I am mostly interested in Machine Learning approaches for solving problems in systems biology. My main research goal is development of efficient machine learning algorithms for supervised learning of regulatory genetic networks from integrated heterogeneous data.

I work for a Developmental Retina group led by Dr. Vasant Honavar which involves a close collaboration with the neurobiology lab of Dr. Heather Greenlee. Our main focus is elucidation of a regulatory genetic network in developing retina. This includes discovering differentially expressed proteins, identifying regulatory mechanisms, such as positive/negative correlation or possible causal relationships between proteins, and building a protein interaction network by learning interactions from diverse data and validating predictions through wet-lab experiments.

Mouse is used as a model organism, and protein expression data are collected for proteins during various developmental stages. Clustering is used as a first step in data analysis. Because of the high-dimensionality of expression data and the combinatorial nature of genetic relationships, uncovering statistically significant relationships between proteins from expression data alone is challenging and likely to produce inaccurate results. Employing additional data, such as protein binding, Gene Ontology annotations, or protein phosphorylation data is one of the ways to improve the predictions. Using machine learning techniques to learn relationships between proteins (for example, physical interactions) from existing training data and predicting unknown protein interactions in a less-studied domain is another approach.

I am currently working on the development of a flexible interactive database quering system that allows to easily design and re-use custom-built queries on genetic networks, and includes several methods for constructing networks from expression data. This tool will provide a simple mechanism for storing and retrieving multi-source protein data and hypothetical interaction networks, facilitate an interactive iterative exploration and analysis of protein networks, and allow to predict and compare regulatory networks using a few well-known machine learning algorithms.

My other interests are in the areas of Developmental Robotics and Complex Adaptive Systems. In 2005, I worked with Dr. Alexander Stoychev on the development of a system that uses a neural network to create mapping between tool shapes and the possible affordances of these tools. In 2004, I worked with Dr. Dan Ashlock to study the effect on environmental stress on aftificial "bacteria" - simple agents who's actions are goverened by a genetically inherited "genome". See the projects page for more information.