What averaging does to G25 coordinates
A set of G25 coordinates is 25 numbers, one per principal component. Averaging works on each of the 25 columns separately: the first value of the result is the average of the first values of the inputs, and so on to the twenty fifth. Most population entries in a G25 sheet are built exactly this way, as the average of the individual samples in that group.
A weighted average sets the share of each input. Because the components are linear projections of the genome, a population formed by mixing two sources sits close to the weighted average of their coordinates. That is the reason a simulated sample is useful, and the same arithmetic an admixture calculator runs in reverse when it searches for the mix of sources that lands closest to you.
What simulated coordinates are good for
- Checking family files. The midpoint of two parents should sit close to their child. A large gap points to a problem with one of the files, or a wrong label.
- Standing in for someone untested. A grandparent who never took a test can be approximated from the relatives who did, then modelled like any other sample.
- Testing a historical scenario. Mix two ancient populations in the proportions a paper proposes and see whether the result lands near the later population it is meant to explain.
- Building a better source. When one group is thinly sampled, an average of close neighbours can make a steadier reference than a single sample.
Where averaged coordinates mislead
- An average hides spread. A population average is a centre point. Real members of the group can sit well away from it, and a person who matches the average closely is not more typical of the group than one who does not.
- Distant inputs make an empty point. Averaging two very different populations gives coordinates that may match no population that ever lived. That is fine as a model of first generation mixing, not as evidence that such a group existed.
- Subtraction amplifies noise. Taking one sample away from another, as Parent Inference does, magnifies the error in the inputs: any noise in the child's file counts twice. Use it with good files and treat the result as approximate.
- Simulated is not measured. A simulated sample carries every assumption that built it. When a model fits it well, that tells you about the assumptions, not about a real genome.
To judge any fit made with simulated coordinates, read the distance as you would for a real sample: the guide on how accurate an admixture calculator is covers what a good fit distance looks like.
Averaging and simulating in MYDNATLAS
Ghost Builder builds averaged and simulated G25 coordinates in three modes. It is part of the paid plans.
Family Wizard
Pick parents, grandparents and great-grandparents, and each generation is weighted 50%, 25% and 12.5% per person. Use it to simulate a relative you never tested or to check that a family's files agree with each other.
Custom Calculator
Add populations, individual samples or a pasted set of 25 coordinates, each with its own weight, and subtract components for de-admixture scenarios. Simplex normalisation makes the added weights sum to 1; Sum to |1| and None are there for advanced formulas. You can name the result and save it.
Parent Inference
Reconstruct an unknown parent from a child and one known parent. In G25 terms the missing parent sits near twice the child minus the known parent, which is why this mode needs good files for both.
Your own coordinates come from your raw DNA file; the raw DNA to G25 guide shows how to download it from each testing company.
Questions
What is a G25 average calculator?
A tool that combines two or more sets of G25 coordinates into one, column by column. A plain average gives every input the same weight; a weighted average lets you set the share of each, for example 75% of one population and 25% of another. The output is a new set of 25 coordinates you can model and compare like any other sample.
How do I average G25 coordinates by hand?
Put each sample on its own row with its 25 values in 25 columns, then take the mean of each column. For a weighted average, multiply each row by its weight, with the weights adding up to 1, and sum each column. Keep the 25 values in their original order, and only combine coordinates of the same kind: scaled with scaled, unscaled with unscaled.
What are simulated G25 coordinates?
Coordinates that were calculated rather than measured from a genome: a weighted average of real samples standing in for a person or population that was never sampled, such as a child of two known parents or a proposed mix of two ancient groups. They describe a hypothesis, not a person.
Can I average my parents' coordinates to check my own?
Yes. A child's coordinates usually sit close to the midpoint of the two parents' coordinates, so averaging the parents and measuring the distance to the child is a useful check on all three files. Expect a small gap: each child inherits a random half of each parent's DNA, and chip coverage differs between kits.
Does MYDNATLAS have a G25 average tool?
Yes. Ghost Builder builds averaged and simulated coordinates in three ways: Family Wizard for relatives, Custom Calculator for any weighted mix of populations, samples or pasted coordinates, and Parent Inference for reconstructing one unknown parent. It is part of the paid plans.