Machine Learning: aplicación a datos RNA-Seq

Abstract

Over the past years, biomedicine has experienced a revolution. This is partly due to the very high volume of information existing today that has been produced by both clinical and pharmacological studies as well as Omics data generation. At the same time, the development of statistical and computer techniques that allow its analysis has led to a new area of knowledge: Bioinformatics. The main aim of this study is to compare different Machine Learning techniques applied to a set of genetic data that were obtained through the RNA-Seq method. This method sequences cDNA in a massive way through a high-performance platform in order to gather global information about the DNA of a sample. Due to the efficacy, reproducibility and performance of the high-throughput sequencing, the RNA-Seq method allows researchers to measure the level of gene expression, detect alternative splicing, mutations, etc. It is also included in this study the theoretic and practical description of several supervised learning methods such Classification Trees, Bagging, Random Forest and Boosting. The efficacy of these methods has been measured by different rates that can be subsequently compared. The said rates are: sensitivity, specificity, area under the ROC curve, global hit rate, positive predictive value (PPV), negative predictive value (NPV) and balance accuracy. This study is complemented by the implementation of R software. It presents different graphs (bar charts, box-andwhisker plot, etc.) as well as numerical data of the obtained predictions for each of the methods.