Detecting a wide range of epitranscriptomic modifications using a nanopore-sequencing-based computational approach...

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Over 40 epigenetic and 300 epitranscriptomic modifications are known.
• Current short-read sequencing methods detect less than 10% of these modifications.
• Existing supervised machine learning approaches are limited to well-characterized modifications.

Purpose of the Study:

• To introduce Modena, an unsupervised learning approach for detecting a broad range of epigenetic and epitranscriptomic modifications.
• To leverage long-read nanopore sequencing for enhanced modification detection.
• To present dynamic thresholding as a novel computational strategy.

Main Methods:

• Utilized long-read nanopore sequencing data.
• Developed Modena, an unsupervised machine learning algorithm.
• Implemented a dynamic thresholding method based on 1D score-clustering.

Main Results:

• Modena outperformed existing methods on five out of six benchmark datasets.
• Modena demonstrated consistent accuracy on DNA modification detection.
• Dynamic thresholding significantly improved the performance of existing algorithms, tripling F1-scores in some cases.

Conclusions:

• Modena offers a powerful new approach for broad-spectrum epigenetic and epitranscriptomic modification detection.
• Dynamic thresholding represents a broadly applicable and effective computational strategy for modification analysis.
• Long-read sequencing integrated with advanced machine learning holds significant promise for advancing epigenetics and epitranscriptomics research.