A gene regulatory network (GRN) records the interactions among transcription factors and their target genes. GRNs are useful to study how transcription factors (TFs) control gene expression as cells transition between states during differentiation and development. Scientists usually construct GRNs by careful examination and study of the literature. This process is slow and painstaking and does not scale to large networks. In this thesis, we study the problem of inferring GRNs automatically from gene expression data. Recent data-driven approaches to infer GRNs increasingly rely on single-cell level RNA-sequencing (scRNA-seq) data. Most of these methods rely on unsupervised or association based strategies, which cannot leverage known regulatory interactions by design. To facilitate supervised learning, we propose a novel graph convolutional neural network (GCN) based autoencoder to infer new regulatory edges from a known GRN and scRNA-seq data. As the name suggests, a GCN-based autoencoder consists of an encoder that learns a low-dimensional embedding of the nodes (genes) in the input graph (the GRN) through a series of graph convolution operations and a decoder that aims to reconstruct the original graph as accurately as possible. We investigate several GCN-based architectures to determine the ideal encoder-decoder combination for GRN reconstruction. We systematically study the performance of these and other supervised learning methods on different mouse and human scRNA-seq datasets for two types of evaluation. We demonstrate that our GCN-based approach substantially outperforms traditional machine learning approaches. / Master of Science / In multi-cellular living organisms, stem cells differentiate into multiple cell types. Proteins called transcription factors (TFs) control the activity of genes to effect these transitions. It is possible to represent these interactions abstractly using a gene regulatory network (GRN). In a GRN, each node is a TF or a gene and each edge connects a TF to a gene or TF that it controls. New high-throughput technologies that can measure gene expression (activity) in individual cells provide rich data that can be used to construct GRNs. In this thesis, we take advantage of recent advances in the field of machine learning to develop a new computational method for computationally constructing GRNs. The distinguishing property of our technique is that it is supervised, i.e., it uses experimentally-known interactions to infer new regulatory connections. We investigate several variations of this approach to reconstruct a GRN as close to the original network as possible. We analyze and provide a rationale for the decisions made in designing, evaluating, and choosing the characteristics of our predictor. We show that our predictor has a reconstruction accuracy that is superior to other supervised-learning approaches.

A representação do Sistema de Regulação Gênica por meio de uma Rede de Regulação Gênica (GRN) pode facilitar a compreensão dos processos biológicos no nível molecular, auxiliando no entendimento do comportamento dos genes, a descoberta da causa de doenças e o desenvolvimento de novas drogas. Através das GRNs pode-se avaliar quais genes estão ativos e quais são suas influências no sistema. Nos últimos anos, vários métodos computacionais foram desenvolvidos para realizar a inferência de redes a partir de dados de expressão gênica. Esta pesquisa apresenta uma análise comparativa de métodos de inferência de GRNs, realizando uma revisão do modelo experimental descrito na literatura atual aplicados a conjuntos de dados contendo poucas amostras. Apresenta também o uso comitês de especialistas (ensemble) para agregar o resultado dos métodos a fim de melhorar a qualidade da inferência. Como resultado obteve-se que o uso de poucas amostras de dados (abaixo de 50) não fornecem resultados interessantes para a inferência de redes. Demonstrou-se também que o uso de comitês de especialistas melhoram os resultados de inferência. Os resultados desta pesquisa podem auxiliar em pesquisas futuras baseadas em GRNs. / The representation of the gene regulation system by means of a Gene Regulatory Network (GRN) can help the understanding of biological processes at the molecular level, elucidating the behavior of genes and leading to the discovery of disease causes and the development of new drugs. GRNs allow to evaluate which genes are active and how they influence the system. In recent years, many computational methods have been developed for networks inference from gene expression data. This study presents a comparative analysis of GRN inference methods, reviewing the experimental modeling present in the state-of-art scientific publications applied to datasets with small data samples. The use of ensembles was proposed to improve the quality of the network inference. As results, we show that the use of small data samples (less than 50 samples) do not show a good result in the network inference problem. We also show that the use of ensemble improve the network inference.

