Inferência de genes diferencialmente expressos utilizando modelos lineares generalizados mistos

• Autor(a) principal: Machado, Douglas Terra
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do LNCC
• Texto Completo: https://tede.lncc.br/handle/tede/348
• topic: Bioinformática; Expressão gênica; Expressão gênica diferencial; Modelo em escala genômica; CNPQ::CIENCIAS BIOLOGICAS::BIOLOGIA GERAL

With the technological advance involving the sequencing of RNAs (RNA-Seq) and with the Bioinformatics tools, it is possible to quantify the transcriptional levels of genes in cells, tissues and cell lines, allowing the identification of Differentially Expressed Genes (DEGs). In the scientific literature, DESeq2 and edgeR tools are used for this inference through generalized linear models, which are models that consider only fixed effects in the experimental design. However, the non-inclusion of random effects, which are effects that cause more variability in the experimental design, raises the possibility that possible DEGs that could be important in the context of the biological phenomenon under study are missed. This dissertation proposes the development of two computational tools that use fixed effects and random effects for an accurate inference of DEGs: DEGFiE (Differentially Expressed Genes with Fixed Effects) and DEGRE (Differentially Expressed Genes with Random Effects). The coefficient of biological variation was inferred in the counting matrices to analyze the technical and biological variability of the data sets before and after the pre-processing step. The computational validation of the DEGFiE and DEGRE tools was performed through the simulation of counting matrices, which have biological variability related to fixed and random effects. In addition to the simulated matrices, public RNA-Seq data were retrieved from both patients with bipolar disorder and healthy subjects. These data were used as a preliminary analysis of the application of the DEGRE tool in real experiments, helping to validate this tool. The performance in detecting the DEGs of the DEGFiE and DEGRE tools was compared with the DESeq2 and edgeR tools and evaluated through the evaluation metrics: accuracy, precision and sensitivity. The DEGFiE tool presented the equivalent evaluation metrics with the DESeq2 and edgeR tools, showing that the pre-processing developed in this dissertation was efficient in removing technical noise from the matrices. The tool DEGRE presented better evaluation metrics in the detection of DEGs in matrices with greater biological variability of random effects. This tool also detected new DEGs in the transcriptome data of patients with bipolar disorder when considering the sex of patients as a random effect. Due to the aspects discussed involving the pre-processing stage and the inference of DEGs, the DEGRE tool, proposed in this dissertation, shows promise in the identification of possible DEGs in experiments derived from more complex experimental designs. Keyworkds: gene dispersion, differentially expressed genes, random effects, generalized linear mixed model, pre-processing.