Machine learning-driven bioavailability prediction in early-stage drug development: a KNIME-based computational...

Area of Science:

• Computational chemistry
• Pharmacokinetics
• Machine learning in drug discovery

Background:

• Bioavailability prediction is crucial but challenging in early drug development.
• Traditional experimental methods are time-consuming and costly.
• Machine learning offers a potential solution to automate and improve prediction efficiency.

Purpose of the Study:

• To explore machine learning for enhanced bioavailability prediction.
• To automate bioavailability assessment using computational workflows.
• To reduce reliance on in vitro and in vivo studies.

Main Methods:

• Analyzed 475 drug-like compounds using molecular descriptors.
• Applied Random Forest, Gradient Boosting, Decision Trees, k-NN, and neural networks.
• Utilized 5-fold cross-validation for model performance assessment.

Main Results:

• Ensemble models outperformed linear and neural network approaches.
• Random Forest achieved the highest predictive performance (R² = 0.87, RMSE = 0.08).
• Topological polar surface area and solubility were key predictive factors.

Conclusions:

• Machine learning, integrated with open-source tools like KNIME, improves pharmaceutical research efficiency.
• This approach supports FAIR data principles and cost-effective bioavailability assessment.
• Facilitates AI-driven predictive modeling for digital health applications in drug development.