Mamba-Based Deep Learning Method for Solid-State Nanopore Signal Classification

Solid-state nanopore detection technology is a single-molecule signal detection method based on nanoscale pores. This technology has been widespread applied in biomedical research, environmental monitoring, and other fields. However, due to the significant variations in signal length of nanopore signal, the development of classification method for solid-state nanopore signals remains a challenging task. This study proposes a solid-state nanopore signal classification algorithm based on the Mamba architecture. As a cutting-edge model for long-sequence processing, Mamba could effectively capture long-range dependencies. Furthermore, for enhancing feature extraction capabilities, this study integrates a multi-head attention mechanism into the Mamba framework. Experiments were conducted on the Oxford Nanopore Technologies (ONT) dataset, and the results demonstrate that the proposed method outperforms traditional machine learning approaches and most existing deep learning models in terms of classification accuracy, recall, and F1 score. Specifically, the proposed method achieves a classification accuracy of 97.1%, which verifies its effectiveness in solid-state nanopore signal classification tasks.