Automatic design of decision-tree induction algorithms for detecting patterns of functional gene-gene interactions in observational gene expression data

• Grant number: 12/51416-0
• Support Opportunities: Regular Research Grants
• Start date: February 01, 2013
• End date: January 31, 2015
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Agreement: BAYLAT/StMBW - Bavarian Academic Center for Latin America and Bavarian State Ministry of Science and the Arts
• Principal Investigator: Márcio Porto Basgalupp
• Grantee: Márcio Porto Basgalupp
• Principal researcher abroad: Rainer Spang
• Institution abroad: University of Regensburg, Germany
• Host Institution: Instituto de Ciência e Tecnologia (ICT). Universidade Federal de São Paulo (UNIFESP). Campus São José dos Campos. São José dos Campos , SP, Brazil

Abstract

Predicting an organism's or a cell's response to a therapeutical intervention lies at the heart of medical research. In a functional cellular experiment, one can intervene in the expression of a gene x using RNA interference technologies and collect the downstream effects in the entire transcriptome using high throughput technologies like microarrays or RNAseq. This experiment gives us a class of genes that respond to the intervention in x (class A) and a class of genes that do not (class B). We hypothesize that also observational data that does not involve perturbations of gene x holds information whether a gene is in class A or B, since the biological mechanisms that drive the expression of these classes of genes must be different. This yields a standard classification problem: Predict whether a gene belongs to a class A or B by considering only observational data. In this project, we will investigate machine learning-based solutions to this problem by evolving decision tree induction algorithms. Decision tree induction is one of the most employed methods to extract knowledge from data, since the representation of knowledge is very intuitive and easily understandable by humans. The most successful strategy for inducing decision trees, the greedy top-down approach has been continuously improved by researchers over the years. This work, following recent breakthroughs in the automatic design of machine learning algorithms, proposes a hyper-heuristic evolutionary algorithm for automatically generating decision-tree induction algorithms, named HEAD-DT. We aim to automatically designing a decision tree induction algorithm tailored to a specific domain: Detecting functional gene-gene interactions from observational gene expression data. If successful, both the designed algorithm and their induced classifiers could be seen as a first step to develop a platform for "virtual intervention experiments" that could be used to prioritize genes for further biological experimental studies. (AU)