NanoSSL: attention mechanism-based self-supervised learning method for protein identification using nanopores.

Area of Science:

• Biophysics
• Computational Biology
• Proteomics

Background:

• Nanopore technology enables single-molecule analysis of biomolecules, with growing applications in proteomics.
• Challenges in nanopore analysis include noisy data and data scarcity, hindering machine learning development.
• Self-supervised learning offers a promising approach to address these limitations in nanopore data processing.

Purpose of the Study:

• To introduce and validate Nanopore analysis using Self-Supervised Learning (NanoSSL) for protein identification.
• To leverage self-supervised learning and attention mechanisms to improve the analysis of nanopore-based protein signals.
• To address the challenges of noisy data and data scarcity in nanopore proteomics.

Main Methods:

• Developed NanoSSL, a generative self-supervised learning framework utilizing attention mechanisms.
• Employed a two-step approach: self-supervised pre-training followed by supervised fine-tuning.
• Implemented a masked autoencoder for reconstructing fragmented nanopore translocation events during pre-training.

Main Results:

• NanoSSL achieved unprecedented accuracy, precision, recall, and F1 scores in classifying mutated amyloid-beta 1-42 proteins.
• Demonstrated superior performance on a public nanopore dataset and homemade solid-state nanopore measurements.
• Verified that self-supervised learning and attention mechanisms are key contributors to performance enhancement.

Conclusions:

• NanoSSL effectively enhances protein identification from nanopore data, outperforming previous methods.
• The framework successfully addresses data scarcity and noise issues inherent in single-molecule nanopore analysis.
• This approach holds significant potential for advancing nanopore-based proteomics and biomarker detection.