligand-Based Drug Discovery Leveraging State-of-the-Art Machine Learning Methodologies Exemplified by Cdr1 Inhibitor...

Area of Science:

• Computational chemistry and cheminformatics
• Machine learning applications in pharmacology
• Drug discovery and development

Background:

• Artificial intelligence (AI) is transforming drug discovery, offering speed and efficiency.
• Ligand-based strategies are crucial when target 3D structures are unavailable.
• Predictive modeling is essential for identifying potential drug candidates.

Purpose of the Study:

• To develop and evaluate novel AI models for ligand-based drug discovery.
• To establish a generalizable workflow for building predictive AI models.
• To demonstrate the efficacy of ensemble machine learning for inhibitor prediction.

Main Methods:

• Utilized target-specific experimental data and diverse molecular features.
• Employed multiple state-of-the-art machine learning algorithms, including a 3D graph neural network.
• Implemented Bayesian hyperparameter tuning, stacked generalization, and soft voting for ensemble modeling.

Main Results:

• The ensemble model achieved high performance metrics, including an average precision of 0.755 and an F1-score of 0.714 on an external test set.
• Demonstrated a low false positive rate (0.1236) on data outside the training chemical space.
• Validated the model's ability to generalize and avoid false positives.

Conclusions:

• Stacking ensemble machine learning offers significant potential for drug discovery.
• The presented workflow provides a rigorous framework for developing ligand-based predictive AI models.
• This approach can be adapted for predicting inhibitors against various other drug targets.