NanoBoost: a wavelet transform-enhanced machine learning algorithm for nanopore sensing

Nanopore sensing has evolved into a powerful technique for single-molecule analysis, with molecular carriers playing a pivotal role in enhancing signal specificity and enabling multiplexed diagnostics. Among these, nanoparticles offer unique advantages due to their tunable size and shape, which allow distinct electrical signatures during translocation and expand the scope of biomarker detection. In this study, we present a machine learning pipeline that combines the eXtreme Gradient Boosting (XGBoost) framework with a novel application of the Discrete Wavelet Transform (DWT) to classify nanosphere and nanorod nanopore detection with high accuracy. By extracting nanopore event-specific wavelet features, the model achieved 91% accuracy in size classification (5 nm vs 10 nm nanospheres) and 99% accuracy in shape classification (nanospheres vs nanorods). Feature importance analysis suggested that the model’s decisions were grounded in physical signal characteristics. Unsupervised learning revealed five distinct nanorod translocation profiles, linked to anisotropic behaviours such as tumbling and pore-wall interactions. Transmission Electron Microscopy (TEM) analysis validated structural heterogeneity, supporting the classification results. The pipeline was further applied to DNA translocation events, identifying key conformational states and demonstrating its versatility. This work establishes a framework for nanoparticle-based molecular carrier design for nanopore diagnostics and future applications, enabling real-time, high-accuracy biomarker screening at the single-molecule level.