Anthony Francis Herzig

Recently, a number of studies have looked at the problem of privacy and data-sharing restrictions in the context of missing-genotype imputation servers. This relates to the most typical imputation pipelines which involve a whole-genome sequenced haplotype reference panel being compared to genotyped study individuals (who have missing data to be imputed). Hence, involving two datasets from separate sources coming together in one informatic environment, where relatively complicated statistical models are applied: specifically, hidden Markov modelling. We give a short review of the current literature in this domain, observing three prevalent strategies: complicated data encryption, technical solutions to secure computation environments, and rearrangements of haplotype data to provide anonymisation. We embarked on a thought experiment to provide a potential fourth type of solution involving federating the different internal tasks within the statistical methods used for imputation. This idea is relevant considering there is currently motivation for federated analyses platforms in Europe for making combined inference across multiple genomic data resources. Our solution allows for very simple manipulations to protect sensitive individual level data, which enable imputation algorithms to complete on simple plain-text files. We provide here an illustration of how such a federated imputation server could be put in place, along with associated code, including a simple implementation of the Li-Stephens haplotype mosaic model to achieve the imputation of missing genotypes. We name our general framework ANONYMP for anonymised imputation. A demonstration of the concept is given involving simulated data generated with msprime. We show that dividing different parts of the required calculations for statistical imputation between several sites is a valuable new avenue in the field of privacy-preserving imputation server development.

Genetic polymorphisms are common in pharmacogenes, with sometimes important implications for drug metabolism. Assessing the correct enzyme phenotype from genetic data is thus a crucial step into the development of personalized medicine. Many bioinformatics star‐allele callers have been developed for this purpose of identifying the correct star alleles and the associated phenotype, each of them having their specific method and limitations. Despite the important benchmarks that have been made so far, their performances have not yet been fully explored depending on various parameters, such as the type of genetic data provided as input or the individuals' ancestry. Hence, we provide a multi‐gene, multi data‐type comparison of the accuracy of four commonly used and open‐access star‐allele callers: PyPGx, ursaPGx, PharmCAT, and Aldy. We found that PyPGx and Aldy are overall more performant than the others, except for CYP2D6 where ursaPGx was the most accurate with its CYP2D6 dedicated caller that relies on the Cyrius software. Comparing to the commercial solution DRAGEN, PyPGx, and Aldy showed better results, except for CYP2D6 where DRAGEN performed best. When only SNP‐chip or low‐pass sequencing data is available, the use of imputation greatly improves the performance of star‐allele callers, allowing performance comparable to that achieved with sequencing data. We also analyzed how concordance between star‐allele callers varies depending on population ancestry. Our findings offer guidance on the choice of star‐allele caller, depending on the pharmacogene being studied and the resolution of available genetic data.

Etude de la composante génétique de caractères complexes dans une population isolée: Mon projet de thèse vise à exploiter le potentiel des isolats de population pour étudier la composante génétique des maladies multifactorielles. En effet, les isolats peuvent faciliter l'identification des facteurs génétiques habituellement trop rares en population générale. Cette thèse est composée de deux études principalement : l'imputation génétique et l'analyse de l'héritabilité. Chacune de ces études ont été abordée sous deux angles : l’un théorique, s’appuyant sur une vaste étude de simulations basée sur les caractéristiques de la population isolée du Cilento, permettant d’évaluer des stratégies d’analyse et de déterminer la plus adéquate ; l’autre appliqué, s’appuyant sur l’analyse de données génétiques réelles issues de la même population.L'imputation génétique est une étape cruciale pour effectuer des analyses d'association dans un isolat et représente une méthode peu couteuse pour obtenir les séquences complètes du génome ou de l’exome des individus de la population. L'efficacité de cette approche dépend de la précision de l’imputation ; nous avons donc étudié plusieurs stratégies pour obtenir une précision d'imputation maximale dans un isolat. Nous avons montré que les logiciels utilisant des algorithmes qui s’appuient sur les caractéristiques particulières des isolats n’étaient pas, de façon inattendue, aussi performants que ceux conçus pour les populations générales. De plus, malgré la disponibilité de panels de référence publics contenant plusieurs milliers de chromosomes, nous avons confirmé qu’un panel de référence spécifique de la population d’étude, même de taille très réduite, était essentiel pour la qualité de l’imputation. Ceci était d’autant plus vrai pour les variantes rares.Pour de nombreux traits, il existe des discordances entre les estimations de l'héritabilité obtenues à partir d’individus apparentés et à partir d’individus non apparentés. En particulier, la plupart des chercheurs considère que les effets dominants (non additifs) ne jouent pas un rôle majeur malgré les résultats contrastés des études sur les isolats. Notre deuxième analyse a révélé des mécanismes possibles pour expliquer la disparité de ces estimations publiées entre populations isolées et populations générales. Cela nous a permis de faire des déductions intéressantes pour nos propres analyses dans le Cilento. En particulier, nous avons identifié la possibilité d'une composante de dominance non nulle pour les niveaux de lipoprotéines de basse densité (LDL). Cela nous a amenés à effectuer des analyses d'association pan-génomique des composantes additives et non-additives pour LDL dans le Cilento et nous avons pu identifier des gènes qui avaient déjà été liés au trait dans d'autres études.Dans le contexte de nos deux études, nous avons observé l'importance de conserver l'incertitude génotypique (dosage pour l’imputation, vraisemblance des génotypes pour les données de séquençage). Dans la perspective de cette thèse, nous avons proposé des moyens d’incorporer cette incertitude à certaines méthodes utilisées dans ce projet.Nos résultats concernant les stratégies d'imputation et l'analyse de l'héritabilité seront très utiles pour la poursuite de l'étude de l'isolat de Cilento. Mais, ils seront également instructifs pour les chercheurs travaillant sur d'autres populations isolées et également applicables plus généralement à l'étude des maladies complexes.

