A hybrid supervised and unsupervised machine learning approach for identifying nucleoside drugs using nanopore readouts

Nucleoside drugs, mimics of natural nucleosides, have become cornerstone treatments in clinical approaches to combat cancer and viral infections. The analysis of nucleoside drugs is commonly performed using liquid chromatography-tandem mass spectrometry (LC-MS/MS), which requires bulky, expensive instruments and is time-consuming. Notably, while detection methods for natural nucleosides have advanced to their 'next generation' with commercialization, the analysis of nucleoside drugs continues to depend on complex analytical tools. Here, we propose a highly portable, next-generation nanopore-based approach for the rapid and precise high-throughput identification of nucleoside drugs. A large number of both anti-viral and anti-cancer drugs can be simultaneously identified and distinguished from nanopore transmission readouts. To facilitate rapid detection, a hybrid supervised and unsupervised machine learning (ML) workflow is demonstrated, enabling the single-molecule detection of each drug over a dynamic configuration space with good accuracy. By implementing the robust ML-assisted nanopore approach, we can overcome previous obstacles, enabling rapid and precise detection of nucleoside drugs.