Benchmarking variant identification tools for plant diversity discovery.

Area of Science:

• Plant Genomics
• Bioinformatics
• Next-Generation Sequencing (NGS)

Background:

• Accurate genomic variation identification is vital for plant studies using high-throughput sequencing.
• Existing bioinformatics tools, often developed for human studies, may not be optimal for plant research.
• A comprehensive evaluation of variant calling pipelines in plants is needed.

Purpose of the Study:

• To evaluate the performance of different bioinformatics tools in plant variant calling pipelines.
• To compare alignment, variant calling, filtering, and imputation methods using simulated and real plant genomic data.

Main Methods:

• Comparative analysis of alignment tools (SOAP2, Bowtie2, BWA-MEM).
• Evaluation of variant callers (GATK HaplotypCaller, SAMtools mpileup).
• Assessment of single vs. multiple reference genomes.
• Comparison of machine-learning-based vs. hard-cutoff variant filtering.
• Analysis of a 2-step imputation method using high-confidence SNPs.

Main Results:

• BWA-MEM showed high accuracy in read alignment; Bowtie2 had the highest overall accuracy.
• Variant caller performance varied based on genetic diversity, sequence coverage, and genome complexity.
• A single reference genome was insufficient for distantly related species.
• Machine learning significantly improved variant filtering accuracy (more true positives, fewer false positives).
• The 2-step imputation method achieved up to 60% higher accuracy.

Conclusions:

• Bioinformatics tool performance differs significantly for plant genomic datasets.
• Tool selection should align with study objectives and available resources.
• This evaluation provides guidance for plant biologists using NGS for diversity and crop improvement.