How to refine variant quality with GATK VQSR?

Prepare Your Data

• Ensure you have your reference genome, known sites of variation, and target variants ready.

 

• Use the `HaplotypeCaller` to generate GVCF files for each sample. Merge these if you have multiple samples using `CombineGVCFs` and then run `GenotypeGVCFs` for joint genotyping.

Run BaseRecalibrator

• Before variant recalibration, use the `BaseRecalibrator` tool to correct any systematic biases in base quality scores.

 

• Provide it with known sites of variation to avoid skewing recalibration estimates.

Build the Recalibration Model

• Use the `VariantRecalibrator` tool on your raw VCF file to build a model based on a set of highly-confident variants.

 

• Choose appropriate truth and training datasets, like HapMap and 1000 Genomes Project, depending on your project specifications.

Apply the Recalibration Model

• Run the `ApplyVQSR` command with the model generated, adjusting the level of stringency with the `--ts_filter_level` argument as needed (commonly 99.0 for SNPs and 90.0 for indels).

 

• Separate SNPs and indels for recalibration, as they typically require different weighting of annotations.

Inspect Results

• Examine your recalibrated VCF file to ensure the variant quality scores (VQSLOD) align with your expectations.

 

• Visualize results using tools like IGV and summarize variant quality using `grep` or shell scripts to extract variants marked as PASS or filtered.

Fine-Tune Settings

• If initial results are unsatisfactory, consider adjusting the set of annotations used in `VariantRecalibrator` or the `--ts_filter_level` to better suit your data's properties.

 

• Review GATK documentation and community forums for insights on parameter adjustments specific to your dataset characteristics.

Final Validation

• Perform a final validation step using independent, high-confidence datasets or experimental methods to ensure the integrity and accuracy of the filtered variants.

 

• Document and save the final version of the recalibrated VCF file for downstream analysis and reporting.