G25 proximity heatmap

Every reference population in the region you pick is scored against your row by Euclidean distance across all 25 Global25 dimensions, the same arithmetic the distance calculator uses. Each population then sits on the map at a curated point, the place its samples are drawn from. Between those points the map interpolates: every spot of land takes a weighted average of the distances around it, with nearer populations weighing far more than distant ones. The locality control sets how steeply that weight falls off and the smoothing control sets how wide a footprint each population has, so a dense cluster of samples reads as one region rather than as a dozen pinpoints.

The colour scale runs from the closest end of the ramp at the minimum you set to the farthest end at the maximum. Auto-scale fits that range to the 5th and 95th percentile of the populations actually on the map, which is what you want when one region is uniformly close or uniformly far. Land with no reference population within the coverage radius fades toward the farthest colour rather than inheriting the average of whatever is nearest, because an empty stretch of map is honestly unknown, not honestly distant.

What the picture is not. Interpolation is not sampling: a colour between two populations is an arithmetic blend of their distances, not a measurement of anyone who lives there, and the land clip draws the field to modern coastlines and borders that the samples know nothing about. A population's point is one curated location for a whole average, and an average can sit close to your row because you share ancestry with it or because it is itself a mixture that happens to land near your coordinates. Read the map as a neighbourhood, never as a route, and compare regions and eras against each other rather than treating any one frame as the answer.

Everything runs in your browser: the row you paste is not uploaded and nothing is stored. The reference set is a community-compiled collection of Global25 population averages built from public ancient and modern DNA studies, including the Moriopoulos collection and Keltyse's ancient averages; some entries are modelled estimates rather than genotyped samples, and the diaspora filter is a curated name list that a mislabelled population can slip through. Land geometry is Natural Earth.

Suppose a row from the southern Balkans is plotted over Europe. The closest readout names a population from the same broad region at a distance near 0.02, the farthest a Volga or Saami average past 0.15, and the map between them shows the warm end of the ramp pooled over the Balkans and Italy, cooling through central Europe and reaching its far colour in the north-east. That gradient is the finding. The exact population that wins by a few thousandths is not: reference entries are averages of individuals, and two averages a hair apart are not distinguishable by a distance.

Now change one control. Set locality to Broad and the pools blur into one another because distant populations keep a say at every pixel; set it to Tight and each population claims only its own surroundings, so the map shows islands of colour with sharper edges. Neither is more true. Broad answers where you sit across a continent, Tight answers which specific neighbours you resemble, and the same row should be read both ways before either reading is trusted.

The cross-frame read is where the tool earns its place. Plot the same row over the Bronze Age and Iron Age eras and watch whether the warm pool moves. A modern row whose nearest ancient populations cluster in the same geography as its nearest modern ones tells a coherent story; a pool that jumps continents between adjacent eras usually indicts the paste, an unscaled row or the expectation, not the ancestry. Ancient distances run systematically wider than modern ones, so compare eras with each other, never with the modern numbers.

A ranked distance list, this tool's own list included, answers the same question in a different shape: which averages are nearest, in order. The list is exact and the map is not, because the map adds an interpolation step between the points. What the map adds is geography, which the list cannot show: two populations at rank 3 and rank 4 may be neighbours or a thousand kilometres apart, and only the map says which. Use the list to read the numbers and the map to read the shape.

An admixture calculator answers a different question altogether. It models your row as a mixture of chosen sources and reports proportions; this tool fits nothing and reports resemblance. A population can colour warm on the map without belonging in your ancestry at all, and a source that a calculator needs can sit far from you on the map. Closeness is resemblance, not descent, and no distance-based picture can test whether a population belongs in a model. That is the admixture calculator's job, and the tested version of it is qpAdm.

The limit every heatmap shares, this one included: it draws a smooth surface over sparse points. Where the reference set is dense, in Europe and South Asia, the surface is well constrained; where it is sparse, in much of Africa, Siberia and the Americas, wide areas take their colour from a handful of averages and the fade to the farthest colour is doing most of the work. The number of populations plotted is shown beside every map for that reason.