The third Genomic History Inference Strategies Tournament is live, and registration is open through November 20 at the Synapse site. Each year we release simulated population genomic data sets and ask the community to infer what produced them, so that we learn which approaches work well in which circumstances rather than relying on developers to grade their own methods. This year's challenges span demographic history, selective sweep detection, and genealogical relationship inference. Several add complications that real data present and most benchmarks avoid: background selection in both the demographic and sweep challenges, low-coverage data, and complex pedigrees. Any approach is welcome. If you are new to this kind of inference, we have built web apps for bottleneck inference, sweep detection, and relatedness that produce files you can submit directly, and past competitions have proven to be good training for students. Top competitors are offered co-authorship on the paper describing each year's results.

Ancient DNA and new inference methods have transformed the study of human origins, but consensus has not followed. In a new review, I argue that the evidence for pervasive structure and admixture in hominin history makes complexity, not simplicity, the appropriate prior, and I identify two blind spots that impede resolving it. First, every inference passes through summaries of the data, and those summaries bound what can be recovered. Second, the space of candidate models is vast, and we lack principled ways to explore it. That is a limit of practice rather than of data. Because omitted processes bias the ones we do estimate, and because competing model classes are rarely fit to common data, a reported best model carries little evidence about the models nobody tried, though withholding a summary before fitting and fitting competing classes to common data both help. The field needs to study the model exploration process itself.

Does the immune system actually shape which mutations rise to high frequency in a growing tumor? Evidence from human cancers has been largely correlational, in part because the alternatives are hard to rule out. In a new paper in Immunity, led by Elizabeth Borden and Karen Hastings, cutaneous squamous cell carcinomas from immunocompetent and immunosuppressed patients were compared to isolate immune-mediated effects. Our group contributed the two null expectations the comparison needs. David Castellano built a mutation model that predicts each patient's expected neoantigen profile from the trinucleotide mutation spectrum and gene-level mutability, so that apparent depletion of MHC-binding neoantigens can be tested against what mutation alone would produce. He also treated the variant allele frequency spectrum as a noisy site frequency spectrum and used approximate Bayesian computation to estimate tumor growth rate and the ratio of clonal to subclonal mutations, separating immune selection from the excess of low-frequency variants that fast growth produces on its own. Growth rates did not differ; the clonal fraction did. The immune system is not eliminating binding neoantigens from these tumors. It is keeping them from becoming clonal.

Congratulations to Sam Gibbon, named the College of Science Outstanding Senior in Mathematics for spring 2026 and awarded a Simons Graduate Fellowship in Ecology and Evolution. Sam came to us as a mathematics major looking for problems where analysis and computation meet biology, and he found them in polyploid population genetics, working with us and with Michael Barker and Justin Conover. That work became his honors thesis on inference methods for auto- and allopolyploids and a preprint on bistable mutation-selection equilibria in tetraploids. The Simons fellowship supports students bringing mathematics, physics, and computer science into evolutionary biology, which is Sam's path in a sentence, and it provides three years of support as he begins his PhD in Human Genetics at the University of Chicago this fall.

Our previous work on joint DFEs between populations only considered deleterious mutations, but adaptive mutations may be important in many cases, including domestication. In our just published collaborative paper we develop a joint DFE model that includes differential positive selection between populations and apply it to simulations designed to mimic the domestication process. The most interesting class of mutations here are those that change sign from deleterious before domestication to adaptive afterward, and we find generally good power to identify that component of the DFE if adaptation is driven by alleles of moderate effect. On the other hand, if adaptation is driven by a few alleles of strong effect, hitchhiking causes strong biases in our estimates, and those of other DFE inference tools.