Merfin: improved variant filtering, assembly evaluation and polishing via k-mer validation

Area of Science:

• Genomics
• Bioinformatics

Background:

• Variant calling is crucial for genotyping and long-read genome assembly.
• Current hard-filtering methods for variant calls lack universal optimal cutoffs due to read quality and caller dependency.
• Existing methods struggle with accuracy in complex genomic regions and diverse sequencing technologies.

Purpose of the Study:

• Introduce Merfin, a k-mer based algorithm for accurate variant filtering.
• Improve precision in genotyping and consensus accuracy in genome assembly polishing.
• Provide robust variant evaluation independent of read alignment and caller scores.

Main Methods:

• Developed Merfin, a k-mer multiplicity-based variant-filtering algorithm.
• Evaluated Merfin on human and nonhuman genome assemblies from Pacific Biosciences HiFi, continuous long reads, and Oxford Nanopore reads.
• Introduced new assembly quality and completeness metrics considering expected genomic copy numbers.

Main Results:

• Merfin significantly increased the precision of genotyped calls across multiple benchmarks.
• Demonstrated improved consensus accuracy and reduced frameshift errors in long-read assemblies.
• Successfully applied to the first complete human genome assembly, showcasing its broad applicability.

Conclusions:

• Merfin offers a robust and accurate approach to variant filtering for genotyping and genome assembly.
• The algorithm's k-mer based evaluation provides a reliable alternative to alignment-dependent scores.
• Merfin advances the accuracy of long-read genome assembly and analysis, including complete genomes.