The thesis describes algorithms for detecting regions of recent identity by descent (IBD) from human genetic data and its applications in optimising resequencing studies, genomic predictions and detecting Mendelian subtypes of diseases. Firstly, we describe the algorithm ANCHAP, which scans pairs of multi-point SNP genotypes for sharing IBD of long haplotypes. A comparison with other methods shows that ANCHAP outperforms them in terms of speed or accuracy. We demonstrate the algorithm on data from population isolates - from Orcades, Croatian islands, and from a population of unrelated individuals. We compare the abundance of IBD segments between cohorts, and identify genetic regions where IBD is most common. Secondly, we verify the IBD regions detected from array data against exome sequence data. We estimate that where sharing IBD between a pair of individuals is inferred, this is confirmed by exome data in 98% of cases. Correctness of IBD detection varies with settings of ANCHAP, length of IBD segments, and position with respect to segment endpoints. We find that with sample sizes of 1000 individuals from an isolated population genotyped using a dense SNP array, and with 20% of these individuals sequenced, 65% of sequences of the un-sequenced subjects can be partially inferred. Implementation of such resequencing strategies requires an IBD-based imputation algorithm, which is outlined. Thirdly, we use recent IBD to detect carriers of Mendelian subtypes of colon cancer. We show this with the example of Lynch syndrome, which accounts for about 3% of colon cancer patients. We detect IBD sharing between known and unknown carriers around DNA mismatch-repair genes. Using the IBD relationship, we build and evaluate a model that predicts presence of Lynch Syndrome mutations. Finally, we discuss whether regions of identity by descent can be used for genomic predictions. We conclude that the utility of the inferred IBD regions depends on accuracy of detection, time to most recent common ancestors and mutation rates since.

In large-scale population genomic data sets, individual chromosomes are likely to share extended regions of haplotype identity with others in the sample. Patterns of local haplotype sharing can be highly informative about many processes including historical demography, selection and recombination. However, in outbred diploid populations, the identification of extended shared haplotypes is not straightforward, particularly in the presence of low levels of genotyping error. Here, we introduce a model-based method for accurately detecting extended haplotype sharing between sets of individuals from unphased data. We describe two implementations of the algorithm that can be applied to data sets consisting of thousands of samples. The method leads naturally to an approach for statistical haplotype estimation, which is shown to be comparable in accuracy to current methods. By applying the method to genome-wide SNP data from over 5,000 samples from the UK we show that the N50 maximal haplotype sharing between unrelated samples is typically 2cM, consistent with a population history of rapid exponential growth that started approx. 125 generations ago. In contrast, within two Greek population isolates of approx. 700 individuals the N50 for maximal haplotype sharing is 12.5cM, while for an unrelated Greek sample of the same size the N50 is 1.3cM. By assessing the size and geographical distribution of maximal haplotype sharing within and between all Greek cohorts, we make inference on the extent of isolatedness of each cohort and on recent migration. We additionally date recent coancestry to about 10 generations for the isolates and 90 generations for the unrelated sample, and finnally attempt to date the time of divergence between them.

