DeepMP

DeepMP applies a convolutional neural network to Nanopore long-read sequencing raw signal and basecalling error data to detect DNA methylation of specific motifs and enable sensitive epigenetic analysis.

Key Features:

• Convolutional Neural Network (CNN) architecture: DeepMP employs a CNN that integrates Nanopore raw signal data and basecalling error information to classify methylated versus unmethylated DNA motifs.
• Threshold-free position modification calling: The method implements a threshold-free model for position-level methylation calling, enabling detection of low-frequency methylation sites across cells.
• Comprehensive benchmarking: DeepMP was evaluated on E. coli, human, and pUC19 datasets and demonstrates superior performance in both read-based and position-based methylation detection across varying methylation frequencies.

Scientific Applications:

• Heterogeneous cell populations: Detects low-frequency methylation events in mixed cell populations.
• Cancer research: Enables profiling of methylation heterogeneity relevant to tumor biology.
• Developmental biology: Supports mapping methylation changes during development.
• Single-cell epigenomics: Facilitates detection of methylation at low frequencies suitable for single-cell analyses.
• Epigenetic regulation studies: Provides methylation maps for investigating gene regulation and cellular differentiation.

Methodology:

DeepMP trains a convolutional neural network on Nanopore sequencing raw signal and basecalling error features to learn patterns distinguishing methylated from unmethylated motifs and performs position-level calling without predefined thresholds.