Recent advances in various technologies that perform measurements at the molecular level have generated large volumes of gene expression levels (hundreds of thousands), pushing in the last decade the development of various techniques for analysis. Genes have a measurable activity by observations in a number of time instants, thus constituting a time series called expression profile. A hot topic is the reconstruction of the network of relationships between a large number of genes from these long time series, discovering the gene regulatory network for gene underlying data. This project proposes the development of new models, algorithms and computational tools based on neural networks, supervised learning and unsupervised, for mining relationships between multiple data series with temporal evolution. Particularly, it is intended to apply these models and algorithms for the analysis and discovery of previously unknown relationships among large numbers of genes, which may allow inference network discovery and regulatory genes underlying data. Unsupervised neural networks could be used for appropriate preprocessing of the data, by grouping of genes having similar behavior. Then supervised models to model gene networks be used, observing the activity of a gene pair in a given number of instants of time. This would require training them with these data as time series, that is, to predict the time profile (regulation) of a gene from the temporal profile of a candidate for possible regulatory gene, with efficient and fast algorithms. It has artificial and real data used in the literature in recent years, and it will be used data emerging from the collaboration with biologists. Among the results is expected that the developed models can provide an important contribution to current treatment and analysis of large volumes of data with temporal dynamics, not only within the Bioinformatics area, but also for the treatment of various Big Data problems that have temporal evolution in other application domains.
CAID 2016 UNL