A circos plot is most likely not the most appropriate solution here. What I would suggest is a confusion matrix, of which you can find an example here:
For every variant in your vcf you'll add a number in this matrix. One sample is the columns, the other is the lines. If your variant is homozygous in both, then you add in that square +1 (the cell with 5845 in the example).
A perfect concordant sample pair will have only variants on the diagonal.
Here is some python code to get such a matrix. It uses cyvcf2 and pandas, and expects as input a vcf file with both samples.
from argparse import ArgumentParser
from cyvcf2 import VCF
import pandas as pd
def main():
args = get_args()
confusion_matrix(args.vcf)
def confusion_matrix(vcff):
"""
First level of the dict is the "first" call, second level is the "second" sample
0: hom_ref
1: heterozygous
2: unknown/nocall
3: hom_alt
"""
zygosities = {0: {0: 0, 1: 0, 2: 0, 3: 0},
1: {0: 0, 1: 0, 2: 0, 3: 0},
2: {0: 0, 1: 0, 2: 0, 3: 0},
3: {0: 0, 1: 0, 2: 0, 3: 0},
}
for v in VCF(vcff):
zygosities[v.gt_types[0]][v.gt_types[1]] += 1
zygs = [2, 0, 1, 3]
df = pd.DataFrame(index=zygs, columns=zygs)
for tr in zygs:
for te in zygs:
df.loc[tr, te] = zygosities[tr][te]
df.columns = ['nocall', 'hom_ref', 'het', 'hom_alt']
df.index = ['nocall', 'hom_ref', 'het', 'hom_alt']
print(df)
def get_args():
parser = ArgumentParser(description="Create confusion matrix of SNV calls")
parser.add_argument("vcf", help="vcf containing two samples")
return parser.parse_args()
if __name__ == '__main__':
main()
