Living organisms portray diverse patterns of growth and developmental regulation. The entire process, beginning from a single cell to the formation of tissues and organ systems and the final culmination in a form that characterizes a fully grown organism is closely guarded by numerous molecular pathways. Like a master conductor, the genetic material of the cell which is stored in its DNA (deoxyribonucleic acid), determines the fate of each individual cell and demarcates its developmental direction. Although in an organism, this genetic material is normally identical in every cell, there are differences in the ways cells respond to them. For example, skin cells behave differently from muscle cells and their developmental processes are highly variable in space and time. It is now known that based on intrinsic or extrinsic environmental cues, the information passed on from DNA is converted into a functional protein product through an RNA (ribonucleic acid) intermediate. Thus, different genes fire at different times leading to diverse patterns of developmental regulation among cells. However, there exists a stark contrast between the size of the genome (DNA), the transcriptome (transcribed RNAs) and the proteome (functional protein products) inside a cell. While there are abundant RNA molecules that are transcribed from the DNA, only few give rise to proteins. The search for function of RNAs that do not code for proteins is a relatively new topic in molecular biology. With advancements in sequencing methodologies, there is a rapid surge in the discovery of such molecules but due to the nonavailability of systematic tools to study them, the functional characterization of these RNAs has been relatively slow. Even among the non coding RNAs, there exists small and long varieties of which the long non coding RNAs (lncRNAs) have more heterogenous functional attributes. The roles that these lncRNAs play in development is only recently emerging, especially in the field of embryonic stem (ES) cell biology. ES cells are of particular interest to researchers due to their properties of replicating indefinitely in culture and giving rise to all the germ layers that eventually constitute an organism. These unique abilities make them perfect models to study essential cellular developmental processes and also contribute to the understanding of the molecular pathways that ultimately lead to diseases like like other processes, is orchestrated by a host of different factors in which lncRNAs are slowly emerging as important players. Although there are thousands of lncRNAs identified, only a few have been implicated in pluripotency. I reasoned that there should be more such candidates and to study them one needs to develop a strategy to functionally investigate several lncRNAs simultaneously. Loss-of-function screens have been extremely successful for dissecting the functions of protein coding genes. Among the triggers for conducting such screens, endoribonuclease-prepared small interfering RNAs (esiRNAs) have been demonstrated as effective mRNA depletion agents with minimum silencing of non-intended targets. Since these RNA interference (RNAi) agents had not been comprehensively tested on lncRNAs, I used them for conducting a screen to discover lncRNAs involved in pluripotency. Using a combination of RNAi and localization strategies, I here report the discovery of a novel lncRNA called Panct1 which through interaction with other factors takes part in the ES cell pluripotency programme. In the process of characterization of Panct1, I have also identified and partially characterized a potential DNA binding protein called CXORF23 which might emerge as an important player in the determination of stem cell fate. These discoveries hint towards the presence of more such lncRNA protein interactions and further widen our understanding of stem cell biology.

