Problem upgrading from Perl Vardict to Java Vardict

[Graeme-Smith]

Hi.
I'm trying to upgrade a bioinformatics pipeline that uses a really old Perl version of Vardict from 2017. (This predates versioned releases of the tool, we're using the code at commit 328e00a). I'd like to upgrade to the latest Java Vardict version v1.8.3. My problem is that when I compare the two versions using a standard benchmark (Horizen HD200 FFPE control) I miss several variants which I am able to detect with the older tool. I think I'm probably failing to set a command line parameter required by the newer version, any pointers to the likely problem would be appreciated:

I'm running the identical commands.

# Command run for perl version:
git clone https://github.com/AstraZeneca-NGS/VarDict.git
git reset --hard 328e00a116 #revert all files to the production version currently in use.
perl_328e00a116/vardict -G genome.fa -f 0.01 -b ONCAA_02_EK6037_SWIFT57_Pan2684_S2_R1_001.refined.primerclipped.bam -c 1 -S 2 -E 3 -g 4 -q 22.5 -Q 10  Pan4081_flat.bed | tee perl_328e00a116_h200_results.txt | teststrandbias.R | var2vcf_valid.pl > perl_328e00a116_h200_results.vcf

# Command run for java versions:    
java_v1.8.2/VarDictJava/build/install/VarDict/bin/VarDict -th -G genome.fa -b ONCAA_02_EK6037_SWIFT57_Pan2684_S2_R1_001.refined.primerclipped.bam -f 0.01 -c 1 -S 2 -E 3 -g 4 -q 22.5 -Q 10 Pan4081_flat.bed | tee java_v1.8.2_results.txt | teststrandbias.R | var2vcf_valid.pl java_v1.8.2_results.vcf

• The differences in the files produced seems to be due to the main vardict script and not teststrandbias.R or var2vcf_valid.pl as the missing variants aren't present in the intermediate files. I have also run ~400 clinical samples between the two version and I am seeing similarly missed variants.
• I have run these commands with several versions of both the perl and java versions of vardict, as shown in this table. I get similiar results as the perl version 328e00a up to JavaVardict v1.5.4, but not after this.

[PolinaBevad]

Hi @Graeme-Smith ,

We changed some realignment methods during these years, improved merging of SNV and complex variants and changed filtering of reads, so some changes are expected. I see that most of the variants you show are SNVs - didn't they combine in some complex variants now? Can you recheck this region in IGV?