DeepSelectNet

DeepSelectNet performs high-accuracy classification of Oxford Nanopore current signals to enable selective sequencing and species identification from mixed samples.

Key Features:

• Deep Learning Architecture: DeepSelectNet employs an advanced neural network architecture with novel data preprocessing techniques and regularization strategies for classifying nanopore current signals.
• Accuracy and Performance: In comparative studies on five datasets it achieved accuracies of 91%–99% (average 95%), outperforming existing methods that ranged 77%–97% (average <89%) and showing approximately a 12% increase over SquiggleNet.
• Precision and Recall: Precision and recall metrics were reported to average 95% and generally exceed 89%.
• Execution Performance: Execution speed improved by 13% on average compared to SquiggleNet.
• Selective Read Rejection: Enables programmatic rejection of irrelevant reads during Oxford Nanopore selective sequencing to focus on target genomic sequences.

Scientific Applications:

• Species classification from mixed samples: Classifies species-specific signals in mixed-sample datasets generated by Oxford Nanopore sequencing for downstream analysis.
• Selective sequencing in genomics research: Supports targeted sequencing workflows by identifying and retaining reads of interest while rejecting non-target reads in real time.
• Real-time species identification: Facilitates rapid species identification applicable to ecological studies and biodiversity assessments using Oxford Nanopore selective sequencing.

Methodology:

Uses an advanced neural network architecture with novel data preprocessing techniques and regularization strategies to classify Oxford Nanopore current signals; evaluated on five datasets with comparative accuracy and execution-speed measurements against SquiggleNet.