Search

Continuous technological innovation is enveloping all areas of Information Technology. Among the many, one of the main objectives of this evolution remains process optimization. In recent years, the advent of new hardware technologies, such as Graphical Processing Units (GPUs), has made this type of optimization more and more feasible, introducing in practice a change in the development paradigm. This change has forced a switch from sequential implementation, typical of Central Processing Units (CPU), to a parallel implementation. This type of approach is particularly efficient given the presence, within this hardware, of a large number of microprocessors, called threads. In the field of bioinformatics, in particular, this paradigm shift allows to obtain great results in speeding up the execution of algorithms typically expensive at the computational level, bringing the decrease in the time of execution of DNA analysis from several months to a few hours. The objective of this thesis project is to carry out a study in order to theorize, implement and test techniques for the optimization of algorithms for DNA analysis. In detail, the project is part of a series of projects for the optimization of a particular analysis tool: the Germline Pipeline. Previous projects to this one have focused on the phase of Alignment and Sorting of reads, while this one has as focus the Variant Calling. The development phase included the acceleration of the Pari-HMM Forward Algorithm (PFA). The algorithm has been analyzed in all its components and, following a new logic of parallel implementation, a new version has been developed that would allow, returning the same results, to decrease considerably the execution time. The analysis of the obtained results, compared both with a CPU version of the algorithm and with other known state-of-the-art results, has confirmed how it is possible to obtain extremely encouraging results through the use of graphical hardware. In addition, he pointed out that it is not necessary to use very expensive hardware in order to execute some algorithms in reasonable time, emphasizing how the assembly of hybrid architectures, combined with GPU-optimized algorithms can lead not only to a gain in terms of execution time but also to a reduction in expenses for the hardware needed to perform the analysis. On this basis, it is recommended that researchers take advantage of these types of technologies in order to optimize execution time. Future research could be aimed at completing the optimization of the algorithms that form the Germline Pipeline, followed by a subsequent integration with the work of this thesis project in order to compare the whole analysis process with other GPU approaches and find the most efficient approach.

Background Next generation sequencing (NGS) technologies have improved the study of hereditary diseases. Since the evaluation of bioinformatics pipelines is not straightforward, NGS demands effective strategies to analyze data that is of paramount relevance for decision making under a clinical scenario. According to the benchmarking framework of the Global Alliance for Genomics and Health (GA4GH), we implemented a new simple and user-friendly set-theory based method to assess variant callers using a gold standard variant set and high confidence regions. As model, we used TruSight Cardio kit sequencing data of the reference genome NA12878. This targeted sequencing kit is used to identify variants in key genes related to Inherited Cardiac Conditions (ICCs), a group of cardiovascular diseases with high rates of morbidity and mortality. Results We implemented and compared three variant calling pipelines (Isaac, Freebayes, and VarScan). Performance metrics using our set-theory approach showed high-resolution pipelines and revealed: (1) a perfect recall of 1.000 for all three pipelines, (2) very high precision values, i.e. 0.987 for Freebayes, 0.928 for VarScan, and 1.000 for Isaac, when compared with the reference material, and (3) a ROC curve analysis with AUC > 0.94 for all cases. Moreover, significant differences were obtained between the three pipelines. In general, results indicate that the three pipelines were able to recognize the expected variants in the gold standard data set. Conclusions Our set-theory approach to calculate metrics was able to identify the expected ICCs related variants by the three selected pipelines, but results were completely dependent on the algorithms. We emphasize the importance to assess pipelines using gold standard materials to achieve the most reliable results for clinical application.

Le syndrome d’interruption de la tige hypophysaire (PSIS) est un désordre rare qui affecte la fonction du système endocrinien. Jusqu’à nos jours, l’imagerie par la résonance magnétique (IRM) demeure la méthode la plus entreprise afin d’évaluer in vivo l’anomalie d’organogenèse de la tige hypophysaire chez les patients. L’absence de la tige caractérise une déficience permanente en hormone de croissance (GH) pendant que l’étiologie du syndrome demeure inconnue. PSIS se définit comme l’hypopituitarisme congénital et il se caractérise soit par une ectopique post- hypophysaire, soit par une hypoplasie antéhypophysaire, ou encore par une hypoplasie de la tige hypophysaire. Notre objectif consiste à déterminer les mutations génétiques partagées entre les sujets affectés de l’étude et qui pourraient expliquer les causes du syndrome. Pour y parvenir, nous avons analysé les données de séquençage d’exomes (WES) provenant de sept familles caucasiennes simplex, une famille d’origine arabique et cinq autres dont la généalogie est incomplète. Ces données ont été précédemment analysées, pour le même but, par d’autres membres de l’équipe en utilisant le pipeline bio-informatique standard basé sur l’utilisation du logiciel GATK. Nous avons préférentiellement opté pour une nouvelle analyse en utilisant deux différents pipelines bio-informatiques indépendants, pour ensuite comparer conjointement les résultats obtenus. Notre protocole consiste à assembler : d’abord, deux pipelines alternatifs de détection de mutations génétiques ponctuelles (SNV). Ils sont composés d’un logiciel d’alignement de séquence (Bowtie2) et deux logiciels d’appel de variantes (Freebayes et SAMtools). Ensuite, nous avons assemblé trois pipelines de détection de variations du nombre de copies (CNV) génomiques composés communément d’un logiciel d’alignement de séquence (BWA) et trois logiciels d’appel de CNV (CoNIFER, fishingCNV, xHmm). Nos résultats nous ont permis d’identifier des mutations candidates additionnelles qui n’ont jamais été identifiées. De plus, notre méthodologie nous a permis de caractériser certains résultats faux positifs, par conséquent elle pourra nous aider à améliorer la performance des pipelines de détection de variations génomiques existantes., Pituitary stalk interruption syndrome (PSIS) is a rare disorder that affects the function of the endocrine system of the affected individuals. The absence of the pituitary stalk, assessed by MRI, characterizes patients with permanent growth hormone deficiency while the etiology of the syndrome remains unknown. PSIS is defined by clinical hypopituitarism together with anatomical findings including a hypoplastic anterior pituitary, ectopic posterior pituitary and reduced or hypoplastic pituitary stalk. We aim to find shared variations (SNP, CNV) among affected patients in coding regions which could explain the origin of the syndrome. We analyzed the exome NGS data from 8 affected French Canadian trio families, with one additional consanguineous Arabic trio family and 5 families with incomplete pedigree. These data were previously analyzed, for the same objective, by other members of the team using a standardize GATK based bioinformatics pipeline. It was desired to reanalyze the complete data set with two other independent pipelines, followed by a comparison of the SNP discovery results. In the present aspect of the study, we built two SNP discovery pipelines, both composed of a different NGS data aligners (Bowtie2) and each composed a different variant caller (Freebayes, SAMtools), then a CNV discovery pipeline which is composed of three different CNV callers (CoNIFER, fishingCNV, xHmm). In addition to the candidate mutations identified in the previous analysis, we identified additional candidate mutations which had not been detected and never been reported. Furthermore, our method helps to discover the sources of variation false discovery which could help to improve existing genomic mutation discovery pipelines.