Step 36: PGx Consensus (PharmCAT Outside Calls)¶

What This Does¶

Decides what the BAM-based callers tell PharmCAT. PharmCAT (step 7) reads a VCF, and from a VCF it types neither the HLA genes nor CYP2D6, so its drug guidance for them (abacavir, carbamazepine, allopurinol, codeine, tramadol, tamoxifen and many more) could never fire. This step writes PharmCAT's outside-call file:

HLA-A and HLA-B from T1K (step 8), cut to two fields (*57:01), when every allele has a T1K quality above 0. When step 8 also ran OptiType, the opt-in second caller (HLA_QC=true, --hla_qc true in Nextflow), a gene passes only when both callers give the same pair at two fields; see step 8. One allele reads as homozygous, the way T1K reports a homozygous gene.
CYP2D6 only when pypgx (step 32) and Cyrius (step 21, once setup.sh --cyrius installed it) give the same diplotype and the CYP2D6 depth check passed. One caller alone, a disagreement, or multi-mapped reads leave CYP2D6 indeterminate, and nothing reaches PharmCAT.

It also writes a consensus table that says, for each of the three genes, what every caller said and what was passed on. Step 27 copies it into the CPIC recommendations.

Why¶

A single depth-based CYP2D6 call can be wrong in ways that change a prescription: on a BAM aligned to a reference with ALT contigs, the depth at CYP2D6 drops and pypgx can report a whole-gene deletion that is not there (lessons learned). Two independent callers agreeing, on depth that passed its check, is the bar for letting a call steer drug guidance. PharmCAT 3.4 makes no CYP2D6 call of its own from a VCF (its report gives CYP2D6 callSource NONE), so it cannot be the second caller.

Tool¶

bin/pgx_outside_calls.py (standard library only), the same code as the PGX_CONSENSUS Nextflow module

Docker Image¶

PYTHON_IMAGE

Input¶

Each one is read when it exists, and the table says when it does not:

File	From
`${SAMPLE}/hla_t1k/${SAMPLE}_hla_genotype.tsv`	step 8
`${SAMPLE}/pypgx/${SAMPLE}_pypgx_summary.tsv`	step 32
`${SAMPLE}/cyrius/${SAMPLE}_cyp2d6.tsv`	step 21 (after `setup.sh --cyrius`)
`${SAMPLE}/pypgx/${SAMPLE}_cyp2d6_depth_check.tsv` (else step 21's)	the CYP2D6 depth check of step 32 or 21

Command¶

./scripts/08-hla-typing.sh your_name
./scripts/32-pypgx.sh your_name
./scripts/21-cyrius.sh your_name          # setup.sh --cyrius first
./scripts/36-pgx-consensus.sh your_name
./scripts/07-pharmacogenomics.sh your_name   # reads the outside calls
./scripts/27-cpic-lookup.sh your_name        # lists them

The Nextflow pipeline (and so run-all.sh) runs PGX_CONSENSUS for every sample with a BAM when pharmcat and any of hla_typing, pypgx or cyrius are in --tools, and PharmCAT waits for it.

Output¶

All output is written to ${GENOME_DIR}/${SAMPLE}/pgx_consensus/.

File	Contents
`${SAMPLE}_outside_calls.tsv`	PharmCAT's outside-call file: `gene<TAB>diplotype`, one line per gene passed on; empty when none was. Step 7 passes it with `-po` when it is not empty.
`${SAMPLE}_pgx_consensus.tsv`	`Gene`, `Result` (the diplotype, `indeterminate` or `not typed`), `Outside_call` (`yes` or `no`), `Reason`, `Evidence` (what each caller said)

For CYP2D6 the Reason is one of: the depth check's message (multi-mapped reads), no caller made a call, one caller only: a second caller must agree (Cyrius: setup.sh --cyrius), pypgx and Cyrius disagree, the CYP2D6 depth check did not run, or pypgx and Cyrius agree; the depth check passed.

Runtime¶

1 to 7 seconds, measured in three 30-40x WGS runs (runtime). PharmCAT (step 7) then took 6 to 16 seconds in the pipeline.

Notes¶

Without Cyrius, CYP2D6 is always indeterminate here: that is the point, not a fault. Install Cyrius (setup.sh --cyrius, non-commercial licence) to get an agreed call; run-all.sh then runs step 21 in a default run.
A probable call, never an outside call. When pypgx alone called CYP2D6 (Cyrius did not run, or made no usable call), the depth check passed, and every variant that defines its star alleles is in the VCF as a PASS call at or above the GQ floor (step 32's Defining_filter and Quality columns), both reports show pypgx's diplotype and phenotype labelled probable (one caller: pypgx). This table still says indeterminate for it and nothing reaches PharmCAT, so no drug guidance rests on it. The rule is probable_cyp2d6 in bin/pgx_outside_calls.py; a failed depth check, a usable Cyrius call (then the two callers decide), a defining variant that is missing, not PASS, below the GQ floor or without a GQ, a star allele pypgx's table lacks, or a summary from before these columns gives no probable call.
A Cyrius call counts only with Filter PASS; None, an ambiguous call (two diplotypes) or CYP2D6_depth_unreliable is no call.
The two diplotypes must be the same up to the order of the alleles. The one passed to PharmCAT is pypgx's.
PharmCAT turns the HLA alleles into its own phenotypes (*57:01 positive, *58:01 negative and so on) and gives the drug guidance for them; step 27 marks those genes [outside call].
Step 7 run before this step gives a report without the outside calls. Run step 7 again after it, as the command order above does.
Steps 8, 21 and 32 delete this step's two files when they start: a call made from an earlier result never reaches PharmCAT. After running any of them again, run this step and step 7 again.