UNIVERSITY OF MALTA LIFE SCIENCE RESEARCH SEMINARS Web: http://www.um.edu.mt/events/scisem/ Email: scisem@um.edu.mt Abstract form Title: Measures to characterize variable importance in classification of microarray datasets Presenter: Contact address: Tel: Fax: Email: Presentation date: Clint Mizzi 99905912 mizziclint@gmail.com 7 November 2011 Abstract In recent years the field of molecular biology has evolved rapidly thanks to a number of significant projects including the human genome projects. This research was only possible with the help of computational techniques, which have opened the doors to a new interdisciplinary field named bioinformatics. This specialisation is well integrated in molecular biology and is used in solving problems related to genome analysis, functional prediction and protein prediction. Classification algorithms have been used in different bioinformatics fields including gene expression microarray datasets. One of the challenges regarding this area is identifying feature importance by, determining the most relevant attributes in a classifier. The aim of this project is to study the use of feature selection methods as a solution to identify the most important genes. Different cancer datasets were used to compare and contrast different classification and feature selection algorithms, including the new random jungle methodology. This project used Classification and Feature Selection techniques to generate gene expression networks. The aim of this report is to present a novel algorithm, which improves on GENIE3 by using Correlation Feature Selection Subset Evaluator to generate gene expression networks. The networks created were analysed for biological meaning.