Ferran Reverter

The main areas of my scientific activity are: 1. Development and implementation of statistical tools for data analysis in Biosciences, 2. Enhancing kernel-based methods to increase its interpretability, 3. Implementation of image analysis tools for wide range of applications (Health,Digital humanities, ...). I participated as a postdoctoral researcher at the Center for Genomic Regulation (CRG, http://www.crg.eu/) in the research program of Bioinformatics and Genomics at the Barcelona Biomedical Research Park (PRBB, https://www.prbb.org/), a center that leads biomedical translational research in Southern Europe. In particular, I was working in the Computational Biology of RNA Processing group. Group lead by Dr. Roderic Guigó Serra. My background in Biology and Statistics, has been key in integration into a competitive research group where the needs of statistical tools and intensive computing are growing. Currently, I am collaborating in four projects: 1) The Transcriptional Landscape of Repetitive Elements in Human Tissues. 2) Whole genome diagnosis of coronary artery disease (project funded by La Marató de TV3 Foundation). 3) Histology and pathology phenotypes (GTEx project) 4) sQTLseekeR (extension). A method for detecting associations of genetic variants with the vectors of isoform splicing ratios of genes (sQTL). I am participating in the Genotype and Tissue Expression project (GTEx). The GTEx project aims to provide to the scientific community a resource with which to study human gene expression and regulation and its relationship to genetic variation. The multidimensional character of the GTEx project; dozens of tissues, hundreds of individuals, thousands of genes and millions of snps, and multiple diseases, requires robust techniques to analyze data in all its complexity, avoiding the risk of overfitting and simultaneously releasing sparse solutions that produce a reduced list of biomarkers that facilitate the biomedical knowledge. My vocation for Statistics and the Data Analysis starts in the Department of Statistics of the UB to which I have been connected since the 1994. Training in mathematical statistics derived from a Postgraduate Diploma in Data Analysis and Statistics (UPC, 1994) and enlarged later during my PhD studies and the doctoral thesis (UB, 2000) has allowed me to make a more methodological research within various competitive groups: Multivariate Geometric Models with Applications to Bioinformatics and Biomedicine (MTM2008-00642, 2008-2011), Group of Multivariate and Computational Statistical Modeling (2009-SGR-932, 2009-2012). I did the PhD research in the field of Differential Geometry methods in Statistics. PhD Thesis, entitled: Métodos computacionales para el cálculo de distancias Riemannianas. Aplicaciones a la geometría informativa correspondiente a la distribución Gamma. University of Barcelona, 2000. Under the project: Multivariate Geometric Models with Applications to Bioinformatics and Biomedicine I did a 3-month stay at the Laboratoire de Statistisque et Probabilités. Institut de Mathématiques de Toulouse. Toulouse, 2008. We investigated the integration of microarray and biochemical data using Kernel Canonical Correlation Analysis, and we compare its performance with Regularized Canonical Correlation Analysis. In order to design data analysis methods, the first question to be addressed is how to represent the data set for further processing. The vast majority of data analysis methods, outside kernel methods, have a natural answer to this question: first define a representation for each object, and then represent the set of objects by the set of their representations. Kernel methods are based on a radically different answer to the question of data representation. Data are not represented individually anymore, but only through a representation set of pairwise comparisons. Although these methods have provided valuable models for data analysis (high predictive performance, ability to integrate different types of data) the resulting models have scarce interpretability in terms of the original variables. Therefore, it is not easy to measure the importance of the variables in the final model, so the interpretation of that model is difficult. Accordingly, a line of research that we are developing consists to produce new methods that preserve the advantages of kernel-methods without losing the interpretability of the estimated models. I was the responsible of the pre-competitive project, entitled: Cataloguing work of artist based on the automatic image annotation incorporating semantic content. VR de Política Docent i Científica. UB. 2011-2012.