MONSTROUS: a web-based chemical-transporter interaction profiler.

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Discovery

Background:

• Membrane transporters are crucial for cellular function, mediating the transport of endogenous and exogenous chemicals.
• Drug interactions with transporters significantly impact pharmacokinetics, potentially causing adverse drug-drug interactions, toxicity, or reduced therapeutic efficacy.
• Regulatory agencies recommend screening new molecular entities for transporter interactions due to their importance in drug efficacy and safety.

Purpose of the Study:

• To develop a publicly available, web-based tool for predicting chemical interactions with key drug transporters.
• To aid in the rapid screening and prioritization of drug candidates, identifying those without transporter liabilities.
• To provide a resource for predicting inhibitors and substrates for 12 transporters recommended for regulatory testing.

Main Methods:

• Developed machine learning and similarity-based classification models using publicly available data.
• Utilized graph convolutional neural networks (GCNNs) for transporters with sufficient bioactivity data.
• Implemented a 2D similarity-based approach for transporters lacking sufficient data, alongside applicability domain calculations.

Main Results:

• Achieved an average five-fold cross-validated ROC-AUC of 0.85 ± 0.07 for GCNN inhibitor models.
• Achieved an average ROC-AUC of 0.79 ± 0.08 for GCNN substrate models.
• Integrated predictive models and applicability domain calculations into an accessible web interface.

Conclusions:

• The MONSTROUS tool effectively predicts drug transporter interactions, assisting in early drug candidate assessment.
• Public availability of MONSTROUS facilitates rapid screening and prioritization, potentially reducing drug development risks.
• The predictive models demonstrate robust performance, supporting regulatory agency recommendations for transporter screening.