Des études récentes ont mis en évidence qu’au moins 70% du génome humain est transcrit et produit une myriade d’ARN non codants. Au début de ma thèse j’ai utilisé des données de RNA-Seq sens-spécifique pour identifier les transcrits divergents dans les tissus primaires de souris. J’ai utilisé aussi des données ChIP-Seq afin d’analyser leurs caractéristiques épigénétiques. Nous avons trouvé que la transcription divergente est associée de manière significative à des gènes liés à la régulation de la transcription et le développement.Dans un deuxième temps, je me suis intéressé à l'identification et la caractérisation des lncRNA chez l'homme. J’ai appliqué des approches statistiques pour quantifier leur expression et identifier ceux qui sont (dé)régulés dans un contexte normal ou leucémique Dans un troisième temps. Au cours de ma thèse, je me suis attaché à étudier le mécanisme moléculaire épigénomique ainsi qu'à développer un pipeline bioinformatique permettant d'identifier les gènes (codant ou non codant) associés à des profils H3K4me2/3 étendus. Ainsi, j’ai mis en évidences que ces profils étendus étaient directement dépendants d'un processus transcriptionelle impliquant des nouveaux mécanismes de régulation. Cette étude a donné aussi lieu à une publication dont je suis cosignataire en premier auteur. (Zacarias, Belhocine et al. Journal of Immunology 2015). Cette nouvelle approche devrait s'avérer très utile dans d'autres modèles développementaux et/ou pathologiques et peuvent être utilisé comme outil de prioritisation des candidats les plus relevant dans des approches plus globale. / Recent studies indicate that at least 70% of the human genome is transcribed into a myriad of both coding and non-coding RNAs. at the beginning of my thesis I used RNA-Seq data to identify divergent transcripts in mouse primary tissues. I also used the ChIP-Seq data to analyze their epigenetic characteristics. The results demonstrated that divergent transcription was significantly associated with genes related to transcription regulation and development. In a second phase, I was interested in the LncRNAs identification and characterization during the development of human T lymphocytes and in T acute lymphoblastic leukemia (T-ALL). I applied statistical approaches to quantify their expression and identify those that are regulated in a normal or leukemic contextSubsequently, I determined the most appropriate approach to prioritize the functional role of LncRNAs. Indeed, I focused on studying the molecular epigenomic mechanism marking and developed a bioinformatics pipeline to identify genes (coding or non-coding) associated with the extended profiles of H3K4me2/3. Evidence generated through the pipeline demonstrated that these extended profiles were directly dependent on specific transcriptional process involving new regulatory mechanisms.In conclusion, this body of work has resulted in a more comprehensive approach to determining the true functional role of LncRNAs in both normal biological context and in human disease.

The onset of immune response against microbial stimuli activates induction of many anti- inflammatory genes and ISGs for effective clearance of the pathogen. This response includes transcriptional activation of several non-coding transcripts such as miRNAs and long non-coding RNAs (lncRNAs). LncRNAs constitutes the largest class of non-coding genome and are arbitrarily described as transcripts greater than 200 base pairs. Similar to protein coding mRNAs, lncRNAs are RNA polymerase II transcripts and undergo mRNA processing such as capping, splicing and polyadenylation. In recent years, high throughput sequencing has enabled an in-depth exploration of the human genome and subsequent discovery of lncRNAs. Several studies have highlighted the crucial role of lncRNAs in many biological processes including as regulators of gene expression as well as molecular effectors of host-pathogen driven immune responses. To date, majority of lncRNAs have been studied in murine models with limited understanding in human cells. In order to elucidate the role of lncRNAs in human immune cell regulation, the goal of this thesis is to identify and characterize novel lncRNAs critical to host-pathogen innate immune responses. RNA sequencing in LPS, IAV and HSV stimulated cells revealed lncRNA LUCAT1 as most differentially regulated lncRNA. CRISPR-cas9 and shRNA mediated depletion of LUCAT1 showed enhanced IFN-I genes signature, which was suppressed upon overexpression of LUCAT1. Additionally, LPS stimulated hDCs showed enrichment of LUCAT1 in the nucleus and its association with the chromatin markers. Further, LUCAT1 depletion contributed to enhanced occupancy of transcriptional coactivators at the promoters of IFN-I genes. Global identification of RNA associated proteins revealed LUCAT1 association with STAT1 in the nucleus thus emphasizing its role in transcriptional regulation of Type I IFN genes in inflammatory responses. This thesis furthers the understanding about the molecular factors affecting immune regulation and describes the novel role of LUCAT1 as an attenuator of immune cell response to pathogens.

Long non-coding RNAs (lncRNAs) are non-protein coding transcripts longer than 200 nucleotides. Many of these lncRNAs have regulatory functions and have recently emerged as major players in governing fundamental biological processes. Here, we review the definition, distribution, identification, databases, analysis, classification, and functions of lncRNAs. We also discuss the potential roles of lncRNAs in the etiological processes of psychiatric disorders and the implications for clinical diagnosis and treatment. Psychiatr Genet 26:109-116 .