Large-scale human genomic projects have generated whole-genome sequencing (WGS) data from hundreds of thousands of individuals, primarily to study host genetic variation. When saliva is the DNA source, the resulting datasets also contain microbial reads that are routinely discarded. Here, we investigate whether these host-centric WGS workflows can yield reliable microbiome profiles, effectively doubling the research value of existing data without additional sampling. We compared non-human reads from 39 deeply sequenced saliva samples from the GAZEL cohort (miG dataset; median ~43 million reads/sample) with 14 samples processed with microbiome-optimized extraction (ASAL; median ~4.3 million reads/sample), using two complementary classifiers: meteor, a coverage-based mapper against a curated saliva-specific database, and sylph, a k-mer classifier against the Genome Taxonomy Database (GTDB). Despite the absence of microbial lysis optimization, miG samples showed up to 3-fold higher species richness, ~10-fold greater sequencing depth, and significantly lower inter-sample variability (PERMANOVA R² = 0.10, p = 0.001; BETADISPER p = 0.0036). Rarefaction to 10? reads eliminated most compositional differences, demonstrating that sequencing depth is the primary driver of community stability. Only ~2% of detected taxa (12 of 592) showed extraction-related differences. The two classifiers exhibited fundamentally different depth-sensitivity profiles, with sylph retaining systematic detection asymmetries even after depth normalization, highlighting that classifier choice introduces biases that affect cross-study comparisons. These results show that biobank WGS data from saliva can be repurposed for robust, population-scale oral microbiome analyses, enabling simultaneous investigation of host genomic variation and the microbiome from the same archived samples.

Recently, a number of studies have looked at the problem of privacy and data-sharing restrictions in the context of missing genotype imputation servers. This relates to the most typical imputation pipelines which involve a whole-genome sequenced haplotype reference panel being compared to genotyped study individuals (who have missing data to be imputed). Hence, involving two datasets from separate sources coming together in one informatic environment, where relatively complicated statistical models are applied; specifically, hidden Markov modelling. We give a short review of the current literature in this domain, observing three prevalent strategies: complicated data encryption, technical solutions to secure computation environments, and rearrangements of haplotype data in an effort to provide anonymisation. We embarked on a thought experiment to provide a potential fourth type of solution involving federating the different internal tasks within the statistical methods used for imputation. This idea is relevant considering there is currently motivation for federated analyses platforms in Europe for making combined inference across multiple genomic data resources. This allows for very simple manipulations to protect sensitive individual level data, which enable imputation algorithms to complete on simple plain-text files. We provide here an illustration of how such a federated imputation server could be put in place, along with associated code, including a simple implementation of the Li-Stephens haplotype mosaic model to achieve the imputation of missing genotypes. We name our general framework ANONYMP for anonymised imputation. A demonstration of the concept is given involving simulated data generated with msprime. We show that dividing different parts of the required calculations for statistical imputation between several sites is a valuable new avenue in the field of privacy-preserving imputation server development.

