TraitPredictor of Bronze age Greek (Log04) - Greek DNA

Main Ideas / Concepts Conveyed

• The video analyzes a specific ancient DNA sample (“Log04”), described as:
  • From Middle Bronze Age Greece
  • Early Greek / Indo-European context
  • Female
• The host uses multiple genetic analysis tools to infer:
  • Ancestry components / closest modern populations
  • Physical traits (e.g., eye color, hair color/texture, skin tone, nose shape)
  • Medical and biomarker predispositions (e.g., vitamin D, lipids, glucose, blood counts)
  • Disease risk panels, including:
    • monogenic-style predictions
    • HLA-based autoimmune risk
• A recurring theme is how genotypes explain predicted phenotypes, and how different tools can disagree because they:
  • use different variant sets
  • use different models
• The host emphasizes data quality concerns:
  • Auto-uploaded/archived datasets (especially 1240k-style files) can contain genotyping “mis-calls”
  • These mis-calls may falsely generate rare-disease risk signals
  • The host argues many “rare variant” findings are likely artifacts, not genuine biology
• The host demonstrates a practical workflow:
  1. Identify genotypes/variants
  2. Run phenotype/biomarker/disease predictors
  3. Compare predictions across tools
  4. Interpret what’s plausible vs. implausible

---

Detailed Methodology / Instruction-Like Workflow (as Presented)

1) Start With Ancestry Model

• Use a tool/model referred to as the “Davitz Global model” with “G25”:
  • Interprets ancestry likelihood as a mixture
  • Example claim: the individual scores mainly Anatolian Farmer-type ancestry
  • No West Hunter-Gatherer or additional components are present in the calculator
• Use distance / closest-population logic:
  • Closest modern matches include Greeks and various Balkan groups
  • Places/groups mentioned in subtitles include (spelled variably): Pomaks/Bulgarians/Turkish from Bulgaria/Macedonia Greeks/Italians/Albanians/Gagou* (from Romania)
  • Conclusion: she appears more Northern than Greeks generally

2) Run Y-DNA vs mtDNA Logic (Sex-Specific)

• Use a “MRE predictor / m… predictor”:
  • Clarifies it does not predict mitochondrial DNA, only Y-DNA
  • Since the sample is female, no Y-DNA prediction is shown

3) Run Physical Trait Prediction Tools

• Use a “Nasak cot calculator”:
  • Eye color likelihood: brown ~80.9%
  • Hair color likelihood: dark brown ~97%
  • Skin tone likelihood: olive / “Mediterranean” ~89%
  • Hair texture: curly ~53%
  • Nose shape: discussed as intermediate, host selects “snub” based on a threshold > 50%
• Validate using direct genotype inspection:
  • Notes absence of certain “light color variant” entries in specific genotype categories
  • Argues this supports dark predicted traits and makes very pale skin predictions implausible
• Compare results across predictors (model disagreement):
  • The host compares WEYC outputs to their tool predictions
  • Host argues WEYC predicts very pale skin and red hair, but this conflicts with:
    • underlying HERC2/OCA2 pigmentation effects
    • the sample’s genotype makeup
  • Pigmentation reasoning mentions:
    • HERC2 and OCA2
    • “light color” variants in SLC45A2 and OCA-related positions (subtitles list these imprecisely)

4) Use an Additional Hair/Eye Phenotyping Tool (“Snipper 3”)

• Snipper 3 for eye color:
  • Host claims “brown is obvious/useless” (as summarized in the video)
• Snipper 3 for hair color:
  • Despite MC1R evidence suggesting red-hair genotype, Snipper 3 predicts brown hair
• Host’s interpretation:
  • Disagreement stems from different numbers of relevant variants in each tool

5) Use “Phenotype Oracle” and Face-Morph Visualization

• Use “Phenotype Oracle”:
  • Discusses closest predicted phenotypes and “distance” scores (e.g., ~0.5 and ~0.6 range)
• Create a blended phenotype:
  • Takes the top and bottom female phenotype images
  • Then “snaps” and morphs them (“face morph”)
• Host adjusts interpretation:
  • Claims the face-morph may lighten eye color, so visuals may differ slightly from genetic predictions

