Tools like HaploAI, Morley's MDKA predictor, and similar services promise to extract your Y-DNA or mtDNA haplogroup from the raw data file of a standard DNA ancestry test. They are useful, but they operate under a fundamental constraint that is getting worse every year: the major testing companies have been systematically reducing the number of Y-chromosome and mitochondrial markers on their SNP arrays. This is not an accident. It is, in part, a business decision. Understanding why, and what it means for haplogroup resolution, is essential for any serious genetic genealogy researcher.
1. The Difference Between an Autosomal Test and a Dedicated Haplogroup Test
When you purchase a consumer DNA ancestry test from AncestryDNA, 23andMe, MyHeritage, or LivingDNA, you receive what is technically an autosomal SNP genotyping array. The chip scans somewhere between 600,000 and 900,000 predetermined positions (SNPs, Single Nucleotide Polymorphisms) distributed across your 22 pairs of autosomes. These are the chromosomes responsible for your admixture composition, ethnicity estimates, and relationship matching, the core product of consumer testing.
Your Y chromosome and mitochondrial DNA are also present in your cells, but they are not the primary target of these arrays. They are added as a secondary layer, covering a curated subset of diagnostic positions. This means:
What You're Actually Buying
~600K, 900K SNPs across 22 autosomes. The commercial core product: ethnicity estimates, relative matching, admixture analysis.
Optional Bonus (for men)
A small subset of Y-SNPs included on the array, historically hundreds to a few thousand, now often dramatically reduced.
Optional Bonus (for all)
A subset of mitochondrial positions. Coverage varies widely by chip version and company. Full mt genome has 16,569 bp; arrays cover a fraction.
By contrast, a dedicated Y-DNA test (FTDNA's Y-37, Y-111, or BigY-700; YSeq's targeted panels; Nebula Genomics' whole-genome sequencing) is specifically engineered to interrogate Y-chromosome positions at much greater depth. A full mitochondrial sequencing (FMS) at FTDNA reads the entire ~16,569 bp mtDNA genome, resolving haplogroups to the deepest known branches.
Consumer autosomal tests were never designed for Y-DNA or mtDNA haplogroup resolution. Any haplogroup information derived from them is a bonus by-product, not the primary purpose, and its precision is limited by design.
2. The Progressive Erosion of Y & mtDNA Markers Across Chip Generations
Consumer DNA testing chips are not static. The major companies periodically switch to new versions of their arrays, driven by cost, new scientific priorities, and increasingly, by commercial strategy. Crucially, each successive generation of the most widely used chips has reduced the number of Y-chromosome markers.
Figure 1. Approximate number of Y-chromosome SNPs included in major commercial chip iterations. Note the strong downward trend from Ancestry v2 (~2,800 Y-SNPs) toward modern GSAMD-based chips (~200). Values are estimates based on published chip documentation and community-sourced analysis; exact counts vary by lot and company configuration.
The trend is unambiguous. The chips used by the major companies from approximately 2020 onward, predominantly variants of Illumina's Global Screening Array (GSA) and its successor the GSAMD, were designed for large-scale medical and biobank research. Their Y-chromosome content is optimized for population-level GWAS studies, not for genealogical haplogroup resolution. The consequences for haplogroup prediction are severe.
mtDNA: A Different but Related Problem
Mitochondrial DNA presents a somewhat different picture. The mtDNA genome is small, just ~16,569 base pairs, and is present in thousands of copies per cell, which means it tends to be well-covered even on arrays not specifically designed for it. However, coverage is not the same as density of diagnostic SNPs. What matters for haplogroup assignment is whether the array interrogates the phylogenetically informative positions that define branches in the global mt haplogroup tree (as curated by Phylotree). Older chip versions covered several hundred mt positions; newer versions have also seen reductions in the number and phylogenetic utility of mt SNPs included.
A chip may report 3,000 mtDNA positions yet still be unable to distinguish between two closely related haplogroups if the specific discriminating SNP is not on the array. Marker count alone does not determine haplogroup resolution. It is the phylogenetic informativeness of the included SNPs that matters, and this declines as chips shift toward medical rather than genealogical priorities.
3. The Business Logic Behind the Reduction
It would be naive to assume that the reduction in Y and mt markers is purely a technical consequence of chip optimization. There is a clear commercial incentive: every autosomal test customer who wants a precise Y-DNA haplogroup must, eventually, purchase a dedicated Y-DNA test.
Y-SNPs are absent from newer chips because the GSA was designed for medical research and simply prioritizes disease-associated variants over genealogical markers. This is a neutral technical trade-off.
Companies have no interest in limiting haplogroup resolution from autosomal kits.
