Proto Indo Europeans 23andme format

In this video, I use the YSEQ Phenotype Predictor and Snipper 4 (the newer version of Snipper 3) to classify eye color, hair color, and skin pigmentation in several proto–Indo-European DNA samples I collected from the European Nucleotide Archive. I also briefly touch on why YSEQ’s phenotype predictor is inferior to Snipper 3/4 and to my own Trait Predictor, especially in cases where ancient samples suffer from insufficient coverage or SNP dislinkages. Here, I want to expand on what those cases actually mean.

YSEQ and HIrisPlex rely almost entirely on a single SNP in the HERC2 gene: rs12913832. One major failure case is insufficient coverage, when a sample simply does not contain this SNP. When that happens, YSEQ displays the familiar “sunglasses” icon. This is a serious problem, because rs12913832 is the single most decisive SNP affecting eye color and hair color, and strongly influences skin pigmentation as well. Lacking this marker severely degrades prediction accuracy.

The second issue, dislinkages, is more subtle but just as important. rs12913832 is normally in very tight linkage with several other SNPs in HERC2, most notably rs1129038 and rs1667394. These SNPs also have independent effects on pigmentation and are usually inherited together as part of a haplotype. For example, if a sample has rs12913832 = GG, it will very often also carry rs1667394 = TT and rs1129038 = TT.

However, there are cases, especially in ancient DNA, where this linkage is broken. In these dislinkage scenarios, the expected haplotype structure is missing. YSEQ, which relies on a single SNP and assumes the presence of the full haplotype, performs very poorly under these conditions and can produce highly misleading results.

Snipper 3/4 and my Trait Predictor are designed to handle both of these problems. Snipper 4 uses seven SNPs in HERC2 and seven additional SNPs in the nearby OCA2 gene for eye color prediction, along with two SNPs in HERC2 and two in OCA2 for hair color. My Trait Predictor goes even further, capturing more SNPs across this region and including an imputation mechanism that can infer missing variants when key HERC2 SNPs are absent entirely.

This makes Snipper 4 and my Trait Predictor far more robust for ancient samples, especially those with low coverage or unusual haplotype structures, exactly the situations where single-SNP methods like YSEQ tend to fail.