DeepCorr: a novel error correction method for 3GS long reads based on deep learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Third-generation sequencing (3GS) technologies produce ultra-long reads crucial for biological analyses.
• High error rates in 3GS long reads hinder downstream applications.
• Existing error correction methods face limitations in accuracy and resource consumption.

Purpose of the Study:

• To propose DeepCorr, a novel deep learning-based algorithm for accurate long-read error correction.
• To address the challenge of high error rates in PacBio and Oxford Nanopore Technologies (ONT) data.
• To improve the utility of long reads in various genomic studies.

Main Methods:

• Developed DeepCorr, utilizing a recurrent neural network (RNN) for long-read error correction.
• Framed error correction as a multi-classification task, leveraging long-term dependencies within reads.
• Integrated high-precision short reads to generate feature vectors and labels for neural network training.

Main Results:

• DeepCorr effectively corrects errors in both PacBio and ONT long reads.
• The algorithm improves alignment identity while preserving the length advantage of 3GS data.
• Demonstrated superior performance compared to state-of-the-art methods on benchmark datasets with reduced computational resource usage.

Conclusions:

• DeepCorr offers a comprehensive and accurate deep learning solution for long-read error correction.
• The tool enhances the reliability of 3GS data for diverse biological analyses.
• DeepCorr provides an efficient and effective method for polishing long reads, even in regions with limited short-read coverage.