Gene expression is tightly regulated by complex transcriptional regulatory mechanisms to achieve specific expression patterns, which are essential to facilitate important biological processes such as embryonic development. Dysregulation of gene expression can lead to diseases such as cancers. A better understanding of the transcriptional regulation will therefore not only advance the understanding of fundamental biological processes, but also provide mechanistic insights into diseases. The earlier versions of high-throughput expression profiling techniques were limited to measuring average gene expression across large pools of cells. In contrast, recent technological improvements have made it possible to perform expression profiling in single cells. Single-cell expression profiling is able to capture heterogeneity among single cells, which is not possible in conventional bulk expression profiling. In my PhD, I focus on developing new algorithms, as well as benchmarking and utilising existing algorithms to study the transcriptomes of various biological systems using single-cell expression data. I have developed two different single-cell specific network inference algorithms, BTR and SPVAR, which are based on two different formalisms, Boolean and autoregression frameworks respectively. BTR was shown to be useful for improving existing Boolean models with single-cell expression data, while SPVAR was shown to be a conservative predictor of gene interactions using pseudotime-ordered single-cell expression data. In addition, I have obtained novel biological insights by analysing single-cell RNAseq data from the epiblast stem cells reprogramming and the leukaemia systems. Three different driver genes, namely Esrrb, Klf2 and GY118F, were shown to drive reprogramming of epiblast stem cells via different reprogramming routes. As for the leukaemia system, FLT3-ITD and IDH1-R132H mutations were shown to interact with each other and potentially predispose some cells for developing acute myeloid leukaemia.

A representação do Sistema de Regulação Gênica por meio de uma Rede de Regulação Gênica (GRN) pode facilitar a compreensão dos processos biológicos no nível molecular, auxiliando no entendimento do comportamento dos genes, a descoberta da causa de doenças e o desenvolvimento de novas drogas. Através das GRNs pode-se avaliar quais genes estão ativos e quais são suas influências no sistema. Nos últimos anos, vários métodos computacionais foram desenvolvidos para realizar a inferência de redes a partir de dados de expressão gênica. Esta pesquisa apresenta uma análise comparativa de métodos de inferência de GRNs, realizando uma revisão do modelo experimental descrito na literatura atual aplicados a conjuntos de dados contendo poucas amostras. Apresenta também o uso comitês de especialistas (ensemble) para agregar o resultado dos métodos a fim de melhorar a qualidade da inferência. Como resultado obteve-se que o uso de poucas amostras de dados (abaixo de 50) não fornecem resultados interessantes para a inferência de redes. Demonstrou-se também que o uso de comitês de especialistas melhoram os resultados de inferência. Os resultados desta pesquisa podem auxiliar em pesquisas futuras baseadas em GRNs. / The representation of the gene regulation system by means of a Gene Regulatory Network (GRN) can help the understanding of biological processes at the molecular level, elucidating the behavior of genes and leading to the discovery of disease causes and the development of new drugs. GRNs allow to evaluate which genes are active and how they influence the system. In recent years, many computational methods have been developed for networks inference from gene expression data. This study presents a comparative analysis of GRN inference methods, reviewing the experimental modeling present in the state-of-art scientific publications applied to datasets with small data samples. The use of ensembles was proposed to improve the quality of the network inference. As results, we show that the use of small data samples (less than 50 samples) do not show a good result in the network inference problem. We also show that the use of ensemble improve the network inference.

Many diseases, such as cardiovascular diseases, cancer and diabetes, originate from several malfunctions in biological systems. The human body is regulated by a wide range of biological systems, composed of biological entities interacting in complex networks, responsible for carrying out specific functions. Some parts of the networks, such as hubs serving as master regulators, are more important for maintaining a function. To find the cause of diseases, where hubs are possible disease regulators, it is critical to know the structure of these biological systems. Such structures can be reverse engineered from high-throughput data with measured levels of biological entities. However, the complexity of biological systems makes inferring their structure a complicated task, demanding the use of computational methods, called network inference methods. Today, many network inference methods have been developed, that predicts the interactions of biological networks, with varying degree of success. In the DREAM5 challenge 35 network inference methods were evaluated on how well interactions in gene regulatory networks (GRNs) were predicted. Herein, in contrast to the DREAM5 challenge, we have evaluated network inference methods’ ability to predict hubs in GRNs. In accordance with the DREAM5 challenge, different methods performed the best on different data sets. Moreover, we discovered that network inference methods were not able to identify hubs from groups of similarly expressed genes. Also, we noticed that hubs in GRNs had a distinct expression in the data, leading to the development of a new method (the PCA method) for the prediction of hubs. Furthermore, the DREAM5 challenge showed that community predictions, combining the predictions from many network inference methods, resulted in more robust predictions of interactions. Herein, the community approach was applied on predicting hubs, with the conclusion that community predictions is the more robust approach. However, we also concluded that it was enough to combine 6-7 network inference methods to achieve robust predictions of hubs.

