Técnicas de machine learning aplicadas al diagnóstico y tratamiento oncológico de precisión mediante el análisis de datos ómicos

Abstract

As sequencing costs have been dramatically reduced, an increasing amount of omics data have been generated to molecularly characterise cancer. Large consortiums are generating large amount of this data and making them publicly available. In addition, Machine Learning (ML) models offer a significant advantage extracting complex patterns from biomedical data. A study of their application in this field is necessary in order to obtain more robust and generalised results. This thesis studies the application of ML models to omics data analysis. Performing a review of previous work, certain limitations in terms of reproducibility and validation of the methodologies were identified. From this study, a set of guidelines for robust and reproducible ML analysis of omics data have been established, allowing to identify altered biomarkers and pathways in colon cancer patients, predict clinical conditions relevant to tumour development, and develop an automatic anti-tumour drug screening model. These results are presented as a compendium of three scientific manuscripts. In conclusion, this thesis provides a variety of computational approaches to improve diagnosis and precision oncological treatment