The GSA design choice is real, but FTDNA, 23andMe, and AncestryDNA all sell or promote separate Y-DNA and mtDNA products. FTDNA's business model is almost entirely built on dedicated haplogroup and matching products.
When autosomal raw data no longer permits reliable haplogroup inference, the commercial pathway becomes: buy a separate test. The market structure creates a structural incentive to not saturate autosomal chips with haplogroup-resolving Y markers.
This is not a conspiracy, it reflects rational product differentiation. But it is a structural reality that users and tool developers must understand. The companies are not obligated to include genealogically useful Y-SNPs on their autosomal chips. As those chips are replaced by medical-purpose arrays, the haplogroup resolution available from standard autosomal raw data files continues to decline.
FTDNA BigY-700: Sequences ~200,000 Y-SNP positions. Defines haplogroups to the deepest available branch, including private variants. Gold standard for genealogical Y-DNA.
YSeq: Targeted Sanger sequencing of specific SNPs. Excellent for confirming or refuting a particular sub-clade after a prediction.
Nebula Genomics / Dante Labs (WGS): Whole-genome sequencing at 30× coverage reads the entire Y chromosome. Haplogroup resolution limited only by the reference panel used for annotation.
FTDNA mtFull Sequence (FMS): Reads all 16,569 bp of the mt genome. The only way to achieve Phylotree-level resolution for mtDNA.
4. What This Means for Haplogroup Prediction Tools
Tools that extract haplogroups from autosomal raw data files are performing a fundamentally constrained task. They take whatever Y or mt SNPs happen to be present in your file, a number that varies by company, chip version, and even individual lot, and compare them against known haplogroup-defining markers. The result is only ever as good as the underlying data.
HaploAI (haploai.exploreyourdna.com)
HaploAI uses a machine-learning approach to infer haplogroups from raw autosomal data. It is designed to extract the maximum possible resolution from whatever Y and mt markers are present in a given file. When a customer uploads a file from an older chip (e.g., AncestryDNA v2, 23andMe v4), the prediction can be remarkably precise. For a file from a modern GSAMD-based chip with only ~200, 600 Y-SNPs, the tool will produce a result, but the confidence interval necessarily widens and the terminal branch assignment becomes shallower.
HaploAI reports prediction confidence alongside haplogroup assignments. A result showing R-M269 from a modern chip with few Y markers does not mean R-M269 is the terminal haplogroup, it means R-M269 is the deepest node that the available SNPs can reliably confirm. The true terminal clade might be R-BY15773, R-FGC23343, or any of hundreds of downstream branches. Only dedicated sequencing can resolve this.
The Morley / MDKA Predictor
The MDKA (Most Distant Known Ancestor) predictor developed by Tim Janzen and the community around James Lick's mtDNA haplogroup predictor use rule-based SNP matching against haplogroup trees. These tools are highly dependent on the phylogenetic informativeness of the SNPs present in your file. With an older 23andMe v4 file containing ~900 Y-SNPs, the predictor might place you at a specific sub-clade four or five levels deep. With a modern v5 file containing ~400 Y-SNPs, the same individual might only be assigned to a haplogroup two levels deep, not because their ancestry changed, but because the relevant discriminating SNPs are simply absent from the file.
HaploGrep and PhyloTree-Based Tools
HaploGrep (designed for mtDNA) performs haplogroup assignment based on maximum parsimony against the Phylotree reference. Its quality score metric explicitly reflects the number of expected mutations found versus expected. When an autosomal raw file contains only a fraction of the diagnostic mt positions, the quality score degrades and the assigned haplogroup may be several branches too shallow, or, worse, a nearby incorrect branch may be ranked equally.
Figure 2. Illustrative comparison of haplogroup tree depth achievable by data source. Modern autosomal chips may only resolve R-M269; older chips with more Y-SNPs can reach R-Z2103; dedicated BigY-700 or WGS reaches the terminal private branch. The same prediction tool produces qualitatively different results depending on the chip version of the raw data file.
5. Practical Implications: What Can You Actually Trust?
The following table summarizes what level of haplogroup assignment is realistically achievable from each data source, and what tools are best suited for each scenario.
