2008-09 Research Projects

In 2008-09, four collaborative research proposals have been funded that focus on:

• Data mining methods of clinical data
• Machine learning to predict disease state
• Computational methods for rational drug design

The following projects are being supported by the University of Minnesota Rochester Biomedical Informatics and Computational Biology Program through its University of Minnesota/Mayo/IBM Collaboration Seed Grant Program:

1. Reliable Biomarkers for Prediction of Transplant-Related Mortality
2. Identification of Small-Molecule Viral Replication Inhibitors
3. Cancer Drug Target Discovery through Computational Prediction of Genetic Interactions
4. New Models for Cancer Targets from Clinical Data

Reliable Biomarkers for Prediction of Transplant-Related Mortality

This project will investigate application of predictive data-analytic modeling to diagnosis of Graft-versus-Host Disease (GVHD) which is a common side effect of an allogenic bone marrow or cord blood transplant. The disease involves about 50% of transplant recipients and is the leading cause of early (acute GVHD) and late (chronic GVHD) mortality post transplant. GVHD occurs because some of donor's immune cells (known as T cells) attack cells in recipient’s body. Hence, there is a need to analyze the association of genetic variants of donor and transplant recipient genes with acute and chronic graft versus host disease. Additional clinical and demographic factors that may impact the outcome (of successful transplants) include: recipient and donor age, type of donor (related/unrelated), donor-recipient gender mismatch, disease status prior to transplant, stem cell source (peripheral blood vs marrow), etc. This project will incorporate both genomic and non-genomic information into a predictive diagnostic model for predicting transplant-related mortality. As the task of utilizing clinical, genomic and demographic data is generic for biomedical applications, this study will lead to improved understanding of machine learning approaches appropriate for this task. 

This project will investigate application of machine learning methodologies to predicting transplant-related mortality using two clinical data sets of donor/recipient transplant patients from the pilot studies conducted at the University of Minnesota and at the Mayo Clinic. Proposed data-modeling approaches include existing methods such as statistical subset selection, Support Vector Machines (SVM), and emerging newlearning approaches such as SVM+, and multi-task learning, which are appropriate for sparse heterogeneous data.

The outcome of this project will lead to identification of genetic variants that will help to identify high-risk groups for GVHD and transplant-related mortality. Moreover, proposed advanced learning methodologies are expected to result in innovative diagnostic techniques that integrate patients’ data from different modalities.

Investigators

• Vladimir Cherkassky, U of M
• Daniel Weisdorf ,U of M
• Mukta Arora, U of M
• Brian Van Ness, U of M
• Mark Litzow, Mayo Clinic
• Walter Kremers, Mayo Clinic
• Brooke. L Fridley, Mayo Clinic

(back to top)

Identification of Small-Molecule Viral Replication Inhibitors

Viral infections are a significant cause of morbidity and mortality throughout the world. While many small molecule inhibitors of viral nucleic acid replication and assembly have been successfully developed, targets in other stages of the life-cycle remain to be exploited. Our proposal encompasses a tripartite approach to discovery of viral replication inhibitors:

1. The application of current in silico docking methodologies to the discovery of human immunodeficiency virus (HIV) and Ebola virus replication inhibitors.
2. Evaluation and validation of in silico results through laboratory based in vitro high-throughput screening.
3. Improving in silico docking by theoretical research leading to refinement in docking algorithms.

This work combines expertise in theory, computation and laboratory science to comprehensively approach the identification and development of viral replication inhibitors. Concurrently, this project creates a foundation for the interdisciplinary collaboration necessary to fundamentally improve drug discovery.

Recent developments in genomic and proteomic technologies have enabled high-resolution measurement of a wide variety of cellular phenomena from DNA sequences to interactions between proteins. A recurring theme from these studies is that most cellular functions are supported by a complex network of interactions that yield a surprising amount of robustness and plasticity in cellular systems. This network-level view of the cell is also informing our understanding of human disease. Recent studies suggest that many diseases, including cancer, result from negative interactions among multiple genes that transform the genetic network into a perturbed “disease state. ”Developing effective cancer treatments requires a holistic understanding of these disease-specific networks.

Investigators

• Yiannis Kaznessis, U of M
• David Katzmann, Mayo Clinic
• Jean-Pierre Kocher, Mayo Clinic
• Eric Poeschla, Mayo Clinic
• Carlos Sosa, IBM

(back to top)

Cancer Drug Target Discovery through Computational Prediction of Genetic Interactions

The proposed research addresses this challenge in the context of cancer with the development of computational strategies for integrating diverse human genomic data to infer perturbed sub-networks responsible for disease. The proposed models will integrate gene expression, protein-protein interactions, sequence and copy number polymorphism data, and existing knowledge of pathways to perform differential network analysis between tumor and normal states. Putative models of the disease-specific network underlying specific cancers can enable the prediction of high-specificity drug targets. Disease-specific networks both give rise to destructive phenotypes (e. g. uncontrolled cell growth) and also present unique vulnerabilities. For example, a second perturbation in the context of the disease state might be lethal while the same perturbation is harmless in normal cells.

This proposal suggests a machine learning approach to predicting such disease-specific vulnerabilities by leveraging large-scale genetic interaction studies in model organisms. A critical step in assessing the clinical relevance of disease network hypotheses and predicted drug targets is experimental validation in the context of cancer. The proposed work couples the computational predictions with in vitro and in vivo tumor models in the zebrafish, which will be used for model validation and iterative refinement.

Investigators

• Chad Myers, U of M
• Dennis Wigle, Mayo Clinic

(back to top)

New Models for Cancer Targets from Clinical Data

The purpose of the proposed work is to demonstrate proof-of-concept for a new paradigm utilizing theoretical models to connect clinical/experimental data to structure-function relationships, and provide guidelines for drug design. The molecular target is the p53 tumor-suppressor protein that has been implicated in cancer. Clinical data available at the Mayo Clinic and corresponding mutation data measured at the Hormel Institute will be collected and analyzed. Mutations with high correlations with cancer phenotypes will be selected as the molecular simulation targets. Simulations using new methods developed at UMTC will be performed on the BlueGene at Hormel and IBM-Rochester with parallel codes optimized at IBM. Conformational events will be studied with long-time classical molecular dynamics simulations, and used as a starting point for detailed investigation into biochemical mechanisms explored with recently developed combined quantum mechanical/molecular mechanical (QM/MM) models. X-ray structure determination at Hormel will provide valuable structural information that will serve as a starting point for theoretical investigations and facilitate validation of the simulations. Biochemical assays will provide further information that, together with the theoretical simulations, will provide insight into the origin of the mutational effects.

Investigators

• Darrin York, U of M
• Tai-Sung Lee, U of M
• Zigang Dong, Hormel Institute
• Ann Bode, Hormel Institute
• Paul Limburg, Mayo Clinic
• Carlos Sosa, IBM

(back to top)