Advances in high-throughput genomic technologies have facilitated the collection of DNA information for thousands of individuals, providing unprecedented opportunities to explore the genetic architecture of complex disease. One important finding has been that the majority of variants in the human genome are low in frequency or rare. It has been hypothesised that recent explosive growth of the human population afforded unexpectedly large amounts of rare variants with potentially deleterious effects, suggesting that rare variants may play a role in disease predisposition. But, importantly, rare variants embody a source of information through which we may learn more about our recent evolutionary history. In this thesis, I developed several statistical and computational methods to address problems associated with the analysis of rare variants and, foremost, to leverage the genealogical information they encode. First, one constraint in genome-wide association studies is that lower-frequency variants are not well captured by genotyping methods, but instead are predicted through imputation from a reference dataset. I developed the meta-imputation method to improve imputation accuracy by integrating genotype data from multiple, independent reference panels, which outperformed imputations from separate references in almost all comparisons (mean correlation with masked genotypes r²>0.9). I further demonstrated in simulated case-control studies that meta-imputation increased the statistical power to identify low-frequency variants of intermediate or high penetrance by 2.2-3.6%. Second, rare variants are likely to have originated recently through mutation and thereby sit on relatively long haplotype regions identical by descent (IBD). I developed a method that exploits rare variants as identifiers for shared haplotype segments around which the breakpoints of recombination are detected using non-probabilistic approaches. In coalescent simulations, I show that such breakpoints can be inferred with high accuracy (r²>0.99) around rare variants at frequencies <0.05%, using either haplotype or genotype data. Third, I show that technical error poses a major problem for the analysis of whole-genome sequencing or genotyping data, particularly for alleles below 0.05% frequency (false positive rate, FPR=0.1). I therefore propose a novel approach to infer IBD segments using a Hidden Markov Model (HMM) which operates on genotype data alone. I incorporated an empirical error model constructed from error rates I estimated in publicly available sequencing and genotyping datasets. The HMM was robust in presence of error in simulated data (r²>0.98) while nonprobabilistic methods failed (r²<0.02). Lastly, the age of an allele (the time since its creation through mutation) may provide clues about demographic processes that resulted in its observed frequency. I present a novel method to estimate (rare) allele age based on the inferred shared haplotype structure of the sample. The method operates in a Bayesian framework to infer pairwise coalescent times from which the age is estimated using a composite posterior approach. I show in simulated data that coalescent time can be inferred with high accuracy (rank correlation >0.91) which resulted in a likewise high accuracy for estimated age (>0.94). When applied to data from the 1000 Genomes Project, I show that estimated age distributions were overall conform with frequency-dependent expectations under neutrality, but where patterns of low frequency and old age may hint at signatures of selection at certain sites. Thus, this method may prove useful in the analysis of large cohorts when linked to biomedical phenotype data.

Das Ziel der vorliegenden Dissertation war die Erweiterung der Betrachtungseinheit des Abstammungskoeffizienten von einzelnen Loci auf Chromosomensegmente der Länge x in Morgan. Das neue Maß mit der Bezeichnung epistatische Kinship beschreibt die Wahrscheinlichkeit, dass zwei zufällig gezogene Chromosomen-segmente herkunftsgleich sind. In einer Simulationsstudie wurden die genetischen Eigenschaften von Chromosomensegmenten theoretisch untersucht. Im Weiteren wurde die markergestützte Schätzung der epistatischen Kinship untersucht, da Abstammungsinformation in kleinen Nutztieropulationen oft nicht verfügbar ist. Abschliessend wurden die Eigenschaften der epistatischen Kinship anhand von praktischen Daten untersucht. Dazu wurden DNA-Proben aus drei Subpopulationen des Göttinger Minischweines für sechs Chromosomensegmente mit Mikrosatelliten typisiert. Die markergestützte epistatische Kinship variierte zwischen den einzelnen Segmenten. Durch die Verwendung eines Korrekturfaktors für zustands- jedoch nicht herkunftsgleiche Segmente wurde die Variabilität zwischen den Segmenten kleiner. Zur Abschätzung der genetischen Diversität wurde ein genetisches Distanzmass hergeleitet, in welchem die genetischen Diversität zwischen wie auch innerhalb Populationen berücksichtigt wird. Die Reihenfolge der genetischen Distanzen für die markergestützten Schätzungen stimmte mit der Reihenfolge der pedigreebasierten Erwartungswerte überein. Basierend auf den Ergebnissen wird die epistatische Kinship als neues Maß zur Bestimmung der genetischen Diversität in Nutztierpopulationen mit kurzem Entwicklungszeitraum vorgeschlagen.

