How to Determine Your Blood Type From DNA Sequencing Data
Yes, you can determine your ABO and Rh blood type from DNA data, and whole-genome sequencing does it accurately for the common alleles. ABO comes down to four well-characterized variants in the ABO gene on chromosome 9, and Rh(D) status is mostly a copy-number question at the RHD gene on chromosome 1. The important caveat is structural: a single-nucleotide genotype list is not enough, because ABO phenotype depends on which variants sit together on the same chromosome. Consumer genotyping arrays give you a partial and sometimes wrong answer for exactly this reason.
This page walks through the biology and the variants. It then covers the commands we would run on a 30x CRAM file, along with the specific places where a confident-looking call turns out to be incorrect.
The four ABO variants that do the work
ABO spans roughly 20 kb near the end of chromosome 9q34, at about 133.25–133.28 Mb in GRCh38. The gene encodes a glycosyltransferase. The A and B versions of the enzyme add different sugars to a precursor on the red cell surface. The O allele encodes no functional enzyme at all, so the precursor is left unmodified.
Four variants carry nearly all the common information:
- rs8176719, a single-base deletion at c.261 in exon 6. This frameshift defines the O1 allele, by far the most common O allele worldwide.
- rs41302905, c.802G>A in exon 7. This defines O2, a second inactivating allele that is much rarer but not negligible in European populations.
- rs8176746, c.796C>A (Leu266Met).
- rs8176747, c.803G>C (Gly268Ala).
The last two distinguish B from A. The B allele differs from the common A1 allele at four amino acid positions, and these two substitutions are the ones that change acceptor-sugar specificity. Genotyping rs8176719, rs41302905 and either of the B markers resolves the standard A/B/AB/O phenotype in the large majority of people.
Subgroups need more. The A2 allele carries c.1061delC (rs56392308) and produces a weakly expressed A antigen. Cis-AB and B(A) alleles carry single substitutions that let one enzyme make both antigens, so they break the simple two-allele logic entirely. These are rare, and they are the reason that serology and genotyping occasionally disagree. Blood group genotyping is now well enough characterized that discrepancies usually resolve in favor of the DNA once the full allele is sequenced 1.
Phase decides the answer
Suppose your variant call file shows you as heterozygous for the c.261 deletion and heterozygous at rs8176747. Two arrangements are possible. You are group B if the B substitution sits on the chromosome without the deletion. If it sits on the chromosome that already carries the frameshift, that chromosome is dead and the other one is a plain A allele. That makes you group A.
Genotypes alone cannot tell these apart. You need phase. The two sites sit in different exons separated by about a kilobase of intron, which is beyond the span of 150 bp paired-end reads at normal insert sizes. Read-backed phasing with a tool like WhatsHap usually fails here on short-read data.
Three ways out. Use a population reference panel to phase statistically, which works well because ABO haplotypes are old and tightly structured. Sequence with long reads, where a single ONT or PacBio read covers exons 6 and 7 together and the phase is directly observed. Or use a purpose-built caller. The bloodTyper tool was written for this problem and encodes ISBT allele definitions rather than treating each site independently, which we prefer to hand-rolled logic.
Calling ABO from a CRAM file
Start by pulling the locus, then call with explicit allele depths so you can inspect the evidence yourself.
samtools view -T GRCh38.fa -o abo.bam sample.cram chr9:133250000-133285000
samtools index abo.bam
bcftools mpileup -f GRCh38.fa -r chr9:133250000-133285000 \
-a AD,DP -q 20 -Q 20 abo.bam \
| bcftools call -m -Ou \
| bcftools norm -f GRCh38.fa -m -any -Ov -o abo.vcf
Resolve rsIDs to coordinates against your exact reference build before you trust any position you copied from a paper. dbSNP and Ensembl both give GRCh37 and GRCh38 positions, and the ABO literature is full of hg19 coordinates that will silently land in an intron on GRCh38.
Then inspect rs8176719 manually. It is a single-base deletion inside a short guanine run, which is the worst case for short-read indel calling. Aligners place the gap inconsistently, and some pipelines left-align it into a neighboring position. Look at the pileup directly:
samtools tview -p chr9:133257500 abo.bam GRCh38.fa
What you want to see is a clean roughly 50/50 split of reads carrying and lacking the base at 30x depth, with no strand bias. A 70/30 split usually means mapping artifact rather than a real allelic imbalance. If your input is a GVCF from a standard germline pipeline, check whether this indel was emitted at all before concluding you are homozygous for the reference.
Rh(D) is a copy-number problem that a SNP cannot show
RHD sits at about chr1:25.27–25.33 Mb in GRCh38, next to its near-twin RHCE. The two genes are 92 to 94 percent identical in sequence and arranged in opposite orientation between two homologous Rhesus boxes. The common European D-negative phenotype is a complete deletion of RHD produced by recombination between those boxes.
So genotype is not the first test. The first test is depth:
mosdepth --by rhd_exons.bed --fasta GRCh38.fa --mapq 0 rhd sample.cram
Compare median depth across RHD exons to RHCE exons and to the genome-wide median. A ratio near 1.0 indicates two RHD copies, near 0.5 indicates one, and near 0.0 indicates the homozygous deletion that yields a D-negative phenotype.
