MegaTrans-machine learning models for drug transporters corresponding to the FDA guidance

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Drug Discovery

Background:

• Regulatory agencies require understanding drug-transporter interactions to prevent adverse effects.
• Computational models can predict transporter inhibition, avoiding costly late-stage development of problematic compounds.

Purpose of the Study:

• To develop and present MegaTrans, a web-based tool for predicting small molecule inhibition of key human drug transporters.
• To build machine learning models using curated in vitro data for transporter inhibition prediction.

Main Methods:

• Machine learning models were generated using extended-connectivity fingerprint 6 and 5-fold cross-validation on curated transporter datasets.
• Models were applied to FDA-approved drugs, with chemical space overlap analyzed using t-distributed stochastic neighbor embedding.

Main Results:

• Initial models achieved 66% accuracy, with a second round of models reaching 76% accuracy.
• MegaTrans provides predictions, model metrics, atom-bond contributions, and chemical space visualizations.

Conclusions:

• MegaTrans offers a computational approach to profile compounds and minimize in vitro screening for drug transporters.
• This tool can be integrated into early drug design and artificial intelligence approaches to enhance drug development efficiency.