# The Science Behind Baby Predictions

_OCA2/HERC2 two-gene eye color model, Punnett square implementation, Premium's three-generation analysis, and how the confidence score is computed up to its 95% cap._

- Source: https://proba.baby/en/genetics-science
- Language: English
- Last verified: v1.0.4 · 7 Sep 2026

## Eye Color Genetics

Two genes, HERC2 and OCA2, solved as two Punnett squares and combined into six eye colors

- **OCA2 Gene (Primary)** — OCA2 sets how much melanin the iris makes. It carries the P (pigment) and p alleles: two P's give brown or amber, one gives hazel or green, none gives blue. Chromosome 15.
- **HERC2 Gene (Regulator)** — HERC2 switches OCA2 on or off, and it is the gene the blue-eye SNP sits in. A b/b genotype shuts OCA2 down for blue or gray eyes; a B allele lets OCA2's pigment through. Together the two genes produce six eye colors: brown, amber, hazel, green, gray and blue.
- **Punnett Square Engine** — Each of the two eye-color genes is solved with its own 2×2 Punnett square (four outcomes each), then combined into six phenotypes. Hair color runs a dominance series across ten shades, with a separate MC1R square for red — and a parent's childhood hair color is treated as evidence of a hidden recessive allele.

## Genotype to Phenotype Mapping

How gene combinations determine eye color

| HERC2 × OCA2 | Eye Color | Explanation |
| --- | --- | --- |
| BB × PP | Brown | HERC2 fully active and both OCA2 alleles pigment-producing — maximum iris melanin. |
| Bb × PP | Amber | One HERC2 allele switched off, but full OCA2 pigment — the iris lands on amber. |
| BB × Pp | Hazel | Full HERC2 activity with only half the OCA2 pigment — the mixed hazel iris. |
| Bb × Pp | Green | Half the HERC2 activity and half the OCA2 pigment — green. |
| bb × P_ | Gray | HERC2 switched off, but OCA2 still carries a pigment allele — gray rather than pure blue. |
| __ × pp | Blue | No OCA2 pigment allele at all — the iris has no melanin to show, so blue. |

## Polygenic Trait Models

Skin tone is polygenic; height we report as a mid-parental band rather than model from genes

- **Skin Tone (3 Genes)** — An additive model over SLC24A5 (largest effect), SLC45A2 and TYR, producing a 0–100 melanin index reported with 68% and 95% confidence bands rather than a single value. Co-dominant blending inheritance.
- **Height (Mid-Parental Band)** — The prediction engine reads no genotype for height. Each parent picks a height band; we average the two and report a range rather than a number — boys average +3 to +7 cm, girls average −3 to +1 cm. No standard deviation, no single figure: height is shaped by hundreds of variants plus nutrition and environment, so a band is the honest output.
- **Grandparent Evidence** — Grandparent data constrains genotype possibilities. A blue-eyed grandparent proves the parent carries the recessive allele — and each extra generation you enter raises the completeness score behind the confidence figure.

## Confidence Scoring

More family data means higher prediction confidence

| Data Available | Confidence | Explanation |
| --- | --- | --- |
| Parents only | 60-70% | Basic Punnett square, no carrier knowledge |
| Parents + 2 grandparents | 90-95% | Partial genotype constraint |
| Parents + 4 grandparents | 95% (capped) | Strong genotype constraint |
| Parents + 4 grandparents + childhood hair color | 95% (capped) | Every ladder input present — the cap binds. |

## Prediction Modes

Three modes offering different levels of data input and confidence

| Mode | Credits | Data Required | Generations |
| --- | --- | --- | --- |
| Scientific | 4 | Parent traits entered manually (no photos) | 2 (parents → baby) |
| Standard | 8 | Parent traits entered manually (no grandparents) | 2 (parents → baby) |
| Premium | 12 | Full parent traits + all 4 grandparents | 3 (grandparents included) |

Confidence is not set by the mode. It is one server-side score that starts at 60% and rises with the data you give us, hard-capped at 95%.

## Scientific Validation

Built on peer-reviewed genetic research

- **Peer-Reviewed Research** — Based on Sturm et al. (2008) and Eiberg et al. (2008) for the HERC2/OCA2 eye-color model, and Lamason et al. (2005) for SLC24A5 pigmentation. Height is not modeled from genotypes at all — it's a mid-parental band from the parents' own heights, and we say so rather than cite a GWAS we don't use.
- **Model Accuracy** — Where the numbers come from: eye color is solved as two Punnett squares (HERC2 × OCA2); skin tone as a three-gene additive melanin index reported with 68% and 95% bands; height as a mid-parental band from the parents' own heights, not from genotypes. Every probability is computed in code before the AI writes a word — and the overall confidence score is hard-capped at 95%, because genetics is never certain.
- **Full Transparency** — Every prediction includes confidence scores and Punnett square visualizations so users understand the science behind the results.

## FAQ

**How accurate are the genetic predictions?**

Confidence isn't a mode setting — one score starts at 60% and rises with the data you give us, hard-capped at 95%. Parents only sits at 60%, reaching about 70% with ethnicity and both photos; all four grandparents plus childhood hair color reaches the cap.

**Why do grandparents improve accuracy?**

If a parent has brown eyes, they could be BB (cannot pass blue) or Bb (can pass blue). A blue-eyed grandparent settles it: the parent must be Bb. That certainty is what raises your confidence score — any grandparent data adds 20 points and each individual grandparent another 5, up to the 95% cap.

**What is the two-gene eye color model?**

OCA2 sets how much melanin the iris makes; HERC2 switches it on or off. We solve each as its own 2×2 Punnett square and combine them, which is why we can return amber, hazel, green and gray rather than just brown or blue.

**How does skin tone prediction work?**

Skin tone is polygenic. We model three of the best-characterised pigmentation genes — SLC24A5, SLC45A2 and TYR — as a weighted additive score on a 0–100 melanin index, and report it with 68% and 95% confidence bands rather than one number. Unlike eye color, skin tone follows co-dominant blending inheritance.

**How accurate is height prediction?**

We report a range, not a point estimate. Each parent picks a height band rather than typing a measurement; the engine averages the two and adds a sex-specific offset — boys +3 to +7 cm, girls −3 to +1 cm. There's no standard deviation and no variance-explained figure behind it. Height is shaped by hundreds of genetic variants plus nutrition and environment, so treat the band as a rough guide, not a measurement. (The app's separate free Height tool uses the classic mid-parental ±6.5 cm formula — that's a different calculator, not the prediction engine.)

**Is this medical advice?**

No. ProbaBaby uses simplified genetic models for educational and entertainment purposes. Predictions are probabilities, not guarantees. For diagnostic genetic testing, consult a certified genetic counselor.

---

ProbaBaby (listed on the App Store as “Pregnancy & Baby Tracker - PB”) is a free iPhone, iPad and Apple Watch app for pregnancy, newborn and postpartum tracking, built for twins, in 32 in-app languages, with an optional AI baby-face prediction that computes Mendelian genetics (Punnett squares) from the traits you enter before any image is generated. Tracking, tools, PDF reports and family sharing are free; only AI predictions use credits — 5 starter credits cover your first Scientific prediction — and no subscription is required. Requires iOS 26 or later. Not available on Android.