different bcftools consensus versions produce different results at multiple ALT variants

[MarieLataretu]

Hi there,

with two different bcftools versions installed via conda (bcftools 1.11, Using htslib 1.11 and bcftools 1.14, Using htslib 1.14). I get different results for the same command and input (SARS-CoV-2 sample):

bcftools consensus \
                    -I \
                    -o ${vcf.baseName}.iupac_consensus.tmp \
                    -f ${reference} \
                    -m ${mask_bed} \
                    --sample ${name} \
                    ${vcf}

bcftools 1.11 applies the deletion, bcftools 1.14 the substitution for this position:

#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	sample
MN908947.3	11073	.	TC	T,TT	45167.2	PASS	AB=0.155887,0.841349;ABP=1863.62,1833.85;AC=1,1;AF=0.5,0.5;AN=2;AO=282,1522;CIGAR=1M1D1M,1M1X1M;DP=1809;DPB=1724.67;DPRA=0,0;EPP=11.9118,36.8465;EPPR=9.52472;GTI=0;LEN=1,1;MEANALT=4,4;MQM=59.9504,59.9382;MQMR=60;NS=1;NUMALT=2;ODDS=76.3545;PAIRED=0.985816,0.979632;PAIREDR=1;PAO=9.66667,9.66667;PQA=321.333,321.333;PQR=321.333;PRO=9.66667;QA=9034,50976;QR=102;RO=3;RPL=66,473;RPP=176.266,476.363;RPPR=9.52472;RPR=216,1049;RUN=1,1;SAF=186,1021;SAP=65.3824,388.796;SAR=96,501;SRF=0;SRP=9.52472;SRR=3;TYPE=del,snp;technology.ILLUMINA=1,1	GT:DP:AD:RO:QR:AO:QA:GL	1/2:1809:3,282,1522:3:102:282,1522:9034,50976:-4846.67,-4119.65,-4043.04,-719.418,0,-269.525

(from top: reference, 1.11, 1.14)

Variants are called with freebayes and normalized with bcftools.

Is that behavior expected? I haven't found something matching in the change log.

[MarieLataretu]

After some testing, it seems like 1.14 always decides for the longer variant independent of FORMAT/GT.

[pd3]

It was probably this commit that introduced the difference 89d61e6.
Also check the difference between -I and -H I: the former applies REF+ALT regardless of the genotype, the latter uses FORMAT/GT.

[MarieLataretu]

I see, FORMAT/GT is used only with --sample and -H.
And the correct IPAC code for that example with -I is TT