Previous Seminars
Lin Poyraz
Columbia University
A sibling study of variation in parental mutation rates
November 4, 2026
People are born with variable numbers of de novo germline mutations (DNMs), depending primarily on parental ages. To investigate additional sources of variation, this work develops an approach to identify DNMs from nucleotide differences between siblings in genomic regions inherited identical by descent from both parents. Applied to whole-genome sequencing data from 28,985 sibling pairs in the UK Biobank and All of Us, together with 2,330 trios, the study identifies more than 800,000 autosomal DNMs and characterizes mutation phenotypes across 27,645 sets of parents. The results show subtle shifts in mutation spectra but no differences in total DNM rates among genetic ancestry groups or between smokers and non-smokers. Testing parental mutation phenotypes against rare damaging variation in DNA repair and maintenance genes identifies REV1 and LIG1 disruptions that increase germline mutation rates, providing evidence that rare mutator alleles segregate in human populations.
Jiseon Min
University of Oregon
Neural posterior estimation for population genetics
October 7, 2026
Simulation-based inference is attractive in population genetics because it can be used when likelihoods are difficult or impossible to evaluate, but traditional approaches such as approximate Bayesian computation can be computationally expensive and struggle with high-dimensional data. Neural posterior estimation trains neural networks on simulated genetic data to directly approximate posterior distributions over population-genetic parameters. The work evaluates neural posterior estimation across several inference tasks, compares performance using raw genotypes and summary statistics, and applies the framework to demographic inference, including analyses of Drosophila melanogaster. The resulting workflow combines flexible simulation-based modeling with calibrated Bayesian uncertainty estimates.
Hao Shen
University of Chicago
Fine-Resolution Asymmetric Migration Estimation
September 23, 2026
The genetic structure of populations is often shaped by asymmetric gene flow among geographic locations. FRAME (Fine-Resolution Asymmetric Migration Estimation) uses an efficient algorithm for computing pairwise coalescent times in structured populations as the basis for inferring asymmetric migration rates across spatial models. The method reduces the computational burden of modeling directional migration and enables fine-resolution analyses with many demes. Applications to simulated data and empirical datasets from poplar trees, North American gray wolves, and human archaeogenetic samples reveal complex asymmetric migration patterns and provide a more detailed view of the geographic structure of genetic variation.
Alexander Mackintosh
Swedish Museum of Natural History; Uppsala University
Reconstructing the history of hybrid species from genome-wide genealogies
September 9, 2026
Several recent studies have found evidence for ancient gene flow between the ancestors of modern humans and an unsampled super-archaic lineage. This work presents a simple approach for characterising this process using genome-wide genealogies sampled from a single population. Genealogies are summarised as distributions of integer partitions, which capture the temporal signal of tree imbalance left by ancient gene flow. Applied to modern human genealogies, the method finds patterns inconsistent with panmixia but explainable by gene flow from a super-archaic lineage, favouring continuous gene flow over pulse admixture and also recovering a bottleneck in the ancestors of modern humans. The approach provides a complementary way to infer ancient population structure from reconstructed genealogies.
Yulin Zhang
UC Berkeley
Recovering signatures of archaic hominin introgression using ancestral recombination graphs
July 1, 2026
Admixture between modern humans and extinct hominins has shaped the genomes of present-day individuals, but reconstructing this history has been constrained by the scarcity of archaic samples and unadmixed outgroup populations. We introduce TRACE, a reference- and outgroup-free approach that uses features of ancestral recombination graphs to identify archaic ancestry. Simulations show TRACE has high precision and low false discovery rates. Applied to 1000 Genomes, TRACE recovers known Neanderthal and Denisovan introgression and uncovers ghost admixture from uncharacterized hominins in both Africans and non-Africans. Ghost ancestry persists in Neanderthal and Denisovan ancestry deserts, challenging their interpretation as Homo sapiens–specific regions. In Oceanians, TRACE finds deep lineages are enriched in Denisovan compared to Neanderthal regions, supporting super-archaic introgression. TRACE enables mapping archaic introgression without archaic genomes.
Nathiel Pope
University of Oregon
Tracing the evolutionary histories of ultra-rare variants using variational dating of large ancestral recombination graphs
May 20, 2026
Ultra-rare variants dominate whole-genome sequencing datasets, yet their interpretation is limited by allele frequency, which provides little information at very low counts and is highly sensitive to uneven ancestry representation. Allele age offers an ancestry-agnostic alternative but existing methods do not scale to biobank-sized cohorts. Here we present a scalable variational algorithm for dating Ancestral Recombination Graphs (ARGs), implemented in tsdate, together with new distributed methods enabling practical biobank-scale ARG inference using tsinfer. Applied to 47,535 genomes from the Genomics England 100,000 Genomes Project, we infer contiguous ARGs spanning 206 Mb and estimate ages for 23.2 million variants, including 11.8 million singletons. ARG-based allele ages remain accurate under extreme sampling imbalance and, in real data, reveal signatures of purifying selection and clinically relevant heterogeneity among variants with identical observed frequencies. Estimates for recent mutations are precise only at large sample sizes, highlighting the information accessible in the haplotype structure of large datasets. Biobank-scale ARGs therefore enable robust, ancestry-agnostic age estimation for ultra-rare variation with broad utility for statistical and clinical genomics.
