How to play
Sandbox: a sample from Person A, or a mixture of A and B in the ratio you pick. Change how much DNA went in, how degraded it is and the analytical threshold, and watch peaks shrink, drop out or turn into stutter. The table compares the sample with each reference and the lab gives its verdicts and the statistic. Match the sample: a sample and four references. Look through all five dyes, give each reference a verdict (match, excluded or inconclusive) and check against the lab.
The science
STRs. At an STR locus a short piece of DNA (like GATA) repeats a different number of times in different people. Everyone has two copies, one from each parent, so a person is, say, 13,16 at D8S1179. The US CODIS database uses 20 core loci; together they make a profile that, for unrelated people, is far rarer than one in the number of people who have ever lived.
The electropherogram. The lab copies each locus with PCR using a primer tagged with a coloured dye, then runs the copies through a thin capillary that sorts them by length. Each allele is a peak: its position is its length in base pairs, its colour tells which group of loci it belongs to, and its height (RFU) roughly follows how much DNA there was. Amelogenin shows X, or X and Y.
Reading it. Peaks under the analytical threshold are treated as noise. PCR slips now and then and makes a small stutter peak one repeat shorter than a real allele; peaks in that spot under 15% of their parent are filtered out (marked st). A degraded sample is broken into short pieces, so the longer loci fade first and some alleles or whole loci drop out. A single peak where the reference has two could be drop-out, so that locus is inconclusive rather than an exclusion.
The statistic. The random match probability is the chance that an unrelated person picked at random has the same profile. At each locus the genotype frequency is 2pq for two different alleles and p² (plus a small correction, θ = 0.01, from the US National Research Council's 1996 report) for two the same; multiplying across loci is the product rule. A mixture uses the combined probability of inclusion instead: at each locus, the square of the summed frequencies of every allele seen.
Relatives. Brothers and sisters get their alleles from the same two parents, so they share far more than strangers do. A partial profile with only a few loci can fit a sibling as well as the donor, which is why the statistic is quoted for unrelated people and why investigators look for other evidence.
How accurate is it?
The 20 loci, their repeat lengths and Amelogenin are real, and the allele frequencies are rounded from published US population data. Sizes are spaced out for reading rather than copied from a commercial kit, peaks are drawn as clean curves, and heights, stutter and degradation follow simple rules with realistic numbers. Real labs also use stochastic thresholds, probabilistic genotyping software for mixtures and separate tables for each population.
Is anything saved? Only how many samples you have matched, on this device.