search

stat-lab / EvalSVcallers

Evaluate the performances (precision and recall) of structural variation (SV) callers
32 stars 13 forks source link

Inconsistency of output files #21

Open gamzemdn opened 3 months ago

gamzemdn commented 3 months ago

Hi Dear Kosugi,

When I analyse output files I see that caller.TF file has true positive INS type variants but caller.eval file shows lines in below: INS

2 3 4 5 6 7 8 9 10 12 Call (A) 43390 12809 3529 1385 779 579 500 458 425 348 Recall (A) 3 2.6 2.4 2.2 2.1 1.9 1.7 1.6 1.5 1.2 Precis (A) 0.9 2.8 9.4 22.5 37.4 45.7 48.4 48.6 48.9 47.9 convert: '/home/gamze/Downloads/EvalSVcallers-master/scripts/convert_SV_callers_vcf.pl' -t BreakDancer '/home/breakdancer.vcf' >breakdancer.vcf overlap: '/home/gamze/Downloads/EvalSVcallers-master/scripts/evaluate_SV_callers.pl' **-r N** '/home/gamze/Downloads/EvalSVcallers-master/breakdancer.vcf' -of 3 >breakdancer_overlap.vcf Why I only see A here not SS, S, M , L like in the other types? For example how can I interpret Recall(A) in terms of size? Best Regards, ![breakdancer6](https://github.com/stat-lab/EvalSVcallers/assets/81433698/25fe6166-54da-4aee-b01d-d1e9fb7b9e00)
stat-lab commented 3 months ago

Most SV calling tools with short read cannot accurately detect INS sizes, especially those larger than read length. So, our script for INS evaluation does not evaluate precision and recall according to the size of the INS, but only determine them for INS breakpoints.