Stephen Staklinski
Cold Spring Harbor Laboratory
Variational Inference with Node Embeddings (VINE) for Scalable Bayesian Phylogenetics
May 6, 2026
Bayesian methods are now widely used in reconstructing both species and cell-lineage phylogenies, but they remain heavily reliant on computationally intensive Markov chain Monte Carlo sampling. Phylogenetic variational inference (VI) circumvents this dependency but so far has been limited in speed and scalability. Here we introduce Variational Inference with Node Embeddings (Vine), a computational method that combines an embedding of taxa in a high-dimensional space and a distance-based “decoder” with several algorithmic innovations to dramatically improve phylogenetic VI. Vine supports both standard DNA substitution models and CRISPR barcode-mutation models for inference of cell-lineage trees and tissue-migration histories. In extensive simulation experiments, we show that Vine can effectively approximate the results of the best available Bayesian methods with speeds orders of magnitude faster. We then apply Vine to ∼1,000 complete SARS-CoV-2 genomes and ∼900 lung-cancer cell barcodes, showing reductions in compute time from days to hours or minutes.
Relatedness between individuals can be measured at the genealogical level (by describing shared ancestors in a pedigree) or at a genetic level (by describing shared haplotypes across the genome). The shared haplotype structure can be conveniently summarized as a sequence of trees along the genome, where each tree describes the last shared genetic ancestors between individuals at a locus. While many tools exist to infer tree sequences from genetic data, and many large pedigree datasets are available, few tools exist to identify the relationship between the two – finding which genetic ancestor corresponds to which pedigree ancestor, and conversely. In this paper, we propose an algorithm to solve this problem by providing, for each genetic tree, the list of all consistent ancestry paths within a genealogical tree. We also provide variants of the algorithm with moderate robustness to both tree sequence and pedigree errors. We demonstrate the scalability of our approach on the BALSAC genealogical dataset, which includes millions of individuals in Quebec, Canada. We find that 20 carriers usually provide enough information to reliably identify a common ancestor 15 generations ago or find the parent of origin of alleles among probands, but the number of possible ancestry paths within the pedigree can remain large. We apply the method to reconstruct the inheritance of a causal allele for type 1 myotonic dystrophy.
Sara Mathieson
University of Pennsylvania
Interpreting Convolutional Neural Networks in Population Genetics
April 8, 2026
Machine learning approaches have become a powerful alternative to traditional methods in population genetics. Convolutional neural networks (CNNs) in particular have been successful in inferring natural selection, recombination rate estimation, introgression, dispersal distances, and effective population size changes. One limitation of CNNs and other deep learning methods is that they can be difficult to interpret. When they have been shown to be as or more successful than summary-statistic-based methods, what are they learning? Here we investigate CNNs from two different methods: the pg-gan discriminator for identifying real vs. simulated data, and two networks trained to detect selective sweeps. We first compute correlations between learned network features and traditional summary statistics, then assess whether summary statistics can be predicted from the learned features. To understand the learned features, we compute feature importance through SHAP values and feature groupings through dimensionality reduction. Finally we use decision trees and random forests to build an interpretable “model-of-the-model”. Our results reveal that some CNN architectures can implicitly compute summary statistics such as pairwise heterozygosity, while statistics such as the site frequency spectrum are less similar to the network’s learned features. We find that long-range linkage disequilibrium is readily approximated by the networks and may be more efficiently computed by CNNs than traditional methods (which are quadratic in the number of sites). Overall, this work contributes to the interpretability of deep learning methods in population genetics by clarifying the relationships between model architecture, learned network features, established summary statistics, and predicted evolutionary parameters.
Hanbin Lee
University of Michigan
Genetic prediction with ARG-powered linear algebra
March 11, 2026
Ancestral recombination graphs (ARGs) are an attractive means for quantitative genetic analysis of complex traits because they encode the realized genetic relatedness between a sample of individuals in the presence of genetic drift, recombination, and mutation. Data structures for efficiently storing ARGs can also be used to rapidly process millions of genomes, and are thus promising for fitting linear mixed models to large phenotype and genome datasets. Here, we study the problems of variance component estimation and prediction of genetic values with ARGs, by describing a generative model of complex traits with additive effects on an ARG, and then developing algorithms that use the ARG to solve these problems efficiently on biobank-scale datasets. We observe nearly linear scaling of runtime with sample size, which is achieved by using the succinct tree sequence representation of the ARG for implicit matrix-vector products, along with modern randomized linear algebra algorithms. We estimate variance components using restricted maximum likelihood, which we find performs substantially better than the Haseman–Elston method. In simulation tests, both variance component estimation and prediction of genetic values (using the best linear unbiased predictor) perform nearly as well with inferred ARGs as with true ARGs. We also discuss interpretations of the variance component estimates as mutational variance and additive genetic variance. We provide an implementation of the algorithms as a Python package tslmm, which leverages the tree sequence library tskit.
Roshni Patel
University of Oregon
Observational epidemiological studies can mitigate genetic confounding with the genetic relatedness matrix
February 25, 2026
Observational studies are commonly used in psychology and epidemiology to identify risk factors correlated with health outcomes. However, these studies are vulnerable to confounding when shared genetic variation influences both the putative risk factor and outcome. Researchers have often controlled for this type of genetic confounding using polygenic scores, but these scores are noisy and biased estimators of a trait’s genetic component. While some newer methods offer significant improvements over polygenic scores, they still rely on genome-wide association studies (GWAS) summary statistics, which may be untenable for certain datasets. Here, we develop an analogous method that leverages a genetic relatedness matrix to control genetic confounding when testing for nongenetic risk factors. In simulations, we find that our method outperforms existing approaches, particularly at sample sizes that are large by the standards of much human research but smaller than datasets often used in human genetics. We also demonstrate that existing methods are susceptible to poor GWAS portability, whereas our method is inherently robust to such concerns, conditional on the availability of individual genotype data. Finally, we apply our method to the UK Biobank to reanalyze social risk factors for health outcomes in previously understudied cohorts.