Research Abstracts

Feng Cai, U of M

Vladimir Cherkassky, U of M

Daniel Weisdorf, U of M

Mukta Arora, U of M

Brian Van Ness, U of M

Bharat Thyagarajan, U of M

 

We describe application of machine learning approaches for predictive modeling to improve the estimation of risks for complications of allogeneic hematopoietic cell transplantation (HCT) including relapse, Graft-versus-Host Disease (GVHD) and transplant-related mortality (TRM). Clinical disease and demographic factors known to impact the outcome of HCT include: recipient and donor age, type of donor (related/unrelated), donor-recipient gender, diagnosis and disease status pre-HCT, stem cell source (peripheral blood, marrow, umbilical cord blood). However, biostatistical analysis of risk has only limited accuracy in estimating a given patient’s risks of serious post-HCT complications. We describe application of standard SVM classifiers for data-analytic modeling of TRM. The goal is to predict the binary output TRM (alive or dead) from a set of genetic, demographic and clinical inputs. Classification decision rule is estimated using Support Vector Machine (SVM) approach appropriate for such sparse multivariate data. In addition, this study compares the quality of several feature selection techniques for modeling TRM. Finally, we discuss methods for interpretation of high-dimensional SVM models.

Jean-Pierre Kocher, Mayo Clinic

Carlos Sosa, IBM / U of M

Kendall Byler, Mayo Clinic

Andrew Norgan, Mayo Clinic / U of M

Emilia Wu, U of M

Eric Poeschla, Mayo Clinic

Yiannis Kaznessis, U of M

David Katzmann, Mayo Clinic

 

Viral replication relies on the ability of the virus to co-opt a large number of host proteins. This work focuses on the HIV enzyme integrase and its interactions with a host cellular factor p75/LEDGF. Integrase mediates the integration of the HIV genome into the genomic DNA of the host cell, and is a process vital to both active replication of HIV and the establishment of viral latency (a hallmark of HIV that makes it difficult to cure). The enzymatic function of integrase has previously been targeted for disruption by small molecules, resulting in the successful development of the integrase inhibitors raltegravir and elvitegravir. Another potential mechanism to disrupt integrase function is to prevent its interaction with LEDGF, a cellular factor that facilitates its activity. We are approaching this problem using virtual molecular docking to screen diverse ligand libraries for inhibitors of the IN-LEDGF interaction. To validate and refine the screening approach, the results of a subset of the virtual screening are being compared to experimental results from an identical ligand library. A secondary focus of our research has been to address the problem of induced protonation state changes in virtual screening. For a number of proteins, including the HIV enzymes protease and integrase, it is apparent that small molecule binding may induce changes in the pKa and ultimately protonation state of amino acid side chains, altering the electronic and steric environment of the binding pocket. Molecular docking programs rarely account for these changes, and as such may not correctly score or rank ligands. Our group is examining fast pKa prediction methods, with a long-term focus on improving scoring by incorporating pKa modeling.

Chad Myers, U of M

Dennis Wigle, Mayo Clinic

 

Genetic and physical interactions among genes or proteins are central to most processes that support cellular function.   One consequence of this network organization is that many cell or organism-level phenotypes are caused by simultaneous variation in multiple genes. Multigenic phenotypes are behind many common diseases like heart disease and cancer, and their multiple-locus origins often make study and treatment difficult. However, genetic interactions may also hold the key to treatment of complex diseases like cancer. For example, one type of genetic interaction that has been observed commonly in model organisms is where a mutation in a gene A has no effect by itself, but when mutated in combination with another gene, B, this mutation kills the cell ("synthetic lethality"). We hypothesize that understanding this phenomenon may offer a promising approach to treating cancer. For instance, finding a secondary mutation that is lethal in combination with a tumor-associated mutation would be a good candidate for a drug target- inhibiting the function of that gene would be harmful only to cells harboring the cancer-specific mutation.

 

To better understand these types of genetic interactions, we are mining high-throughput experimental data from lower model organisms where millions of combinations of genes have been mutated to assess their potential for genetic interaction. We discuss our progress in this direction based on a large genetic interaction network (~5 million double mutants) from the yeast model system Saccharomyces cerevisiae. We also describe our computational strategy for translating information about genetic interactions across species, including interactions with genes associated with cancer. Finally, we will describe our efforts to validate these genetic interactions in higher order model systems and human cells.

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

 

Despite its biological importance, the constitutive activation mechanisms of p53 are still far from clear. Very recently a crystal structure of the complex of a p53 tetramer and a DNA segment was resolved that provides an excellent departure point for molecular simulations. Most cancer-related p53 mutations are in the p53 DNA binding domain, but most of these sites do not directly contact the DNA. Similarly, the phosphorylation of S215 inactivates p53, although S215 is distant from the DNA binding region. These observations are intriguing and raise the question as to what is the relationship between mutations, phosphorylations and p53-DNA binding.

 

We have performed long-time molecular dynamics simulations on the complex of the p53 DNA binding domain tetramer with DNA, as the first steps to understand the binding and possible activation mechanism due to phosphorylation of S215. Simulation results indicate the effect of S215 phosphorylation is most prominent at the interface of the two p53 monomers, and suggest that electrostatic interactions at the monomer-monomer interface may play an important role in DNA binding. Characterization of the structure and dynamics in the presence and absence of S215 phosphorylation provides insight into the mechanism of inactivation of p53.