Genotype–phenotype association tests are typically adjusted for population stratification using principal components that are estimated genome-wide. This lacks resolution when analyzing populations with fine structure and/or individuals with fine levels of admixture. This can affect power and precision, and is a particularly relevant consideration when control individuals are recruited using geographic selection criteria. Such is the case in France where we have recently created reference panels of individuals anchored to different geographic regions. To make correct comparisons against case groups, who would likely be gathered from large urban areas, new methods are needed. We present SURFBAT (a surrogate family based association test), which performs an approximation of the transmission-disequilibrium test. Our method hinges on the application of genotype imputation algorithms to match similar haplotypes between the case and control groups. This permits us to approximate local ancestry informed posterior probabilities of un-transmitted parental alleles of each case individual. This is achieved by assuming haplotypes from the imputation panel are well-matched for ancestry with the case individuals. When the first haplotype of an individual from the imputation panel matches that of a case individual, it is assumed that the second haplotype of the same reference individual can be used as a locally ancestry matched control haplotype and to approximately impute un-transmitted parental alleles. SURFBAT provides an association test that is inherently robust to fine-scale population stratification and opens up the possibility of efficiently using large imputation reference panels as control groups for association testing. In contrast to other methods for association testing that incorporate local-ancestry inference, SURFBAT does not require a set of ancestry groups to be defined, nor for local ancestry to be explicitly estimated. We demonstrate the interest of our tool on simulated datasets, as well as on a real-data example for a group of case individuals affected by Brugada syndrome.

The demographical history of France remains largely understudied despite its central role toward understanding modern population structure across Western Europe. Here, by exploring publicly available Europe-wide genotype datasets together with the genomes of 3234 present-day and six newly sequenced medieval individuals from Northern France, we found extensive fine-scale population structure across Brittany and the downstream Loire basin and increased population differentiation between the northern and southern sides of the river Loire, associated with higher proportions of steppe vs. Neolithic-related ancestry. We also found increased allele sharing between individuals from Western Brittany and those associated with the Bell Beaker complex. Our results emphasise the need for investigating local populations to better understand the distribution of rare (putatively deleterious) variants across space and the importance of common genetic legacy in understanding the sharing of disease-related alleles between Brittany and people from western Britain and Ireland.

Imputation servers offer the exclusive possibility to harness the largest public reference panels which have been shown to deliver very high precision in the imputation of European genomes. Many studies have nonetheless stressed the importance of ‘study specific panels’ (SSPs) as an alternative and have shown the benefits of combining public reference panels with SSPs. But such combined approaches are not attainable when using external imputation servers. To investigate how to confront this challenge, we imputed 550 French individuals using either the University of Michigan imputation server with the Haplotype Reference Consortium (HRC) panel or an in-house SSP of 850 whole-genome sequenced French individuals. With approximate geo-localization of both our target and SSP individuals we are able to pinpoint different scenarios where SSP-based imputation would be preferred over server-based imputation or vice-versa. This is achieved by showing to a high degree of resolution the importance of the proximity of the reference panel to target individuals; with a focus on the clear added value of SSPs for estimating haplotype phase and for the imputation of rare variants (minor allele-frequency below 0.01). Such benefits were most evident for individuals from the same geographical regions in France as the SSP individuals. Overall, only 42.3% of all 125,442 variants evaluated were better imputed with an SSP from France compared to an external reference panel, however this rises to 58.1% for individuals from geographic regions well covered by the SSP. By investigating haplotype sharing and population fine-structure in France, we show the importance of including SSP haplotypes for imputation but also that they should ideally be combined with large public panels. In the absence of the unattainable results from a combined panel of the HRC and our French SSP, we put forward a pragmatic solution where server-based and SSP-based imputation outcomes can be combined based on comparing posterior genotype probabilities. We show that such an approach can give a level of imputation accuracy in excess of what could be achieved with either strategy alone. The results presented provide detailed insights into the accuracy of imputation that should be expected from different strategies for European populations.

Assortative mating for a given phenotype is the phenomenon by which mates select each other based on their phenotypic similarity. Other phenomena can create positive correlation between the parents’ and the offspring’s environment: vertical cultural transmission, or dynastic effects. When these phenomena occur together, they induce a gene-environment correlation at the population scale. It will impact genetic measures of associations such as SNP effect size and SNP-heritability. In this paper, we provide a complete mathematical modelling of both assortative mating and vertical cultural transmission in the classical framework of the polygenic additive model. We establish for the first time the theoretical evolution and equilibrium values of all quantities of interest involved. We then derive its consequences on typical genetic epidemiology analyses; including both population and family-based study designs. We show the consequences on heritability estimation, Genome-Wide Association Studies (GWAS), and the variance explained by polygenic scores. We validate our calculations through simple forward-time simulations.