SquiggleNet

SquiggleNet classifies nanopore sequencing reads directly from raw electrical signals using a 1D Residual Network (ResNet) deep-learning model to enable real-time sequence classification for enrichment or depletion.

Key Features:

• Deep-learning model: Implements a 1D Residual Network (ResNet) deep-learning architecture to analyze signal-level data from Oxford Nanopore sequencing.
• Signal-level classification: Operates directly on raw electrical signals without performing base calling or sequence alignment.
• Real-time operation: Classifies reads at speeds that surpass DNA translocation through the nanopore, enabling immediate decision-making and potential read ejection (read-until).
• Minimal signal requirement: Achieves substantially higher classification accuracy using only one second of sequencing data.
• Memory efficiency: Reduces memory requirements by an order of magnitude compared to conventional alignment-based approaches.
• Human vs bacterial discrimination: Distinguishes human and bacterial DNA with over 90% accuracy and generalizes across bacterial species in human respiratory metagenomes.
• Element-level classification: Accurately classifies sequences containing human long interspersed repeat (LINE) elements.

Scientific Applications:

• Read-until enrichment/depletion: Enables real-time selection or ejection of reads for targeted enrichment or depletion workflows on Oxford Nanopore devices.
• Metagenomic analysis: Supports classification and taxonomic separation of complex samples such as human respiratory metagenomes.
• Host-pathogen discrimination: Facilitates rapid separation of host (human) and bacterial reads for downstream analysis.
• Genetic element detection: Allows identification of reads containing specific elements such as human long interspersed repeat (LINE) sequences.

Methodology:

Processes raw electrical signal input from Oxford Nanopore and applies a 1D Residual Network (ResNet) deep-learning model to perform real-time classification without basecalling or alignment.