CARE 2.0: reducing false-positive sequencing error corrections using machine learning

Area of Science:

• Genomics
• Bioinformatics

Background:

• Next-generation sequencing (NGS) requires preprocessing for error correction.
• Existing tools correct most errors but introduce false positives, impacting downstream analyses.
• There is a need for more precise sequencing error correction methods.

Purpose of the Study:

• To develop a more precise sequencing read error correction tool.
• To minimize false-positive corrections while maintaining high true-positive rates.

Main Methods:

• Developed CARE 2.0, a context-aware read error correction tool.
• Utilized multiple sequence alignment and a random decision forest classifier trained on Illumina data.
• Implemented in C++/CUDA for CPU and GPU execution.

Main Results:

• CARE 2.0 achieved up to two orders of magnitude fewer false positives than state-of-the-art tools.
• Maintained comparable true-positive correction rates.
• Demonstrated improved de novo assembly and k-mer analysis on simulated and real-world data.

Conclusions:

• CARE 2.0 significantly reduces false-positive sequencing errors, enhancing data quality.
• Machine learning approaches are effective for improving read error correction.
• The tool's precision benefits downstream genomic analyses and is publicly available.