DeePaC-Live

DeePaC-Live detects novel pathogens in real time from Illumina and Nanopore sequencing reads using deep neural networks to predict pathogenic potential during sequencing.

Key Features:

• Real-Time Detection: Performs analysis of sequencing data as it is generated to enable early detection during sequencing runs.
• Deep Learning Integration: Employs deep neural networks trained to classify reads from Illumina and Nanopore platforms and to predict pathogenic potential without requiring complete datasets or reliance on known-pathogen databases.
• Performance Enhancement: Outperforms existing machine learning and sequence-alignment techniques in simulated and real-world scenarios, with a reported ~80-fold increase in sensitivity after 50 Illumina cycles compared to traditional real-time mapping methods.
• Efficient Subsequence Utilization: Uses short subsequences for inference; the first 250 base pairs of Nanopore reads (≈0.5 seconds of sequencing) yield predictions that can surpass complete long-read mappings.
• Integration with HiLive2: Integrates with HiLive2, a real-time Illumina mapper, to process and analyze Illumina data during sequencing runs.

Scientific Applications:

• Outbreak surveillance and early detection: Enables early intervention by detecting novel or rapidly evolving pathogens in real time during sequencing runs.
• Biosecurity screening of synthetic sequences: Can be applied to screen synthetic DNA/RNA sequences for potential biosecurity threats.
• Rapid partial-read analysis: Supports rapid inference from partial reads to accelerate decision-making in Illumina and Nanopore sequencing workflows.
• SARS-CoV-2 sequencing monitoring: Has been evaluated in simulated and real-world SARS-CoV-2 sequencing runs to assess real-time detection performance.

Methodology:

DeePaC-Live uses deep neural networks trained on Illumina and Nanopore sequencing reads to classify and predict pathogenic potential in real time, integrates with HiLive2 for real-time Illumina mapping, utilizes short subsequences (e.g., first 250 bp of Nanopore reads) for early inference, and was benchmarked against traditional machine-learning and sequence-alignment methods with sensitivity measured after 50 Illumina cycles.