A sinalização celular desencadeada na presença do hormônio andrógeno em células da próstata é dependente da ativação do receptor de andrógeno (AR), que é um fator de transcrição codificado pelo gene NR3C4. O AR quando não ligado ao hormônio andrógeno é encontrado no citoplasma, e a ligação do hormônio ao AR promove seu deslocamento para o núcleo. O AR se liga a motivos de DNA, promovendo a transcrição de determinados genes por meio de um complexo mecanismo de regulação ainda não totalmente conhecido. Neste trabalho exploramos a capacidade dos RNAs longos intergênicos não-codificadores (lincRNAs) se ligarem ao AR e contribuírem para a regulação transcricional de genes. Utilizamos a imunoprecipitação de RNA, seguida de sequenciamento em larga escala (RIP-Seq) para a identificação e quantificação de lincRNAs associados fisicamente ao AR na linhagem celular de próstata LNCaP. Em paralelo utilizamos dados de transcriptoma e de marcas epigenéticas para verificar se a interação física dos lincRNAs com o AR influenciaria a composição da cromatina e a expressão dos genes vizinhos desses ARA-lincRNAs (Androgen Receptor Associated LincRNAs). Como resultado deste estudo descrevemos pela primeira vez centenas de LincRNAs associados fisicamente ao receptor de andrógeno após o tratamento com o hormônio. Observamos que uma parte desses ARA-lincRNAs pode modificar in cis a expressão de genes codificadores de proteínas vizinhos e essa capacidade de regulação é reforçada pela presença de marcas epigenéticas que são características de reguladores in cis. Além disso, mostramos que as regiões da cromatina contendo domínios topologicamente associativos (TADs) que possuem ARA-lincRNAs apresentam fatores de transcrição, marcas epigenéticas e nível de transcrição gênica diferenciadas. Os resultados apresentados neste trabalho estendem a importância dos lincRNAs durante eventos regulatórios complexos e mostra pela primeira a atuação dessas moléculas em sinergia com um fator de transcrição modificando a cromatina e alterando a regulação gênica. / The cell signaling events triggered in the presence of the androgen hormone in prostate cells is dependent on the activation of the androgen receptor (AR), which is a transcription factor encoded by the NR3C4 gene. AR when not bound to the androgen hormone is found in the cytoplasm, and binding of the hormone to AR promotes its displacement to the nucleus. AR binds to DNA motifs, promoting the transcription of certain genes through a complex regulatory mechanism not yet fully undestood. In this work we explore the ability of long non-coding intergenic RNAs (lincRNAs) to bind to AR and contribute to the transcriptional regulation of genes. We used immunoprecipitation of RNA, followed by large-scale sequencing (RIP-Seq) for the identification and quantification of lincRNAs physically associated with AR in the LNCaP prostate cell line. In parallel we used transcriptome data and data on epigenetic marks to verify whether the physical interaction of lincRNAs with AR would influence the chromatin composition and the expression of genes neighboring these ARA-lincRNAs (Androgen Receptor Associated LincRNAs). As a result of this study we described for the first time in the literature hundreds of LincRNAs physically associated with the androgen receptor after treatment with the hormone. We have observed that part of these ARA-lincRNAs can modify in cis the expression of neighboring protein coding genes and this regulatory ability is enhanced by the presence of epigenetic marks that are characteristic of in cis regulators. In addition, we have shown that chromatin regions containing topologically associating domains (TADs) that possess ARA-lincRNAs have different transcription factors, epigenetic marks and gene transcription levels. The results presented in this work extend the importance of the lincRNAs during complex regulatory events and shows for the first time the performance of these molecules in synergy with a transcription factor modifying the chromatin and altering gene regulation.