This master's thesis deals with possible applications of Bayesian networks. The theoretical part is mainly of mathematical nature. At first, we focus on general probability theory and later we move on to the theory of Bayesian networks and discuss approaches to inference and to model learning while providing explanations of pros and cons of these techniques. The practical part focuses on applications that demand learning a Bayesian network, both in terms of network parameters as well as structure. These applications include general benchmarks, usage of Bayesian networks for knowledge discovery regarding the causes of criminality and exploration of the possibility of using a Bayesian network as a spam filter.

Les interactions entre micro-organismes sous-tendent de nombreux services écosystémiques, y compris la régulation des maladies des plantes cultivées. Un acteur de cette régulation est le pathobiome, défini comme le sous-ensemble des micro-organismes associés à une plante hôte en interaction avec un agent pathogène. L'un des défis actuels consiste à reconstruire les pathobiomes à partir de données de metabarcoding, pour identifier des agents potentiels de biocontrôle et pour surveiller en temps réel leurs réponses aux changements environnementaux. Plusieurs verrous méthodologiques doivent cependant être levés pour atteindre ces objectifs. Tout d’abord, il n’existe pas de consensus concernant l’approche bioinformatique la plus fiable pour déterminer l’identité et l’abondance des micro-organismes présents dans les échantillons végétaux. De plus, les réseaux microbiens construits avec les méthodes actuellement disponibles sont des réseaux d’associations statistiques entre des comptages de séquences, non directement superposables aux réseaux d’interactions (ex : compétition, parasitisme) entre micro-organismes. L’objectif de la thèse était donc de déterminer les approches bioinformatiques et statistiques les plus pertinentes pour reconstruire des réseaux d’interactions microbiennes à partir de données de metabarcoding. Le modèle d’étude était la vigne (Vitis vinifera L. cv. Merlot noir) et l’oïdium de la vigne, Erysiphe necator. Nous avons tout d’abord déterminé l’approche bioinformatique la plus adaptée pour identifier la communauté fongique associée à ce pathogène, en comparant la capacité de 360 pipelines à retrouver la composition d’une communauté artificielle de 189 souches fongiques. DADA2 est apparu comme l’outil le plus performant. Nous avons ensuite évalué l’influence de la pratique culturale (viticulture conventionnelle vs. biologique) sur les communautés fongiques des feuilles et évalué le niveau de réplicabilité des réseaux microbiens construits avec une méthode d’inférence classique, SparCC. La réplicabilité était très faible, jetant ainsi un doute sur l’utilité de ces réseaux pour le biocontrôle et la biosurveillance. Nous avons donc utilisé une nouvelle approche statistique, le modèle PLN, qui permet de prendre en compte la variabilité environnementale, pour explorer finement le pathobiome d’Erysiphe necator. Les interactions microbiennes prédites par le modèle sont en cours de comparaison avec des expériences de confrontations de levures en co-cultures. Une approche alternative, HMSC, a également été testée sur un autre modèle biologique et certaines prédictions ont été confrontées avec succès aux données de la littérature. Les réseaux microbiens, sous réserve d’amélioration des méthodes de reconstruction, pourraient donc être utilisés pour capturer les signaux des interactions biotiques dans le pathobiome. / Interactions between microorganisms underpin many ecosystem services, including the regulation of crop diseases. An actor in this regulation is the pathobiome, defined as the subset of microorganisms associated with a host plant in interaction with a pathogen. One of the current challenges is to reconstruct pathobiomes from metabarcoding data, in order to identify potential biocontrol agents and to monitor in real time their responses to environmental changes. However, several methodological hurdles must be overcomed to achieve these objectives. First, there is no consensus on the most reliable bioinformatics approach to determine the identity and abundance of microorganisms present in plant samples. In addition, microbial networks built with currently available methods are networks of statistical associations between sequence counts, not directly related to networks of interactions (e. g. competition, parasitism) between microorganisms. The objective of the thesis was therefore to determine the most relevant bioinformatics and statistical approaches to reconstruct microbial interaction networks from metabarcoding data. The study system was grapevine (Vitis vinifera L. cv. Merlot noir) and the fungal agent of grapevine powdery mildew Erysiphe necator. First, we determined the most appropriate bioinformatics approach to identify the fungal community associated with this pathogen, by comparing the ability of 360 pipelines to recover the composition of an artificial community of 189 fungal strains. DADA2 has emerged as the most powerful tool. We then evaluated the influence of the cropping system (conventional vs. organic viticulture) on foliar fungal communities and assessed the level of replicability of microbial networks built with a standard inference method, SparCC. Replicability was very low, casting doubt on the usefulness of these networks for biocontrol and biomonitoring We therefore used a new statistical approach, the PLN model, which allows us to take into account environmental variability, to finely explore the pathobiome of Erysiphe necator. The microbial interactions predicted by the model are being compared with experiments confronting yeasts in co-cultures. An alternative approach, HMSC, was also tested on another biological model and some predictions were successfully compared with the data in the literature. Microbial networks, provided improved reconstruction methods, could therefore be used to capture signals of biotic interactions in the pathobiome.

