BIOINFOMATICS RESEARCH
Our bioinformatics research program is truly multi-disciplinary and involves collaborations Dr. Aidong Zhang from the Department of Computer Science and Engineering and Dr. Alan Forrest from the Department of Pharmacy Practice. The bioinformatics research program is focused on the following areas:

  • Integration of clinical and genomic data
  • Biomarker optimization
  • Multi-dimensional visualization
  • Modeling genomics and proteomics time series
Biomarker Optimization
The biomarker optimization program is focused on innovative approaches to identify informative genetic, genomic and proteomics markers for patient care. Good biomarkers are important in drug discovery and for individualizing treatment regimens. We have developed techniques that involve supervised and unsupervised approaches:
  • Improved metrics for time series analysis, e.g. Kullback-Leibler divergence
  • Improved predictor construction, e.g. Markowitz method for creating optimal portfolios of genes for treatment effect
  • Empirical sample pattern detection
  • 3-D clustering: Gene-Sample-Time data
Supervised approaches are usually more appropriate for analyzing key outcomes and hypothesis testing in well-designed, experimental settings whereas unsupervised approaches are more appropriate for data mining and hypothesis generation.

Visualization of Genomic Data
Visualization can also provide effective tools to summarize and interpret data sets, describe the contents, and expose features in time series data. The key challenges in genomic data visualization are the nature of the data loss that occurs upon mapping to 2-D and understanding how to interpret the 2-D mapping to infer the relationships between the data points in the N-dimensional space. The currently available approaches are: i) the parallel coordinates approach, wherein the data along each dimension is plotted along a separate axis, ii) the multi-dimensional scaling (MDS) approach, in which the presentation in 2-dimensions (2-D) is optimized to preserve a specific aspect of the relationship, e.g., the Euclidean distance, block distance or rank relationships between the points in the N-dimensional space. In many respects, MDS is the current gold standard for multi-dimensional visualization. Visualization has not been extensively investigated in the context of genomic data analysis. We have focus on developing our approach, VizStruct, which is motivated by radial visualization techniques such as Radviz, for visualizing genomic time-series. VizStruct offers substantive and unique advantages over both competing methods such as Sammon's mapping, multi-dimensional scaling and parallel coordinates.

Murali Ramanathan, PhD

Associate Professor of
Pharmaceutical Sciences
and Neurology

543 Cooke Hall
Dept. of Pharmaceutical Sciences
State University of New York
Buffalo, NY 14260

Ph:  (716)645-2842 ext 242
Fax: (716)645-3693

E-mail: murali@buffalo.edu

Publications