| Data Source | Approx. Y-SNPs | Y Haplogroup Depth | mt Haplogroup Depth | Suitable For |
|---|---|---|---|---|
| AncestryDNA v2 (2016) | ~2,500, 3,000 | Good, 4, 6 levels | Good | HaploAI, Morley, HaploGrep |
| 23andMe v4 (2017) | ~700, 1,000 | Moderate, 3, 4 levels | Good | HaploAI, Morley (limited depth) |
| AncestryDNA v3 (2018) | ~2,000, 2,500 | Good, 4, 5 levels | Good | HaploAI, Morley, HaploGrep |
| 23andMe v5 (2018+) | ~350, 500 | Limited, 2, 3 levels | Moderate | HaploAI (broad class only) |
| GSA-based chips (2020, 2023) | ~400, 700 | Limited, 2, 3 levels | Moderate | HaploAI (broad class only) |
| GSAMD (2023+) | ~150, 250 | Very limited, 1, 2 levels | Moderate | Broad assignment only; confirm with dedicated test |
| LivingDNA (dedicated panel) | ~15,000+ | Excellent, 6, 8+ levels | Excellent | Best autosomal chip for haplo depth |
| FTDNA BigY-700 | ~200,000+ | Terminal branch + private SNPs | N/A (separate FMS) | Gold standard Y-DNA; all tools |
| WGS (30×, e.g. Nebula) | Full Y chromosome | Terminal branch | Full mt genome | Best of all worlds; requires annotation |
| FTDNA Full mtDNA Sequence | N/A | N/A | Terminal branch, all 16,569 bp | Definitive mt haplogroup assignment |
The "Chip Version Problem" in Practice
A concrete example: a man who tested with AncestryDNA in 2016 (v2 chip) uploads his raw data to HaploAI and receives a precise prediction of R-Z2103 > CTS1987. His brother, who tested with the same company in 2023 (GSAMD chip), uploads his file and receives only R-M269. The two brothers have identical Y chromosomes, passed down from the same paternal lineage. The difference is entirely in the chip, not in biology. The 2023 file simply lacks the SNPs needed to go deeper.
If you see contradictory predictions between your result and a close paternal relative's result, always check which chip version generated each file before concluding there is a discrepancy in ancestry. The difference is almost always the data source, not the prediction tool.
6. What Prediction Tools Can and Cannot Do
Haplogroup prediction tools are, at their best, sophisticated inference engines that extract the maximum phylogenetic information from imperfect input data. Understanding their design helps users calibrate expectations.
HaploAI PredictionFree
What it does: Machine-learning model trained on thousands of genotyped individuals with known haplogroups. Infers the most probable haplogroup from available SNPs, with confidence scoring. Works for both Y and mtDNA from autosomal raw data. Best results with older chips (AncestryDNA v2/v3, 23andMe v4). Available at haploai.exploreyourdna.com.
Limitation: Result quality degrades proportionally with the number of informative SNPs available. A confident result from a modern GSAMD file should be understood as a probable broad clade, not a terminal haplogroup.
Morley / MDKA Y-DNA Predictor PredictionFree
What it does: Rule-based matching of detected Y-SNPs against a curated list of haplogroup-defining markers. Returns a haplogroup tree path based on the deepest confirmed positive SNP. Reliable and transparent, you can see exactly which SNPs were used.
Limitation: Requires the chip to have tested the relevant diagnostic SNPs. With sparse modern chips, many defining SNPs will be missing, forcing the predictor to stop at a shallow ancestor node.
HaploGrep 3 PredictionFree
What it does: Maximum parsimony assignment of mtDNA haplogroups against Phylotree. Accepts VCF, FASTA, or Haplogrep format. Provides a quality score that directly reflects how many expected phylogenetic markers were found.
Limitation: Quality score from autosomal raw files is often mediocre because the mt SNPs on commercial arrays do not align well with Phylotree's curated discriminating positions. A "good" HaploGrep result from a consumer file may still reflect a wrong branch assignment two levels above the true terminal clade.
YSeq SequencingPaid
What it does: Targeted Sanger sequencing of specific Y-SNPs. Ideal workflow: use a prediction tool to identify a probable haplogroup, then order a targeted panel from YSeq to confirm or deny specific sub-clades. Prices start around $15, 18 per SNP or panel.
Best use case: Confirmation and refinement after an autosomal-based prediction. If HaploAI suggests R-Z2103, a YSeq panel for CTS1987, Z2106, and related SNPs will definitively place you.
FTDNA BigY-700 SequencingPaid
What it does: NGS sequencing of ~15 Mb of the Y chromosome covering ~23,000 named SNPs plus discovery of novel private variants. Assigns terminal haplogroup with certainty. Allows matching with other BigY testers sharing private variants, equivalent to Y-STR surname-level matching but at the SNP level.
Best use case: Any serious genealogical research involving paternal lineage. Non-negotiable for brick-wall paternal lines.
7. How to Get the Most From Your Existing Raw Data
If you already have an autosomal raw data file and want the best possible haplogroup prediction before investing in a dedicated test, the following workflow maximizes your results:
Identify Your Chip Version
Check your testing company's account settings or the file header/metadata in your raw data. This tells you how many Y/mt SNPs to expect and calibrates your confidence in any prediction.