La complexité de l’étude des neuropathies héréditaires provient de leur hétérogénéité clinique et génétique et de la diversité des fibres composant les nerfs périphériques. Cette complexité se reflète dans les nombreuses classifications différentes. Les neuropathies héréditaires se classifient entre autres selon leur mode de transmission et leur atteinte sensitive, autonomique et motrice. Les neuropathies héréditaires sensitives et autonomiques (NHSA) se présentent avec une perte de la sensation distale aux membres, accompagnée d’autres manifestations selon le type de NHSA. L’étude des NHSA est facilitée lorsqu’il existe des grappes de familles originaires de régions du Québec où des effets fondateurs pour des maladies récessives ont déjà été identifiés. Nous avons recruté une grande famille canadienne-française originaire de Paspébiac dans la Baie-des-Chaleurs dans laquelle nous avons identifié quatre cas atteints d’une neuropathie héréditaire sensitive avec rétinite pigmentaire et ataxie (NHSRPA). Nous avons émis l’hypothèse que nous étions en présence d’une nouvelle forme de neuropathie héréditaire sensitive récessive à effet fondateur. Afin d’identifier la position chromosomique du gène muté responsable de la NHSRPA, nous avons tout d’abord complété un criblage du génome en génotypant des marqueurs microsatellites «single tandem repeat» (STR) sur des individus clés et nous avons ensuite procédé à une analyse de liaison génétique paramétrique. Ces études nous ont permis de lier cette famille au chromosome 1 et de définir un premier intervalle candidat de 6,7 Mb. Grâce à un génotypage de marqueurs «single nucleotide polymorphism» (SNP), nous avons réduit l’intervalle candidat à 5,3 Mb au locus 1q32,2-q32,3. Cette région contient 44 gènes candidats. Une revue plus fine de la littérature a fait ressortir qu’une famille espagnole et une américaine de souche hollandaise souffrant de la même maladie avaient déjà été liées au même locus. L’origine possiblement basque de notre famille gaspésienne nous a poussé à comparer l’haplotype porteur avec celui de la famille espagnole qui, quoi que gitane, provient du pays basque espagnol. Ces travaux ont démontré le partage d’une région de 203 kb. Afin de rétrécir davantage notre intervalle candidat, nous avons comparé les haplotypes des cas entre les deux familles et nous avons identifié un dernier intervalle candidat de 60 SNP au locus 1q32,3. Cette région ne contient que quatre gènes candidats dont le plus intéressant est le gène «activating transcription factor» (ATF3). À ce jour, aucune mutation n’a été trouvée dans le gène ATF3 et les gènes FAM71A, BATF3 et NSL1. Des expériences supplémentaires sont nécessaires afin d’identifier le gène muté responsable de la NHSRPA. / Hereditary neuropathies study’s complexity comes from their clinical and genetic heterogeneity and the peripheral nerves fibers’ diversity. This complexity leads to numerous different classifications. Hereditary neuropathies are classified based on the transmission mode and the sensitive, autonomic and motor affection. Hereditary sensory and autonomic neuropathies (HSAN) present themselves with members’ distal loss and other manifestations depending on the HSAN type. HSAN study can be facilitated when there are existing family grapes originating from Quebec regions where recessive diseases founder effects have been identified. We have recruited a large French-Canadian family originating from Paspébiac in the Baie-des-Chaleurs in which we have identified four cases affected by a hereditary sensory neuropathy with retinitis pigmentosa and ataxia (HSNRPA). We have hypothesized that we were in presence of a new form of recessive hereditary sensitive neuropathy with founder effect. In order to identify the HSNRPA causing mutated gene chromosomal position, we first completed a genome wide scan by genotyping microsatellite single tandem repeat (STR) markers on informative individuals and we have then proceeded to a parametric genetic linkage analysis. These studies allowed us to link this family to chromosome 1 and define a first candidate interval of 6.7 Mb. Second to the single nucleotide polymorphism (SNP) markers genotyping, we have reduced the candidate interval at 5.3 Mb on locus 1q32.2-q32.3. This region contains 44 genes. A finer literature review made us realize that a Spanish family and an American from Dutch origin suffering from the same disease had already been linked to the same locus. The possible Gaspesian family’s Basque origins brought us to compare their carrier haplotype with the Spanish family’s, although Gypsy but coming from the Spanish Basque country. This work has demonstrated a shared region of 203 kb. In order to further refine our candidate interval, we have compared the haplotypes of the cases between the two families and we have identified a last candidate interval of 60 SNP at locus 1q32.3. This region contains only four candidate genes, the activating transcription factor (ATF3) gene being the most interesting one. Until today, no mutation has been found in the ATF3 gene and in the FAM71A, BATF3 and NSL1 genes. Further experiments will be necessary in order to identify the HSNRPA causing mutated gene.

No description available.