Noise2read: Accurately Rectify Millions of Erroneous Short Reads Through Graph Learning on Edit Distances

Area of Science:

• Genomics and Bioinformatics
• Computational Biology

Background:

• Short-read sequencing data, despite low per-base error rates (0.1%-0.5%), can contain millions of erroneous reads (10%-15%).
• Existing error correction methods often introduce new errors or fail to fully restore original sequences, compromising data integrity.

Purpose of the Study:

• To develop a novel method, noise2read, for accurate rectification of erroneous reads in sequencing data.
• To preserve data integrity by reverting erroneous reads to their original states without generating non-existent sequences.

Main Methods:

• Leveraged a computable rule derived from polymerase chain reaction (PCR) error mechanisms: a rare read is likely erroneous if it has a highly abundant neighboring read.
• Constructed a graph linking reads with small edit distances to identify erroneous reads and used these pairs as training data to refine error identification.

Main Results:

• Noise2read significantly outperforms state-of-the-art methods across 19 metrics on UMI-based datasets.
• Demonstrated substantial improvements in genome abundance quantification, isoform identification, SNP profiling, and genome editing efficiency estimation.
• Case studies confirm noise2read's capability to enhance short-read sequencing quality.

Conclusions:

• Noise2read effectively rectifies erroneous reads in PCR-amplified sequencing data.
• The method preserves data integrity and offers significant advantages for various genomic applications.
• Noise2read is publicly available, facilitating its adoption in the research community.