Run HaploAI First
Upload to haploai.exploreyourdna.com. Note both the predicted haplogroup and the confidence score. A low confidence score on a modern chip is informative, it tells you where the data runs out.
Cross-Check With Morley
For Y-DNA, the Morley predictor shows you exactly which SNPs were detected. If your result stops at R-M269 despite knowing your father tested as R-Z2103 via FTDNA, you now know why, the chip simply did not test the Z2103 position.
Use HaploGrep for mtDNA
Extract your mt calls and run through HaploGrep 3. The quality score will tell you how confidently the assignment was made. Below 70%, consider the result a broad classification only.
Confirm With Dedicated Testing
If haplogroup resolution matters for your research (genealogy, biogeographic ancestry, deep ancestry), the prediction is a starting point, not a final answer. Budget for a YSeq panel or FTDNA BigY/FMS.
Consider LivingDNA or WGS
For future tests, LivingDNA's chip includes a dedicated Y/mt panel (~15,000+ Y-SNPs). For comprehensive genomics, 30× WGS provides the full Y chromosome with no SNP selection bias at all.
8. The Bottom Line for Genetic Genealogists
Haplogroup prediction from autosomal raw data is a genuinely valuable tool, particularly for the millions of people who tested before dedicated Y or mt products became widely accessible, and for those who want a preliminary orientation before investing in more expensive sequencing. But the landscape is changing in an unfavorable direction for this use case.
The progressive shift toward GSA and GSAMD chips means that new kits purchased today from most major providers will yield less haplogroup information than kits purchased five years ago. This is not a failure of the prediction algorithms, tools like HaploAI, Morley's predictor, and HaploGrep are performing impressively with constrained input. The limitation is upstream, at the chip design level.
Understanding this distinction is the first step toward using these tools wisely. A prediction from a modern autosomal kit is a hypothesis to be tested, not a conclusion. The confirmation requires dedicated sequencing, and that is, increasingly, by design.
1. Consumer autosomal DNA tests include Y-chromosome and mtDNA markers as a secondary layer, not as their primary purpose.
2. The number of Y-SNPs on mainstream chips has declined dramatically since 2016, driven by a shift to medically-oriented GSA chips, and, arguably, by commercial product differentiation.
3. Prediction tools (HaploAI, Morley, HaploGrep) are only as accurate as the input data. A shallow result from a modern chip is not a tool failure, it reflects genuine data scarcity.
4. For confident terminal haplogroup assignment, dedicated Y-DNA testing (FTDNA BigY-700, YSeq) or full mtDNA sequencing (FTDNA FMS) remains the only reliable path.
5. Always check chip version before comparing haplogroup results across individuals or across prediction tools.
References & Further Reading
- ISOGG Wiki. Haplogroup predictor tools. International Society of Genetic Genealogy. Continuously updated. https://isogg.org/wiki/Haplogroup_predictor_tools
- van Oven, M., & Kayser, M. (2009). Updated comprehensive phylogenetic tree of global human mitochondrial DNA variation. Human Mutation, 30(2), E386, E394. Phylotree Build 17. https://doi.org/10.1002/humu.20921
- Weissensteiner, H., et al. (2016). HaploGrep 2: mitochondrial haplogroup classification in the era of high-throughput sequencing. Nucleic Acids Research, 44(W1), W58, W63. https://doi.org/10.1093/nar/gkw233
- Illumina, Inc. (2019). Infinium Global Screening Array-24 v3.0 Product Files. Illumina technical documentation. https://support.illumina.com/downloads/infinium-global-screening-array-v3-0-product-files.html
- Elhaik, E., et al. (2019). The diversity of recent and ancient human genomes and the future of genealogical genomics. Genome Biology and Evolution, 11(4), 1042, 1048. https://doi.org/10.1093/gbe/evz026
- Karmin, M., et al. (2015). A recent bottleneck of Y chromosome diversity coincides with a global change in culture. Genome Research, 25(4), 459, 466. https://doi.org/10.1101/gr.186684.114
- Batini, C., et al. (2015). Large-scale recent expansion of European patrilineages inferred from population genomic data. Nature Communications, 6, 7152. https://doi.org/10.1038/ncomms7152
- Family Tree DNA. BigY-700 Technical Documentation. FTDNA Learning Center. https://www.familytreedna.com/learn/y-dna-testing/y-str/bigy/
- ISOGG. Chip comparison table, Y-chromosome SNP coverage. Community documentation, ISOGG Wiki. https://isogg.org/wiki/Autosomal_DNA_testing_comparison_chart
- ExploreYourDNA. HaploAI, Haplogroup Prediction from Raw Data. https://haploai.exploreyourdna.com/