Identification of Novel High-Affinity Substrates of OCT1 Using Machine Learning-Guided Virtual Screening and Experimental Validation

OCT1 is the most highly expressed cation transporter in the liver and affects pharmacokinetics and pharmacodynamics. Newly marketed drugs have previously been screened as potential OCT1 substrates and verified by virtual docking. Here, we used machine learning with transport experiment data to predict OCT1 substrates based on classic molecular descriptors, pharmacophore features, and extended-connectivity fingerprints and confirmed them by in vitro uptake experiments. We virtually screened a database of more than 1000 substances. Nineteen predicted substances were chosen for in vitro testing. Sixteen of the 19 newly tested substances (85%) were confirmed as, mostly strong, substrates, including edrophonium, fenpiverinium, ritodrine, and ractopamine. Even without a crystal structure of OCT1, machine learning algorithms predict substrates accurately and may contribute not only to a more focused screening in drug development but also to a better molecular understanding of OCT1 in general.

Automated summary

This study used machine learning and in vitro experiments to predict and confirm OCT1 substrates, with 85% of the 16 newly tested substances confirmed as strong substrates for the protein. The findings show the potential of machine learning algorithms in accurate substrate prediction and their contribution to drug development screening and understanding of OCT1.