DeepMP: a deep learning tool to detect DNA base modifications on Nanopore sequencing data.

Area of Science:

• Genomics and Epigenetics
• Bioinformatics and Computational Biology

Background:

• DNA methylation is crucial for biological processes, and Nanopore long-read sequencing offers direct detection of these modifications.
• Existing computational methods for Nanopore-based methylation detection often use fixed thresholds and struggle with low-frequency methylated sites.
• Current methods typically utilize either Nanopore signals or basecalling errors, but not a combination of both.

Purpose of the Study:

• To develop a novel computational model, DeepMP, for accurate DNA methylation detection using Nanopore sequencing data.
• To introduce a threshold-free model capable of detecting methylation at low frequencies across cells.
• To leverage both Nanopore signals and basecalling errors for improved methylation detection.

Main Methods:

• Developed DeepMP, a convolutional neural network (CNN)-based model.
• Integrated Nanopore signal data and basecalling errors as input features.
• Implemented a threshold-free model for position modification calling.

Main Results:

• DeepMP accurately detects methylated motifs within Nanopore reads.
• The model demonstrates high sensitivity for identifying sites methylated at low frequencies.
• Benchmarking on *Escherichia coli*, human, and pUC19 datasets showed DeepMP outperforms existing state-of-the-art methods in both read-based and position-based methylation detection.

Conclusions:

• DeepMP offers a significant advancement in Nanopore-based DNA methylation detection.
• The threshold-free approach and combined feature utilization enhance accuracy and sensitivity, particularly for low-frequency methylation.
• DeepMP is freely available, facilitating broader adoption in genomic research.