Evaluating assembly and variant calling software for strain-resolved analysis of large DNA viruses

Area of Science:

• Virology
• Bioinformatics
• Genomics

Background:

• Human cytomegalovirus (HCMV) poses risks to immunocompromised individuals and congenitally infected infants.
• Analyzing heterogeneous HCMV populations requires accurate strain assembly and variant calling from high-throughput sequencing data.
• Existing methods lack thorough evaluation for complex viral populations like HCMV, especially with low-divergent strains and large genomes.

Purpose of the Study:

• To benchmark the performance of various assemblers and variant callers for HCMV quasispecies analysis.
• To identify optimal bioinformatics tools and best practices for characterizing viral populations.
• To assess challenges in analyzing low-divergent viral strains with large genomes.

Main Methods:

• Evaluated 15 assemblers and 6 variant callers using 10 lab-generated datasets.
• Utilized two different library preparation protocols to create benchmark datasets.
• Developed and provided the QuasiModo benchmarking workflow for reproducibility.

Main Results:

• MetaSPAdes and IVA assemblers demonstrated strong performance in recovering abundant HCMV strains.
• Savage assembler showed limitations in recovering low-abundance strains, resulting in fragmented sequences.
• LoFreq and VarScan2 variant callers performed well across all strain abundances but exhibited systematic T-to-G errors in a specific genomic context.

Conclusions:

• Specific assemblers (metaSPAdes, IVA) and variant callers (LoFreq, VarScan2) are recommended for HCMV quasispecies analysis.
• Awareness of context-dependent systematic errors in variant calling is crucial for accurate HCMV data interpretation.
• The QuasiModo workflow facilitates reproducible benchmarking for viral quasispecies analysis.