Iterative Self-Supervised Signal Curation Enables High-Fidelity Peptide Profiling in Parallel Nanopore Sensing

Area of Science:

• Proteomics
• Biotechnology
• Data Science

Background:

• Nanopore sequencing is increasingly used for high-throughput proteomics, generating large datasets.
• Stochastic translocations and sensor variations introduce significant noise and artifacts in nanopore data.
• Automated and unbiased data curation is crucial for accurate proteomics analysis.

Purpose of the Study:

• To develop an automated and unbiased method for curating nanopore proteomics data.
• To introduce NanoCurator, a self-supervised deep learning model for signal extraction.
• To enhance the reliability and efficiency of nanopore-based peptide identification.

Main Methods:

• Developed NanoCurator, an iterative, self-supervised convolutional neural network-long short-term memory (CNN-LSTM) autoencoder.
• Utilized reconstruction error and Lempel-Ziv complexity with adaptive thresholding for signal segregation.
• Validated NanoCurator on simulated and empirical datasets with diverse noise profiles.

Main Results:

• NanoCurator effectively distinguishes genuine translocation signals from various noise types.
• Achieved significant improvements in classification accuracy: 1.8% for peptide panels, 2.3% for PTMs, and 1.5% for mixtures.
• Reduced data volume requirements while maintaining high-fidelity signal extraction.
• Demonstrated robust generalization across different datasets.

Conclusions:

• NanoCurator provides a robust and generalizable framework for automated signal quality control in nanopore proteomics.
• The method significantly enhances the accuracy of peptide identification and PTM analysis.
• Enables reliable, large-scale application of massively parallel nanopore sensing in proteomics research.