EGGS: Empirical Genotype Generalizer for Samples Authors: T Quinn Smith, Amatur Rahman, Zachary A Szpiech Published: 2026 ABSTRACT AbstractMotivationWe introduce Empirical Genotype Generalizer for Samples (EGGS) which accepts empirical genotypes with missing data and replicates the distribution of missing genotypes along the empirical segment in other replicates. The empirical segment must have a number of sites less than the replicate. In addition, EGGS can remove phase, remove polarization, simulate deamination, simulate sequencing error, create pseudohaploids, and convert between Variant Call Format (VCF), ms-style replicates, and EIGENSTRAT/ANCESTRYMAP. When producing VCF files, EGGS is not limited to biallelic sites and assumes all samples are diploid.AvailabilityEGGS is written in the C programming language. Precompiled executables, source code, the manual, and the analysis conducted in the paper are available at https://github.com/TQ-Smith/EGGS 1 Introduction Recent advantages in both forward and backward in-time simulations have made it possible to generate synthetic genotypes under varying evolutionary assumptions and complicated demographic scenarios (Kelleher et al. 2016, Kern and Schrider 2016, Lauterbur et al. 2023, Haller et al. 2026). Such simulations are essential for rigorously testing computational methods, and more recently, training machine learning models (Schrider and Kern 2018, Flagel et al. 2019, Amin et al. 2024, Rahman et al. 2025). In addition, simulations are often used to test hypotheses concerning the evolutionary histories of human and nonhuman species (Amin et al. 2024, Rayner et al. 2024). Simulated samples are reported under idyllic conditions where each genotype is phased, and in the case of biallelic sites, each site has a known ancestral allele. Simulated genotypes are devoid of technical artifacts that introduce variant uncertainty, such as missing alleles and deamination. Empirical data are often error-prone and violate the assumptions of simulated data. Ignoring this discrepancy in data quality can potentially lead to erroneous inference and unreliable results. Robust inference in the presence of low-quality genotypes is of particular interest when studying ancient DNA (aDNA) (Hofreiter et al. 2001, Williams and Huber 2025). This has motivated the development of methods to account for genotype uncertainty (Achaz 2008, Hellmann et al. 2008, Johnson and Slatkin 2008, Korneliussen et al. 2014, Korunes and Samuk 2021, Bailey et al. 2025). Missing alleles are caused by compounding sources, including sample quality, sequencing technology, and data processing. The inherent stochasticity and complexity of such error are randomly introduced into simulated data (Pandey et al. 2024). This approach assumes that the proportion of missing samples at each site viewed in aggregate reflects the underlying structure of missing genotypes. However, a cumulative summary of missing sites for each sample ignores stretches of missing genotypes across the empirical segment, especially in regions of low complexity. In addition, the random modelling of missing data using a predefined distribution may inaccurately capture the empirical distribution. These difficulties are often exacerbated in non-model organisms lacking a high-quality reference genome (Mora-Márquez et al. 2025). Several recent studies introduce missingness blocks that correspond to regions of low mappability and variant callability (Skov et al. 2018, Arnab et al. 2025). Other current methods that introduce missing genotypes require a FASTA file (Kern and Schrider 2018), ignore the position of the missing sites within the simulated segment, are restricted to specific simulation frameworks, simulate error prone reads directly (Renaud et al. 2017), or are not generalizable to any supplied Variant Call Format (VCF) file (Goulart and Samuk 2026). We introduce Empirical Genotype Generalizer for Samples (EGGS) which extracts the underlying distribution and tracks of missing sites across a genomic region and introduces similar patterns of missing genotypes in replicates. To avoid confusion, we define a missing genotype at a locus where a sample is missing both alleles. In a VCF, this is denoted with ‘./.’. EGGS works on replicates with a number of sites less than the empirical segment and any number of samples. We demonstrate the use of EGGS and illustrate its ability to capture realistic patterns of missing genotypes for a simulated set of samples. 2 Methods 2.1 Replicating missing sites Consider S samples genotyped at N sites. Our goal is to introduce the structure of missing genotypes in a synthetic replicate. Let T be the number of samples in the synthetic replicate, where the T samples are genotyped at M