Picture a wind map over a mountain range. The wind mostly blows downhill, following elevation changes. But if you look closely, in some valleys the air swirls into little eddies that keep the wind from simply flowing down. That is exactly what the mutation field in human DNA looks like.
It turns out that most of this field can be explained by a simple rule: mutations go where the DNA sequence is more stable. This is like water flowing into lowlands. The scientists trained a neural network to predict only such "slopes" and found that it reproduces the genome's composition and the famous Chargaff symmetry, where the frequency of a word equals the frequency of its mirror image.
But a small part of the field remains that cannot be reduced to slopes. These are eddies where mutations swirl into cycles: for example, letter A changes to G, G to T, and T back to A. Such cycles break detailed balance, where each step is balanced by its reverse. The eddies are weak but reproducible: they are visible in African, Asian, and European populations.
The mutation wind map shows that the human genome is nearly at equilibrium: the main flow goes downhill, like water from a mountain. But in CpG valleys, weak eddies remain, reminding us that even in a calm landscape there are irreversible currents. These eddies are traces of molecular processes that can be measured and used to analyze mutations in cancer or when comparing species.
Chargaff's symmetry, which the model reproduces, was discovered by Erwin Chargaff in the 1950s and became a key clue for Watson and Crick in solving the structure of DNA.