A sepse é uma das maiores causas de mortalidade em pacientes hospitalizados, e uma complicação comum, tanto em pacientes clínicos quanto de cirurgias, admitidos em hospitais por causas não infecciosas. A sepse é especialmente comum em pacientes mais velhos, sendo portanto esperado que sua incidência aumente com o envelhecimento da população, e apesar da sua maior taxa de mortalidade, a resposta imune em idosos durante o choque séptico é muito similar à dos pacientes mais jovens. O objetivo desse estudo foi de conduzir uma análise de expressão gênica dos neutrófilos da circulação, tanto de pacientes adultos como de pacientes idosos, observando tanto os mRNAs e as vias em que estão envolvidos, como o papel dos ncRNAs, para um melhor entendimento da resposta imune do indivíduo idoso a infecções severas. Os RNAs de 24 indivíduos, divididos igualmente entre idosos e adultos, e entre pacientes em choque séptico e controles, foram hibridizadas em microarranjos de DNA. Deste experimento foram encontrados genes cuja expressão pode ser utilizada para diferenciar a resposta imune entre adultos e idosos. Estes genes foram observados concentrados em algumas vias, entre elas fosforilação oxidativa, disfunção mitocondrial, sinalização do TGF-, entre outras. Além da análise usando os mRNAs, esse trabalho mostra fortes indicações de interações de mRNAs com RNAs não codificadores longos, dos quais a maioria não têm função conhecida. Para propor uma função aos RNAs não codificadores foi construída uma rede de coexpressão em que alguns RNAs de função desconhecida se mostraram fortemente ligados à genes das vias moleculares do ribossomo e da mitocôndria. Também foi observado que para os idosos a rede de coexpressão é menos centralizada, suportando a hipótese de que alterar a expressão de alguns genes chave pode ser o fator determinante para alterar a expressão gênica e um conjunto maior / Sepsis is one of the major causes of mortality in hospitalized patients, and a common complication, both in clinical patients and in surgeries, admitted to hospital for non-infectious causes. Sepsis is especially common in older patients, and is therefore expected that its incidence increases as the population ages, and despite its higher mortality rate, the immune response in the elderly during septic shock is very similar to that of younger patients. The objective of this study was to conduct a gene expression analysis of circulating neutrophils, both adults and elderly patients, noting both the mRNAs, the pathways in which those are involved, and the role of ncRNAs, for a better understanding of the immune response of the elderly to severe infections. The RNAs of 24 individuals, equally divided among the adults and the elderly, and among patients in septic shock and controls, were hybridized to DNA microarrays. From this experiment many genes whose expression can be used to differentiate the immune response in adults and the elderly were found. These genes were concentrated in some metabolic pathways, including oxidative phosphorylation, mitochondrial dysfunction, TGF- signaling, and others. Besides the analysis using mRNAs, this work shows strong indications of mRNAs interactions with non coding RNAs, most of which have no known function. To propose a role for noncoding RNAs a coexpression network was built in which some RNAs of unknown function showed strongly connected to genes of the molecular pathways of the ribosome and mitochondria. It was also noted that for the elderly, the coexpression network is less centralized, supporting the hypothesis that altering the expression of a few key genes can be a determining factor for altering the gene expression of a larger set

A sepse é uma das maiores causas de mortalidade em pacientes hospitalizados, e uma complicação comum, tanto em pacientes clínicos quanto de cirurgias, admitidos em hospitais por causas não infecciosas. A sepse é especialmente comum em pacientes mais velhos, sendo portanto esperado que sua incidência aumente com o envelhecimento da população, e apesar da sua maior taxa de mortalidade, a resposta imune em idosos durante o choque séptico é muito similar à dos pacientes mais jovens. O objetivo desse estudo foi de conduzir uma análise de expressão gênica dos neutrófilos da circulação, tanto de pacientes adultos como de pacientes idosos, observando tanto os mRNAs e as vias em que estão envolvidos, como o papel dos ncRNAs, para um melhor entendimento da resposta imune do indivíduo idoso a infecções severas. Os RNAs de 24 indivíduos, divididos igualmente entre idosos e adultos, e entre pacientes em choque séptico e controles, foram hibridizadas em microarranjos de DNA. Deste experimento foram encontrados genes cuja expressão pode ser utilizada para diferenciar a resposta imune entre adultos e idosos. Estes genes foram observados concentrados em algumas vias, entre elas fosforilação oxidativa, disfunção mitocondrial, sinalização do TGF-, entre outras. Além da análise usando os mRNAs, esse trabalho mostra fortes indicações de interações de mRNAs com RNAs não codificadores longos, dos quais a maioria não têm função conhecida. Para propor uma função aos RNAs não codificadores foi construída uma rede de coexpressão em que alguns RNAs de função desconhecida se mostraram fortemente ligados à genes das vias moleculares do ribossomo e da mitocôndria. Também foi observado que para os idosos a rede de coexpressão é menos centralizada, suportando a hipótese de que alterar a expressão de alguns genes chave pode ser o fator determinante para alterar a expressão gênica e um conjunto maior / Sepsis is one of the major causes of mortality