Análise de padrões transcriptômicos em pacientes comcâncer de mama: uma abordagem de aprendizado de máquina

• Ano de defesa: 2025
• Autor(a) principal: Padre, Natalia Santos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Bioinformática; UTFPR
• Palavras-chave em Português: Mamas - Câncer; Aprendizado do computador; Bioinformática; Breast - Cancer; Machine learning; Bioinformatics; CNPQ::CIENCIAS SOCIAIS APLICADAS; Engenharia/Tecnologia/Gestão
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/38023

Breast cancer is one of the leading causes of death among women and is among the most prevalent worldwide. Thus, the search for more accurate diagnoses, treatments, and cancer prevention has increased, requiring a better understanding of the genetic alterations of tumors. By combining genomic and transcriptomic data, researchers can better understand the underlying mechanisms of the disease. Transcriptomic profiling involves sequencing transcribed RNAs to understand gene expression, which can lead to the identification of biomarkers for disease progression and aid in the discovery of treatment pathways. Studies analyzing transcriptomes, which encompass transcripts from both cancerous and non-cancerous cells, can be conducted through data analysis. In this work, we propose a pipeline for the transcriptomic analysis of gene expression data obtained from The Cancer Genome Atlas (TCGA) and use machine learning algorithms to classify the data into solid primary tissue, early-stage tumor, and metastasis. The data is pre-processed and normalized using the TCGAbiolinks package, which integrates TCGA data into a unified Bioconductor library, running on R software. TCGAbiolinks prepares the data using pre-processing strategies, allowing users to download a specific version of the data. For the analysis, it was necessary to transpose the datasets so that genes were represented by columns and individuals by rows. We executed seven classification models, with Random Forest and Radial Basis Function Support Vector Machine (RBF SVM) standing out, demonstrating superior performance with accuracy above 98%. To validate the classification models, cross-validation was used. At the end of this analysis, performance metrics (accuracy, precision, recall, and F1-score) were calculated for each iteration, and the average of these values was used to assess the overall performance of the models. Additionally, we applied a feature selector to the best-performing models to identify the most relevant variables in the dataset.