Try DeepVariant as replacement for freebayes

[marc-sturm]

• try DeepVariant
  • benchmark performance with NA12878x2_93
  • benchmark runtime (1, 5, 10 threads, maybe on GPU)
    • test -num_shards option of DeepVariant and check CPU usage
    • test deepvariant-gpu
    • benchmark runtime and performance for gpu version
  • check if gVCF can be created and used to perform multi-sample analysis

    • gVCF can be created

      • gVCF merging with GLnexus (https://github.com/dnanexus-rnd/GLnexus) recommended

          /glnexus_cli --config DeepVariant multisample_test/*.gvcf | /mnt/storage2/megSAP/tools/bcftools-1.20/bcftools view | bgzip -@ 4 -c > dv_merged_trio.vcf.gz
        

      • gVCF merging of DeepVariant output gVCFs with gatk not compatible

          The list of input alleles must contain <NON_REF> as an allele but that is not the case within DeepVariant output gVCFs
        

    • check if dragen gVCFs can be merged with GLnexus

    • validate_NA12878.php on child VCF before and after merging

    • compare Mendelian error rate of merged deepvar gVCFs with vc_freebayes trio output

    • compare performance (validate_NA12878.php) of freebayes, dragen and deepvar on child

    • check if it can be used to perform multi-sample analysis

[KilianIlius]

Runtime/memory (variant calling on NA12878x2_93 WGS):

Algorithm	threads	runtime	RAM
DeepVariant CPU	10	07:32:48	33.956G
DeepVariant CPU	5	21:50:02	14.197G
DeepVariant CPU	1	61:45:49	18.904G
Freebayes	15	00:50:44	n/a

Validation Performance

Freebayes vs Deepvariant on NA12878x2_93

Name	Options	Date	Average Depth	Expected SNVs	Expected Indels	SNV Sensitivity	SNV PPV	SNV F1	SNV Genotyping Accuracy	Indel Sensitivity	Indel PPV	Indel F1	Indel Genotyping Accuracy	All Sensitivity	All PPV	All F1	All Genotyping Accuracy
Freebayes NA12878x2_93_var		2025-01-10 18:13:04	40.98	3,256,373	495,309	0.9934	0.9944	0.9939	0.9993	0.9754	0.9938	0.9845	0.9847	0.9911	0.9927	0.9943	0.9974
DeepVariant vc_deep_NA12878x2_93_t10		2025-01-09 00:47:46	40.98	3,256,373	495,309	0.9941	0.9993	0.9967	0.9997	0.9910	0.9970	0.9940	0.9986	0.9937	0.9963	0.9990	0.9996

Freebayes vs Deepvariant vs merged gVCFs of NA12878x2_80/NA12891_14/NA12892_18 trio

Name	Date	Avg Depth	Expected SNVs	Expected Indels	SNV Sens.	SNV PPV	SNV F1	SNV Genotyping Acc.	Indel Sens.	Indel PPV	Indel F1	Indel Genotyping Acc.	All Sens.	All PPV	All F1	All Genotyping Acc.
glnexus_merged_child_only_no_wt	2025-01-28 16:55:20	113.07	36057	2260	0.9852	0.9978	0.9914	0.9981	0.8805	0.8908	0.8856	0.9950	0.9790	0.9852	0.9915	0.9979
NA12878x2_80_deepvar_var	2025-01-17 20:50:25	113.07	36057	2260	0.9859	0.9984	0.9921	0.9992	0.9695	0.9874	0.9783	0.9963	0.9849	0.9913	0.9978	0.9990
NA12878x2_80_freebayes_var	2023-08-14 15:37:28	113.07	36057	2260	0.9842	0.9832	0.9837	0.9972	0.9593	0.9377	0.9484	0.9673	0.9828	0.9816	0.9805	0.9954
NA12878x2_80_dragen	2023-08-16 09:06:47	112.82	36057	2260	0.9859	0.9863	0.9861	0.9976	0.9221	0.9337	0.9279	0.9933	0.9821	0.9827	0.9832	0.9974

Mendelian Error Rate Comparison (Gonosomes where excluded)

Method	Child/Parent1/Parent2	Mendelian Error Rate	Total Variants	Mendelian Errors
GLnexus merged gVCFs (DeepVariant)	NA12878x2_80 / NA12891_14 / NA12892_18	6.31%	190,410	12,018
vc_freebayes (all.vcf.gz)	NA12878x2_80 / NA12891_14 / NA12892_18	1.73%	201,148	3,481