ReQuant: improved base modification calling by k-mer value imputation.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Nanopore sequencing detects DNA base modifications directly from raw electrical signals.
• Current methods, including deep learning (DL), require extensive training data for all sequence contexts, which is often impractical.
• Research primarily focuses on 5-methylcytosine (5mC) in CpG contexts due to data limitations.

Purpose of the Study:

• To develop a novel algorithm, ReQuant, for imputing DNA modification models from limited training data.
• To enable accurate identification of various base modifications in Nanopore sequencing data, even in unrepresented sequence contexts.
• To overcome the limitations of existing methods requiring comprehensive training datasets.

Main Methods:

• Developed ReQuant, an algorithm that imputes full, k-mer based, modification models from sparse k-mer context training data.
• Applied ReQuant to Lambda Phage R9 and human R10 Nanopore sequencing data.
• Evaluated imputation accuracy for CpG/GpC methylation and CpG glucosylation.

Main Results:

• ReQuant demonstrated high accuracy in calling modifications using ≤25% of all possible 6-mers for training.
• The algorithm successfully generalized to different Nanopore sequencing data (R9 and R10).
• Showcased consistent and predictable effects of DNA modifications on Nanopore current signals.

Conclusions:

• ReQuant effectively imputes DNA modification models, circumventing the need for modification-specific DL tools.
• The approach enables base modification calling even when all sequence contexts are not available.
• This work significantly broadens the scope of base modification research using Nanopore technology.