The output files contain metrics like precision, recall and TP/FP/FN counts for all variants. Also, we a VCF file giving the assessment for each variant call.
Output tables are in CSV format.
| Metric 1 | Metric 2 | ... | |
|---|---|---|---|
| Location/count type 1 | < value > | < value > | ... |
| Location/count type 2 | < value > | < value > | ... |
| ... | < value > | < value > | ... |
Here is an example from hap.py:
| METRIC.Recall | METRIC.Precision | ... | |
|---|---|---|---|
| Locations.SNP | 0.97 | 0.99 | ... |
| Locations.INDEL | 0.8 | 0.95 | ... |
| ... | ... | ... | ... |
The output VCF file is similar to the intermediate VCF file.
We add one additional variant classification: in
addition to FP/FN/TP, each variant call can also be assigned the status
UNK for unknown / outside the regions which the truthset covers.
A simple definition for UNK variants is as follows: We call any variant
unknown if it BT == FP and BK == miss and if the variant is outside
the confident call regions of the truth set. If the truthset does not give
confident call regions, no UNK variants are output.
We introduce the following additional fields:
##INFO=<ID=Regions,Number=.,Type=String,Description="Tags for confident / stratification regions.">
##FORMAT=<ID=BVT,Number=1,Type=String,Description="High-level variant type in truth/query (SNP|INDEL|COMPLEX).">
##FORMAT=<ID=BLT,Number=1,Type=String,Description="High-level location type in truth/query (het|hom|hetalt|halfcall|multiallelic|nocall).">
Optional: variant types seen and counted for this record.
##INFO=<ID=VTC,Number=.,Type=String,Description="Variant types used for counting.">