in hospitalized patients, and a common complication, both in clinical patients and in surgeries, admitted to hospital for non-infectious causes. Sepsis is especially common in older patients, and is therefore expected that its incidence increases as the population ages, and despite its higher mortality rate, the immune response in the elderly during septic shock is very similar to that of younger patients. The objective of this study was to conduct a gene expression analysis of circulating neutrophils, both adults and elderly patients, noting both the mRNAs, the pathways in which those are involved, and the role of ncRNAs, for a better understanding of the immune response of the elderly to severe infections. The RNAs of 24 individuals, equally divided among the adults and the elderly, and among patients in septic shock and controls, were hybridized to DNA microarrays. From this experiment many genes whose expression can be used to differentiate the immune response in adults and the elderly were found. These genes were concentrated in some metabolic pathways, including oxidative phosphorylation, mitochondrial dysfunction, TGF- signaling, and others. Besides the analysis using mRNAs, this work shows strong indications of mRNAs interactions with non coding RNAs, most of which have no known function. To propose a role for noncoding RNAs a coexpression network was built in which some RNAs of unknown function showed strongly connected to genes of the molecular pathways of the ribosome and mitochondria. It was also noted that for the elderly, the coexpression network is less centralized, supporting the hypothesis that altering the expression of a few key genes can be a determining factor for altering the gene expression of a larger set

Living organisms portray diverse patterns of growth and developmental regulation. The entire process, beginning from a single cell to the formation of tissues and organ systems and the final culmination in a form that characterizes a fully grown organism is closely guarded by numerous molecular pathways. Like a master conductor, the genetic material of the cell which is stored in its DNA (deoxyribonucleic acid), determines the fate of each individual cell and demarcates its developmental direction. Although in an organism, this genetic material is normally identical in every cell, there are differences in the ways cells respond to them. For example, skin cells behave differently from muscle cells and their developmental processes are highly variable in space and time. It is now known that based on intrinsic or extrinsic environmental cues, the information passed on from DNA is converted into a functional protein product through an RNA (ribonucleic acid) intermediate. Thus, different genes fire at different times leading to diverse patterns of developmental regulation among cells. However, there exists a stark contrast between the size of the genome (DNA), the transcriptome (transcribed RNAs) and the proteome (functional protein products) inside a cell. While there are abundant RNA molecules that are transcribed from the DNA, only few give rise to proteins. The search for function of RNAs that do not code for proteins is a relatively new topic in molecular biology. With advancements in sequencing methodologies, there is a rapid surge in the discovery of such molecules but due to the nonavailability of systematic tools to study them, the functional characterization of these RNAs has been relatively slow. Even among the non coding RNAs, there exists small and long varieties of which the long non coding RNAs (lncRNAs) have more heterogenous functional attributes. The roles that these lncRNAs play in development is only recently emerging, especially in the field of embryonic stem (ES) cell biology. ES cells are of particular interest to researchers due to their properties of replicating indefinitely in culture and giving rise to all the germ layers that eventually constitute an organism. These unique abilities make them perfect models to study essential cellular developmental processes and also contribute to the understanding of the molecular pathways that ultimately lead to diseases like like other processes, is orchestrated by a host of different factors in which lncRNAs are slowly emerging as important players. Although there are thousands of lncRNAs identified, only a few have been implicated in pluripotency. I reasoned that there should be more such candidates and to study them one needs to develop a strategy to functionally investigate several lncRNAs simultaneously. Loss-of-function screens have been extremely successful for dissecting the functions of protein coding genes. Among the triggers for conducting such screens, endoribonuclease-prepared small interfering RNAs (esiRNAs) have been demonstrated as effective mRNA depletion agents with minimum silencing of non-intended targets. Since these RNA interference (RNAi) agents had not been comprehensively tested on lncRNAs, I used them for conducting a screen to discover lncRNAs involved in pluripotency. Using a combination of RNAi and localization strategies, I here report the discovery of a novel lncRNA called Panct1 which through interaction with other factors takes part in the ES cell pluripotency programme. In the process of characterization of Panct1, I have also identified and partially characterized a potential DNA binding protein called CXORF23 which might emerge as an important player in the determination of stem cell fate. These discoveries hint towards the presence of more such lncRNA protein interactions and further widen our understanding of stem cell biology.

