Please use this identifier to cite or link to this item:
Title: Estudio comparativo de modelos de machine learning para la detección de dianas microARN
Author: Martinez Rodriguez, Jordi
Director: Morán Moreno, José Antonio
Ventura Royo, Carles  
Tutor: Pla Planas, Albert
Others: Universitat Oberta de Catalunya
Keywords: microRNA
machine learning
prediction model
Issue Date: Jan-2018
Publisher: Universitat Oberta de Catalunya
Abstract: MicroRNAs (miRNA) are small, non-coding RNA molecules of an approximate length of 23 nucleotides, which are involved in post-transcriptional regulation by inhibiting genetic expression by binding to specific target mRNA sites. The present document evaluates the viability of different existing machine learning models for the prediction of miRNA targets. The viability of six different models was examined: k-nearest neighbors, Naive Bayes, artificial neural network, support vector machine, decision tree and random forest. For model training, a database was built with more than 14000 miRNA-mRNA interactions with 19 features, from which 3173 real and 3173 unreal interactions were selected. The random forest model showed the highest efficiency in the prediction of interactions, and was as well one of the most solid and simple to train. Considering the characteristics of models based in random forest and due to their ability to use a great number of features for miRNA target predictions, these models are recommended for future studies of detection and prediction of miRNA targets.
Language: Spanish
Appears in Collections:Bachelor thesis, research projects, etc.


This item is licensed under a Creative Commons License Creative Commons