Machine Learning-Driven Nanopore Sensing for Quantitative, Label-Free miRNA Detection

Area of Science:

• Biotechnology
• Nanotechnology
• Molecular Diagnostics

Background:

• Nanopore sensors provide high sensitivity for single-molecule detection, crucial for early disease diagnosis.
• MicroRNAs (miRNAs) are vital biomarkers for various diseases, necessitating accurate detection methods.

Purpose of the Study:

• To develop and evaluate a multiplexed nanopore-based assay for specific and accurate miRNA detection.
• To compare the performance of different computational strategies for analyzing nanopore signals.

Main Methods:

• Utilized DNA-barcoded probes that induce characteristic signal delays in nanopore translocation upon target miRNA binding.
• Evaluated three signal classification methods: moving standard deviation (MSD), spectral entropy (SE), and a convolutional neural network (CNN).
• Trained the CNN on image representations of raw nanopore current traces for enhanced analysis.

Main Results:

• The CNN model achieved near-perfect classification performance (accuracy, precision, recall = 0.99), outperforming MSD and SE.
• Grad-CAM visualization confirmed the CNN's focus on relevant signal features, improving interpretability.
• Nanopore-derived delay metrics correlated well with RT-qPCR validation data, demonstrating assay validity.

Conclusions:

• A CNN-based approach offers superior sensitivity and robustness for analyzing nanopore sensor data in miRNA detection.
• This work establishes a framework for machine learning-driven nanopore diagnostics for single-molecule biomarker detection.
• Advanced data interpretation is key to unlocking the full potential of nanopore sensing for diagnostics.