DNA evolution can be viewed as a random walk over a vast discrete space of sequences. The question is whether the observed directionality of mutations reduces to an effective energy landscape, as in an equilibrium process with detailed balance, or contains an irreversible circulation characteristic of nonequilibrium systems. Context-dependent human mutation probabilities provide a direct test of this idea. The work stands apart: nearby in the archive lie studies of evolution, horizontal gene transfer, and phylogenetic networks, but there are almost no direct topical neighbors. The closest works touch on evolution and cellular processes but do not dissect the mutation field as a thermodynamic system.
The authors took probabilities of single-nucleotide substitutions in seven-letter contexts from population data of African, Asian, and European descent. For each pair of substitutions, they computed the log ratio of forward to backward probabilities, obtaining an antisymmetric field. They then trained a neural network with a siamese architecture that predicted only energy differences, that is, the gradient part of the field. The remainder, which cannot be represented as a gradient, was isolated using the Hodge projection and analysis of cycle affinities. They also checked how well the predicted energies reproduce the frequencies of short words in the genome and Chargaff's symmetry.
The siamese model predicted log ratios for mutations not seen in training with a correlation of about 0.93. This is close to the unconstrained benchmark (0.948) and to the empirical reversibility ceiling obtained by the Hodge projection (about 0.96). The energy landscape, trained only on mutation probabilities, reproduced the short-word composition of the genome and Chargaff's symmetry for sequences up to four letters long. Deviations from equilibrium were small but reproducible: cycle affinities had a standard deviation of 0.773, and the median absolute affinity was 0.451. The average vorticity fraction in the field by the Hodge projection is 0.11. The irreversible circulation is strongest in CpG-related cycles and edges of CpG transitions, consistent with methylcytosine deamination.
The work shows that the human mutation field is nearly at equilibrium but contains a measurable irreversible component. This sets it apart from neighboring studies that consider evolution or cellular processes without a thermodynamic decomposition of mutation probabilities. Here, for the first time, detailed balance is directly tested on empirical human mutation data. The resulting energy landscape links mutation directionality to genome composition, and the residual circulation points to specific biochemical mechanisms such as cytosine methylation.
The topic could develop toward stratifying data by replication timing, chromatin state, methylation, transcription asymmetry, and repair pathways. This would link the irreversible component to specific molecular processes. The method could also be applied to de novo mutations, cancer signatures, and cross-species comparisons.
The results will influence evolutionary biology, genomics, and the statistical physics of nonequilibrium systems. They provide a tool for separating the equilibrium background from directed mutational mechanisms.
The next step is to analyze mutation data accounting for epigenomic factors to uncover the origin of the irreversible circulation. It is also worth checking whether the CpG enrichment persists in other species.
The work connects observed mutational biases to the fundamental problem of detailed balance in biological systems. It shows that even without exact equilibrium, one can isolate a dominant reversible part and a small but structured irreversible component. This contributes to understanding how random mutations shape genome composition.
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The authors represented the probabilities of single-nucleotide substitutions in seven-letter contexts of the human DNA as a directed field and decomposed it using a neural network and the Hodge projection into a gradient part (an energy landscape) and a curl part. The resulting landscape reproduces the composition of short genomic words and the Chargaff symmetry, while a small irreversible circulation, reproducible across populations, is most pronounced in CpG-related cycles.
Where its strength is
A strong point is the direct test of detailed balance against empirical human mutation data, yielding a measurable irreversible component tied to a specific biochemical mechanism (deamination of methylcytosine).
Where it could go next
Worth trying
Why this may matter
If the mutation field is indeed nearly equilibrium but contains a universal irreversible component, this could bridge molecular mechanisms of mutagenesis and large-scale evolutionary patterns, and also provide a tool for estimating the contribution of individual biochemical processes to shaping genomic composition.
Nearby works in our archive
The archive is sparse around this work: the closest review is only 0.54, against a usual 0.64. It looks like few have drilled nearby.
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.