We demonstrate the success of recurrent neural networks in gene network inference and expression prediction using a hybrid of particle swarm optimization and differential evolution to overcome the classic obstacle of local minima in training recurrent neural networks. We also provide an improved validation framework for the evaluation of genetic network modeling systems that will result in better generalization and long-term prediction capability. Success in the modeling of gene regulation and prediction of gene expression will lead to more rapid discovery and development of therapeutic medicine, earlier diagnosis and treatment of adverse conditions, and vast advancements in life science research.

No description available.

Este estudo tem como objetivo melhorar o desempenho computacional dos algoritmos de inferência em redes credais, aplicando técnicas de computação paralela e sistemas distribuídos em algoritmos de fatoração de matrizes esparsas. Grosso modo, técnicas de computação paralela são técnicas para transformar um sistema em um sistema com algoritmos que possam ser executados concorrentemente. E a fatoração de matrizes são técnicas da matemática para decompor uma matriz em um produto de duas ou mais matrizes. As matrizes esparsas são matrizes que possuem a maioria de seus valores iguais a zero. E as redes credais são semelhantes as redes bayesianas, que são grafos acíclicos que representam uma probabilidade conjunta através de probabilidades condicionais e suas relações de independência. As redes credais podem ser consideradas como uma extensão das redes bayesianas para lidar com incertezas ou a má qualidade dos dados. Para aplicar a técnica de paralelização de fatoração de matrizes esparsas na inferência de redes credais, a inferência utiliza-se da técnica de eliminação de variáveis onde o grafo acíclico da rede credal é associado a uma matriz esparsa e cada variável eliminada é análoga a eliminação de uma coluna. / This study\'s objective is the computational performance improvement of credal network inference algorithms by applying computational parallel and distributed system techniques of sparse matrix factorization algorithms. Roughly, computational parallel techniques are used to transform systems in systems with algorithms that can be executed concurrently. And the matrix factorization is a group of mathematical techniques to decompose a matrix in a product of two or more matrixes. The sparse matrixes are matrixes which have most of their values equal to zero. And credal networks are similar to Bayesian networks, which are acyclic graphs representing a joint probability through conditional probabilities and their independence relations. Credal networks can be considered as a Bayesian network extension because of their manner of leading to uncertainty and the poor data quality. To apply parallel techniques of sparse matrix factorization in credal network inference the variable elimination method was used, where the credal network acyclic graph is associated to a sparse matrix and every eliminated variable is analogous to an eliminated column.