Mining statistically-solid k-mers for accurate NGS error correction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Next-generation sequencing (NGS) data is prone to machine-induced errors.
• Effective error correction methods depend on accurately distinguishing 'solid' (frequent, likely correct) k-mers from 'weak' (infrequent, likely erroneous) k-mers.
• Determining an optimal frequency cutoff (f₀) and refining k-mer sets by removing erroneous solid k-mers and including error-free weak k-mers are critical challenges.

Purpose of the Study:

• To develop a robust statistical framework for identifying reliable k-mers in NGS data.
• To improve the accuracy of NGS data error correction by precisely classifying k-mers.
• To introduce a method for refining k-mer sets beyond simple frequency cutoffs.

Main Methods:

• Modeled erroneous k-mer frequencies using a Gamma distribution.
• Modeled correct k-mer frequencies using a mixture of Gaussian distributions to determine the frequency cutoff f₀.
• Employed z-scores to quantify k-mer frequency deviation from the mean, identifying subsets of solid and weak k-mers.
• Constructed a Bloom filter using statistically identified k-mers for error correction.

Main Results:

• The proposed method effectively determines the frequency cutoff f₀ by modeling k-mer distributions.
• Z-scores accurately distinguish between solid and weak k-mers, particularly identifying low-frequency solid k-mers.
• The developed error correction approach demonstrated superior performance compared to state-of-the-art methods on both real and synthetic NGS datasets.

Conclusions:

• The z-score is a powerful metric for differentiating k-mer types and refining k-mer sets.
• This statistical approach significantly enhances the accuracy of NGS error correction.
• The method offers a marked improvement in error correction accuracy for sequencing data.