O câncer de mama é um dos que mais afeta mulheres em todo o mundo e em 2016 estima-se que haverá aproximadamente 58 mil novos casos no Brasil. É uma doença caracteristicamente heterogênea, o que dificulta o diagnóstico, prognóstico e a abordagem terapêutica utilizada. Os biomarcadores já existentes para câncer de mama não são suficientes para explicar a ocorrência de metástase, que é a principal causa de morte entre os pacientes. Neste sentido, os RNAs longos não codificadores (lncRNAs) vêm se estabelecendo como importantes moléculas regulatórias em diversos processos biológicos e alguns têm sido associados com o desenvolvimento e a progressão do câncer de mama. Apesar disso, muito ainda precisa ser elucidado a respeito dessa nova classe de ncRNAs e seu envolvimento na metástase. Desta forma, determinar uma assinatura de lncRNAs em tumores metastáticos de mama pareados com seus respectivos tumores primários pode sugerir novas moléculas envolvidas no mecanismo de metástase. Neste estudo, foram analisadas amostras de cinco tumores primários de mama e seus correspondentes metastáticos em Sistema Nervoso Central (n=2) e em linfonodos (n=3). O RNA total de cada amostra foi extraído e avaliado quanto à sua qualidade para obtenção do transcriptoma pelo método de RNA-Seq. Para o tratamento dos dados foram usados os aplicativos: Tophat para o alinhamento das reads, o Cufflinks para a montagem e quantificação dos transcritos e o pacote DESeq2 para análise de expressão diferencial. O resultado da análise demonstrou que os lncRNAs com íntrons retidos são os mais abundantes e que as amostras de tumores primários e metastáticos compartilham a maioria dos lncRNAs expressos, tornando-os qualitativamente bastante semelhantes entre si. Dentre os lncRNAs diferencialmente expressos, doze são comuns entre os tumores primários, independente dos subtipos moleculares das amostras, e não há transcritos comuns às amostras metastáticas. A análise de agrupamento hierárquico definiu três assinaturas de expressão de lncRNAs para as amostras de tumores primários, metástase em cérebro e em linfonodos, podendo indicar lncRNAs potencialmente envolvidos no mecanismo de metástase. / Breast cancer is the principall cancer that affects women in the world and it is expect to Brazil, 58.000 new cases in 2016. It is a heterogeneous disease, what makes diagnosis, prognosis and therapeutical approach more difficult. Existing biomarkers for breast cancer are not sufficient to explain metastasis occurrence, which is the main cause of mortality. In this way, long non-coding RNAs (lncRNAs) have been establishing as important regulatory molecules in several biological processes and some of them have been associated to breast cancer development and progression. Nevertheless, many aspects about this new ncRNA class and its involvement in metastasis need to be elucidated. Thus, determining a lncRNA signature of metastatic tumors paired with their related primary tumors may indicate new molecules involved in the mecanism of metastasis. In the present study, Five samples of primary breast tumors and their related metastasis in central nervous system (n=2) and in lymph nodes (n=3) were analyzed. Total RNA of each sample was extracted and had its quality evaluated for transcriptome obtainment by RNA-Seq method. Data processing was performed utilizing the applications: Tophat, for read alignment; Cufflinks, for transcripts assembly and quantification; and DESeq2 package, for differential expression analysis. The results revealed that lncRNAs containing retained introns are the most abundant transcripts and that primary and metastatic samples share most of the expressed lncRNAs, what makes them qualitatively similar to each other. Differential expression analysis showed that twelve lncRNAs are common among primary tumors, apart from samples mollecular subtypes, and metastatic tumors do not have any transcript in common. Hierarchical clustering analysis defined three lncRNA expression signatures, for primary tumors, brain and lymph node metastatis, what may indicate lncRNAs potentially involved in metastasis mechanism.

