#27 · Genomics

New software speeds demographic inference, haplotype simulation, and ancestry analysis

The dominant thread is re-engineering population-genetic and genomic analysis as faster, more scalable software. momi3 adds GPU execution, autodiff, and pruning to demographic inference from large allele frequency spectra, reporting speedups up to 1000x. FORGEPHAST provides a Python framework for reference-panel haplotype and phenotype simulation with admixture, QC, and Zarr-based storage at biobank scale. ORIAN offers both reference-based, neural-network plus HMM, and iterative reference-free local ancestry inference. ARGformer differs in kind: it applies transformers to self-supervised embeddings of ancestral recombination graphs for population structure clustering and ancestry retrieval. Across the set, the new contribution is mainly the computational layer of speed, scalability, and storage, with ARGformer adding a learned representation of genealogical structure.