FilterBatch

WDL source code

Filters poor quality variants and outlier samples. This workflow can be run all at once with the top-level WDL, or it can be run in two steps to enable tuning of outlier filtration cutoffs. The two subworkflows are:

1. FilterBatchSites: Per-batch variant filtration. Visualize filtered SV counts per sample per type to help choose an IQR cutoff for outlier sample filtering, and preview outlier samples for a given cutoff.

2. FilterBatchSamples: Per-batch outlier sample filtration; provide an appropriate outlier_cutoff_nIQR based on the SV count plots and outlier previews from step 2. Note that not removing high outliers can result in increased compute cost and a higher false positive rate in later steps.

The following diagram illustrates the recommended invocation order:

Inputs

batch

An identifier for the batch. Should match the name used in GatherBatchEvidence.

*_vcf

Clustered VCFs from ClusterBatch

evidence_metrics

Metrics table GenerateBatchMetrics

outlier_cutoff_nIQR

Defines outlier sample cutoffs based on variant counts. Samples deviating from the batch median count by more than the given multiple of the interquartile range are hard filtered from the VCF. Recommended range is between 3 and 9 depending on desired sensitivity (higher is less stringent), or disable with 10000.

Optional outlier_cutoff_table

A cutoff table to set permissible nIQR ranges for each SVTYPE. If provided, overrides outlier_cutoff_nIQR. Expected columns are: algorithm, svtype, lower_cuff, higher_cff. See the outlier_cutoff_table resource in this json for an example table.

Optional adjudicate_cutoffs

A pre-computed table of random forest cutoffs that override the table that is generated in the workflow. It ensures that the workflow does not re-run the random forest model training if cutoffs are already available, or if setting these in a manual manner is desired.

Optional adjudicate_scores

A pre-computed table of variant scores that override the table that is generated in the workflow. It ensures that the workflow does not re-run the random forest model evaluation if scores are already available, or if setting these in a manual manner is desired.

Optional adjudicate_rf_files

A pre-computed series of interemediete random forest files that override what is generated in the workflow. It ensures that the workflow does not re-save these intermediete files if they are already available.

Outputs

filtered_depth_vcf

Depth-based CNV caller VCFs after variant and sample filtering.

filtered_pesr_vcf

PE/SR (non-depth) caller VCFs after variant and sample filtering.

cutoffs

Variant metric cutoffs for genotyping.

sv_counts

Array of TSVs containing SV counts for each sample, i.e. sample-svtype-count triplets. Each file corresponds to a different SV caller.

sv_count_plots

Array of images plotting SV counts stratified by SV type. Each file corresponds to a different SV caller.

outlier_samples_excluded_file

Text file of sample IDs excluded by outlier analysis.

filtered_batch_samples_file

Text file of remaining sample IDs after outlier exclusion.