Dissertação (mestrado)—Universidade de Brasília, Instituto de Ciências Exatas, Departamento de Ciência da Computação, 2017. / Submitted by Raquel Almeida (raquel.df13@gmail.com) on 2018-04-10T18:44:42Z No. of bitstreams: 1 2017_BrunoCoutoK¨ummel.pdf: 3746010 bytes, checksum: fb3e186abc2f80bf5a5302719a1aa78b (MD5) / Approved for entry into archive by Raquel Viana (raquelviana@bce.unb.br) on 2018-04-12T19:37:08Z (GMT) No. of bitstreams: 1 2017_BrunoCoutoK¨ummel.pdf: 3746010 bytes, checksum: fb3e186abc2f80bf5a5302719a1aa78b (MD5) / Made available in DSpace on 2018-04-12T19:37:08Z (GMT). No. of bitstreams: 1 2017_BrunoCoutoK¨ummel.pdf: 3746010 bytes, checksum: fb3e186abc2f80bf5a5302719a1aa78b (MD5) Previous issue date: 2018-04-12 / RNAs não-codificadores longos (long non-coding RNA - lncRNAs) constitui uma classe heterogênea de RNAs que agrega transcritos com pouca capacidade de codificar proteínas e que possuem mais de 200 nucleotídeos em sua composição. Estudos recentes apontam que essas moléculas possuem funções de regulação de processos biológicos importantes dentro das células. Sabe-se também que o nível de expressão dos lncRNAs está correlacionado com diversas doenças genéticas, tais como câncer e doenças neuro-degenerativas. Este trabalho apresenta um método para seleção das características mais relevantes para modelos de aprendizado de máquina aplicados ao problema de distinguir lncRNAs de transcritos codificadores de proteínas. O método proposto, denominadoSingle Score Feature Selection (S2FS), utilizou como características as frequências de 2-mers, 3-mers e 4-mers dos transcritos, para detectar aquelas mais relevantes para distinguir lncRNAs de transcritos codificadores de proteínas. As características identificadas pelo S2FS foram avaliadas nos datasets obtidos de repositórios públicos de transcritos RNAs codicadores de proteínas e de lncRNAs de Homo sapiens, Mus musculus e Danio rerio. Para o dataset de H. sapiens, também foi utilizada a característica da ORF mais longa de cada transcrito. Os resultados obtidos indicam que o S2FS identificou boas características para os modelos de predição de lncRNAs baseados em Random Forest. Nos modelos de classificação testados, as características selecionadas pelo S2FS possibilitaram resultados melhores do que as características selecionadas por um método de seleção univariada de características baseado no escore da função χ2. / Long non-coding RNA(lncRNAs) constitutes a heterogeneous class of RNAs that includes RNAs with more than 200 nucleotides and poor capacity for coding proteins. Recent studies have indicated that these molecules act on critical biological processes inside the cells. However, their expression levels are also correlated with a number of complex human diseases, such as cancer, neuro-degenerative diseases and others. This work proposes a method for feature selection for machine learning methods applied to the task of distinguishing lncRNAs from protein coding transcripts. The proposed method, called Single Score Feature Selection (S2FS), used as features the 2-mer, 3-mer and 4-mer frequencies of the transcripts, in order to detect those more relevant to distinguish lncRNAs from protein coding transcripts. The features identified by S2FS were evaluated on datasets obtained from public repositories of protein coding transcripts and lncRNAs of Homo Sapiens, Mus musculus and Danio rerio. For the H. sapiens dataset, the longest ORF of each transcript was also used as a feature. The obtained results show that the S2FS identified good features for the lncRNA prediction models based on Random Forest. In the tested classification models, the selected features from S2FS enabled better performance results than the features selected by an univariate selection method based on the scores of a χ2 function.