---

Biomarker / Medical Risk Prediction

Biomarkers

• Vitamin D: claims very low levels due to multiple variants linked to reduced vitamin D
• Lipids: predicts higher LDL and lower HDL
• Glucose: states “very high” glucose (host corrects an earlier mistaken subtitle read)
• Blood pressure: slightly above average; host notes BP is environment-sensitive and not strongly emphasized in ethnicity-focused reporting
• Iron: predicts lower iron; no hemochromatosis variants
• Telomere length: average
• Height: average or slightly above average
• Blood type: predicts Type A, and notes a pattern observed in other Greek samples

Complex Disease Panel Interpretations

• Reports polygenic risk scores for traits/diseases such as:
  • kidney stones
  • “doome” / unknown term (as presented)
  • gout
  • glaucoma (including subtype references)
  • leukemia
  • myopia
  • cardiovascular-related risks (including atrial fibrillation and clotting)
• Relative framing:
  • Low gout odds (very low number)
  • High leukemia odds, framed as more common in European contexts
  • Male pattern hair loss: high (host links this to a stereotype associated with Greek/Italian/Jewish Greek populations)
  • Cardiovascular risks: atrial fibrillation/clotting below average

---

Monogenic / Trait-Specific Panels and Genotype Checks

• “Warrior” phenotype:
  • Predicted as intermediate
  • Linked to dopamine-related enzymes/genes:
    • COMT, MAOA, MAOB
    • dopamine receptor genes including DRD1–DRD5
• Lactase persistence:
  • Host claims she lacks key European lactase persistence variants (therefore lactose intolerant)
• Empathy:
  • Predicts intermediate empathy based on OXTR-associated SNPs

---

Important Quality Control Instructions (Implicit Methodology)

• When seeing extremely rare risk variants across many unrelated rare diseases:
  • Host interprets this as a sign of genotyping errors / miscalls due to file-quality limits
• Suggested approach:
  • Confirm rare calls using higher-quality direct-to-consumer genotyping results (e.g., 23andMe, MyHeritage, AncestryDNA)
• Host’s core distinction:
  • Rare single-variant signals are less suspicious in high-quality datasets
  • But in 1240k/European archive files, many rare signals likely reflect artifacts

---

Main Lessons / Takeaways

• Genotype-informed prediction can be more reliable than cross-tool outputs when models disagree—especially when anchored to known biology like HERC2/OCA2 pigmentation.
• Not all “rare variant disease” outputs should be trusted, particularly from low-coverage or archived ancient DNA datasets, because mis-calls can generate false signals.
• Prediction differences can be explained by:
  • tool/model structure
  • SNP coverage (some tools consider more relevant SNPs than others)
• Ancient samples may still display European-like visible traits, even with what the host describes as more “exotic” ancestry/variant profiles.

---

Speakers / Sources Featured

Primary Speaker

• The video host/author (specific name not provided in the subtitles)

Tools / Models Referenced as Prediction Sources

• Davitz Global model (with G25)
• MRE predictor / m… predictor (Y-DNA only; no mtDNA)
• Nasak cot calculator
• WEYC (compared against the host’s pigmentation interpretation)
• Snipper 3
• Phenotype Oracle
• Face morph / face morphing (visual blending method)
• “RW DNA file” / “raw DNA file” (dataset file referenced)
• A trait predictor executable sold by the host (referred to as “trade predictor / TR/triade predictor” in subtitles)
• Mentions external tools/matching systems:
  • GD Match
  • GEDmatch (spelled variably in subtitles)
  • “GD match or whatever” (as phrased in the video)

Dataset / Repository Mentioned

• European Nucleotide Archive (ENA) (where file-quality issues are discussed)
• 1240k files / 1240k-style datasets (ancient DNA panel type leading to miscalls)