DeepRepeat

DeepRepeat detects short-tandem repeats (STRs) directly from nanopore ionic current signals by converting raw electric signals into RGB-channel images for image-based identification and quantification of STRs, bypassing basecalling.

Key Features:

• Signal-level analysis: Operates on raw nanopore ionic current signals rather than on basecalled reads.
• Signal-to-image conversion: Converts ionic current signals into red-green-blue (RGB) channels to represent signal patterns as images.
• Image-based pattern recognition: Recasts repeat detection as an image analysis task to leverage image recognition methods for STR identification.
• STR identification and quantification: Identifies and quantifies short-tandem repeats from the transformed signal images.
• Long STR accuracy: Achieves superior accuracy for measuring long STRs compared to conventional basecalling-dependent methods.
• Empirical validation: Demonstrated accurate identification and quantification of telomeric repeats in the CHM13 cell line.
• Genome-wide and candidate-region evaluation: Validated on genome-wide datasets and on candidate regions derived from seven distinct sources.

Scientific Applications:

• STR detection in nanopore sequencing: Direct detection and quantification of STRs from raw nanopore signals without basecalling.
• Telomeric repeat analysis: Accurate measurement of telomeric repeats in the CHM13 cell line.
• Genome-wide STR characterization: Evaluation and characterization of STRs across genome-wide nanopore datasets.
• Candidate-region STR assessment: Assessment of STRs in candidate regions derived from seven distinct sources.

Methodology:

Converts nanopore ionic current signals into RGB-channel images and applies image-recognition–based pattern recognition to identify and quantify STRs from the transformed signals.