The failure mode here is mapping quality. Because RHD and RHCE are nearly identical, many reads are assigned MAPQ 0. If you filter at MAPQ 20, as you reasonably would for SNV calling, RHD depth collapses and you will call a deletion that does not exist. Run the depth calculation at MAPQ 0 and read the ratio against RHCE, which suffers the same ambiguity.
Two ancestry-specific alleles break the depth-only approach. RHD*Ψ, the pseudogene common in people of African ancestry, carries a 37 bp duplication in exon 4 and a nonsense change in exon 6. The gene is present, so depth looks normal, but no D antigen is made. The Asian-type DEL allele carries c.1227G>A and expresses D at a level too low for routine serology to detect. Both matter clinically, and both are invisible unless you check the specific variants. Allele frequencies at loci like this vary sharply between populations, which is a strong argument for reading your own genome rather than assuming a European reference frequency 2.
What sequencing gives you that a typing card does not
Serologic typing reports the antigens a specific antibody panel happens to test. Sequencing reports the alleles, which covers the more than forty blood group systems and the hundreds of antigens now defined at the DNA level 1. From one 30x whole genome you can also read several other systems. These include FY (Duffy) with the GATA promoter variant rs2814778, JK (Kidd) in SLC14A1, and KEL. The glycophorin genes GYPA and GYPB are readable too.
The Duffy result is the one most readers will find interesting beyond transfusion, because the erythroid-silencing promoter variant is associated with lower baseline neutrophil counts without any increase in infection risk. Interpreting your own complete blood count in that light is a conversation for a clinician who can see the actual numbers.
Do not use a genotype-derived blood type for transfusion, emergency care, or pregnancy management. Blood banks require serologic typing and a crossmatch. Rh management in pregnancy requires a clinician. A DNA-derived type is a hypothesis about your antigens, useful for understanding and for flagging questions, and the confirmation is a blood draw.
Whole-genome sequencing is the right substrate for all of this because one assay answers many questions at once, including structural variation that targeted tests miss 3. The same data supports metabolic and biomarker interpretation well beyond blood groups 4.
Questions people also ask
Can I get my blood type from 23andMe or AncestryDNA raw data?
Partially, and with real risk of error. Array chips have included rs8176719 on some versions and dropped it on others, and single-base deletions are genotyped poorly by hybridization probes. Arrays also give you unphased genotypes, so the A-versus-B ambiguity described above remains unresolved. Treat an array-derived blood type as a guess.
Which variants do I look up if I only have a genotype file?
Search your raw data file for rs8176719, rs41302905, rs8176746 and rs8176747. If rs8176719 is absent or reported as a no-call, you cannot determine ABO from that file. Some array exports use internal i-numbered identifiers for indels rather than rsIDs.
How accurate is blood typing from whole-genome sequencing?
For common ABO and RHD alleles at 30x coverage with a caller that uses ISBT allele definitions, concordance with serology is high, and disagreements often reflect rare alleles that serology cannot see 1. Accuracy drops for weak subgroups, hybrid RHD-RHCE alleles, and structurally complex glycophorin variants.
Why does my reference genome build matter so much here?
ABO and RHD coordinates differ by roughly 2.9 Mb and 0.4 Mb respectively between GRCh37 and GRCh38, and the Rhesus box region was corrected between builds. A position copied from an older paper into a GRCh38 pipeline will point at the wrong base. The reference itself has been revised repeatedly since the first human genome assembly was declared complete 5.
Does DNA tell me my Rh status reliably?
It tells you RHD copy number reliably, which covers the common European D-negative genotype. It requires targeted checks for the RHD pseudogene and the Asian-type DEL allele, and it cannot predict the antigen strength that serology measures.
Footnotes
-
Connie M. Westhoff. Blood group genotyping. Blood, 2019. https://doi.org/10.1182/blood-2018-11-833954 ↩ ↩2 ↩3
-
Israel Aguilar-Ordoñez, Eugenio Guzman-Cerezo, David Torres-Treviño, et al. Whole genome sequencing of 1427 Mexican individuals from the oriGen cohort. Nature Communications, 2026. https://doi.org/10.1038/s41467-026-77389-0 ↩
-
Sarra L. Ryan, John F. Peden, Zoya Kingsbury, et al. Whole genome sequencing provides comprehensive genetic testing in childhood B-cell acute lymphoblastic leukaemia. Leukemia, 2023. https://doi.org/10.1038/s41375-022-01806-8 ↩
-
Yi-Xuan Wang, Yi-Xuan Qiang, Yi-Jun Ge, et al. Analysis of whole genome sequencing and plasma metabolomics unveil genetic determinants and clinical implications for human health. Nature Communications, 2026. https://doi.org/10.1038/s41467-026-74781-8 ↩
-
Tabitha M Powledge. Human genome project completed. Genome Biology, 2003. https://doi.org/10.1186/gb-spotlight-20030415-01 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.