This is an introduction to the VCF stats report: how to create one and how to read it. The VCF stats report includes visualizations on several overall statistics from a VCF file (output of DeepVariant).
For new DeepVariant runs, this report is included in the outputs by default (to turn this off, add a --novcf_stats_report flag). For previous DeepVariant runs you can generate a report for an existing VCF file using the vcf_stats_report binary:
sudo docker run \
-v "${DATA_DIR}:/input" \
-v "${OUTPUT_DIR}:/output" \
google/deepvariant:"${BIN_VERSION}" \
/opt/deepvariant/bin/vcf_stats_report \
--input_vcf /input/${INPUT_VCF} \
--outfile_base /output/${OUTPUT}
This creates one output file: ${OUTPUT}.visual_report.html
Here we walk through all the charts to discuss how they are generated and how to interpret them.
Insertions have ALT longer than REF, where deletions are the opposite. SNPs have both ALT and REF of one base. Biallelic variants are those where there is only one alternate allele, while multiallelic variants have more than one alternate allele. Each variant is only counted once. Multi-allelic variants will be listed as “Mutliallelic_Deletion/Insertion/SNP” only if all alternate alleles are of the same type (e.g. all insertions) or as “Multiallelic_Complex” if not all alleles are of the same type. RefCalls are candidates that were determined to match the reference and are therefore not variants, although they are included in the VCF file (see FILTER column).
This chart contains a histogram of depths, which come from the DP sub-column of the VCF file. All entries in the VCF are included in this chart, including RefCalls. Any entries without a DP are ignored.
This is the QUAL column of the VCF file. See the VCF specification for how this is calculated, but in short, a high QUAL score indicates a low probability that the call shown in ALT is wrong. This chart shows quality scores from all rows of the VCF file, including RefCalls.
These numbers come from the GQ sub-column, listed in the FORMAT column of the VCF file. An important distinction here is that if the variant-caller is very sure that the variant is there, but not sure if it is heterozygous or homozygous, the QUAL score can be very high, but the genotype quality score (GQ) can be low. GQ is on a Phred scale, calculated as -10*log10(probability that the genotype is wrong). Any entries without a GQ are ignored.
The histograms show the variant allele frequency (VAF) distributions for different genotypes. Black guiding lines are shown to indicate the theoretical VAF for the main genotypes. For example heterozygous variants should have about as many variant-supporting as reference-supporting reads, for a VAF of 0.5. The reference calls will not usually show a VAF as low as 0 because otherwise they wouldn’t have been flagged as candidates in the first place. The genotypes are based on the GT sub-column and consolidated. For example, 0/1 and 0/2 both become Het (0/x). 1/1 and 3/3 are Hom (x/x). Het - both variants (x/y) includes all calls with two different alternate alleles, such as 1/2 or 3/5.
Of all biallelic SNPs, this shows the counts from a particular REF (along the top labeling the four charts) to a particular ALT (each bar within the charts labeled at the bottom). See the Ti/Tv section for a brief explanation of why some of these base changes tend to be more frequent than others. RefCalls and multi-allelic variants are not included.
Transition (Ti) count is the number of biallelic SNPs going from purine->purine or pyrimidine->pyrimidine, where purines are A and G, pyrimidines are C and T. Transversions (Tv) are purine->pyrimidine or pyrimidine->purine. Transitions are biologically more likely to occur than transversions due to the molecular structure of the bases, so a ratio well above one is desirable. This article on the GATK forums has a good discussion of how to interpret the ratio. These include all biallelic SNPs, excluding RefCalls.
The sizes of all biallelic insertions and deletions are shown as histograms. The top and bottom plots show the same data, just on a linear scale on top and on a log scale on the bottom. RefCalls and multi-allelic variants are not included.