Associate Professor of Computer Science and Informatics
Adjunct Associate Professor of Statistics
150 South Woodlawn Avenue
Bloomington, IN 47405
Phone: (812) 856-1851
Fax: (812) 856-1995
Office: Lindley Hall 301F
Post-doctoral fellow, Bioinformatics, Indiana University School of Medicine, Indianapolis, Indiana, 2004
Ph.D., Computer and Information Sciences, Temple University, Philadelphia, Pennsylvania, 2003
M.S., Electrical Engineering, University of Belgrade, Serbia, 1997
B.S., Electrical Engineering, University of Novi Sad, Serbia, 1994
ISCB Board of Directors, 2012-
Editorial Board Member, Bioinformatics, Oxford University Press, 2010-
Guest Associate Editor, PLoS Computational Biology, 2013-
• Protein Bioinformatics
Methods for characterization and prediction of protein's structural and functional properties, both on a whole-molecule and residue level. This includes automated inference of protein molecular and cellular function or disease associations from its sequence/structure/interactions, as well as understanding post-translational modifications, protein-partner binding sites, etc. We are also interested in understanding the molecular basis of disease via studying amino acid substitutions causing or associated with disease and biochemical ways they lead to altered phenotypes. See our algorithms and software for probabilistically identifying disease-associated human genes (PhenoPred) and biochemical basis of disease given a mutation (MutPred).
• Computational Mass-Spectrometry Proteomics
Methods for peptide identification, protein identification and protein quantification from tandem mass spectrometry (MS/MS) data. Each peptide in a mixture of digested proteins can be is associated with a probability to be detected by a mass spectrometry platform (that includes sample preparation, separation, mass spectrometer and software for peptide-to-spectrum matching). We hypothesized that this property, called peptide detectability, can be successfully inferred from amino acid sequence of a peptide and its parent protein. We use peptide detectability to build algorithms for protein inference and label-free quantification. See our algorithms and software for protein identification from MS/MS data (MSBayesPro).
• Machine Learning and Data Mining
Classification methods: prediction from biased, noisy, high-dimensional, class-imbalanced, and heterogeneous data. These methods include feature selection algorithms, estimation, exploiting unlabeled data, etc. See our work involving development of kernel methods for vertex labeling in sparse graphs (Graphlet Kernels), applied to the domain of protein function.
Last modified: March 07, 2014