MSRCall

MSRCall performs basecalling of Oxford Nanopore MinION ionic current signals to infer DNA and RNA nucleotide sequences using a multi-scale recurrent deep neural network.

Key Features:

• Multi-scale deep neural network architecture: Captures short- and long-range dependencies in MinION ionic current signals to represent features across multiple temporal scales.
• Multi-scale recurrent layers: Recurrent layers are redesigned to operate within a multi-scale framework for improved temporal feature extraction.
• Fusion block: Integrates information from different scales to enhance interpretation of complex signal patterns associated with nucleotide sequences.
• Connectionist Temporal Classification (CTC) decoder: Aligns input ionic current signals with output nucleotide labels without requiring pre-segmented data, supporting real-time basecalling.

Scientific Applications:

• Improved basecalling accuracy: Demonstrates higher read and consensus accuracies compared to existing basecallers for Oxford Nanopore data.
• Real-time sequencing: Enables immediate sequence analysis from MinION ionic current signals for applications such as field-based genomic studies and rapid pathogen detection.
• Research and clinical sequencing applications: Supports genomic investigations and clinical diagnostics that require accurate interpretation of nanopore sequencing data.

Methodology:

Applies a multi-scale deep neural network with redesigned multi-scale recurrent layers, a fusion block for scale integration, and a Connectionist Temporal Classification (CTC) decoder to Oxford Nanopore MinION ionic current signals.