Vilas Distinguished Achievement Professor David Page

Congratulations to Professor David Page who has been named a Vilas Distinguished Achievement Professor. This appointment was made by Provost Sarah Mangelsdorf after a nomination by Dean Robert Golden. Provost Mangelsdort, Dean Golden and Dean Moss were instrumental in providing approval and support for this appointment.

The Vilas Distinguished Achievement Professorships recognize professors whose distinguished scholarship has advanced the confines of knowledge, and whose excellence has also included teaching or service. The professorship provides a fixed allocation of flexible funds to be used by David over the next five years. David will carry the title of Vilas Distinguished Achievement Professor for the duration of his career at UW-Madison.

Professor Rick Chappell an Elected Fellow of the Society of Clinical Trials

Congratulations to Professor Rick Chappell for his election as a Fellow of the Society of Clinical Trials, an organization which Rick has previously served as President. Fellows are Society members who have made significant contributions to the advancement of clinical trials and to the Society. Rick was elected for "important contributions to the statistical methodology for the design and analysis of clinical trials, particularly in cancer and aging; for leadership in clinical trial coordination and conduct at the University of Wisconsin Comprehensive Cancer Center; for national leadership on numerous DSMBs and advisory committees, including NIH and FDA; and for distinguished service to the Society."

Congratulations, Rick!

BMI faculty presented at Morgridge Symposium

BMI faculty to present at Morgridge Symposium
The BMI Department was a participating sponsor in a symposium entitled "When is an algorithm a medical device?", primarily organized and sponsored by the Morgridge Institute for Research. From the announcement materials, the "symposium will provide information on the current and potential regulatory framework for medical software development, guidelines for identifying when software becomes a medical device, and guidance on how to integrate the required practices into biomedical research."
Video are now available here:
BMI faculty members David DeMets and David Page are among the panelist who will present their work in the half-day symposium. 
More information is available here:

The Center for Predictive Computational Phenotyping (CPCP) at UW-Madison

Transcription-based cellular phenotyping

The University of Wisconsin, in a partnership with the Morgridge Institute for Research and the
Marshfield Clinic Research Foundation, has received a grant from the NIH to establish the Center
for Predictive Computational Phenotyping (CPCP), which is one of the NIH's
new Centers of Excellence for Big Data Computing in the Biomedical Sciences.
The grant will provide nearly $11 million over a four-year period.

The Center for Predictive Computational Phenotyping will develop innovative computational and statistical
methods and software for a broad range of problems that can be cast as computational phenotyping.
The term phenotype, which is derived from the Greek word phainein meaning‚ "to show"
refers to the observable properties of an organism that result from the interaction of its genotype
and its environment. Some phenotypes are easily measured and interpreted, and are available in an accessible
format. However, a wide range of scientifically and clinically important phenotypes do not satisfy these
criteria. In such cases, computational phenotyping methods are required either to extract a relevant phenotype
from a complex data source or collection of heterogeneous data sources, and to predict clinically important
phenotypes before they are exhibited. The Center will have a particular focus on screening for breast cancer
and Alzheimer's disease, and it will investigate how to exploit a wide array of data types for these tasks,
including molecular profiles, medical images, electronic health records, and population-level data. The Center
will also provide training in biomedical Big Data analysis to scientists and clinicians, and it will
investigate the bioethical issues surrounding the technology being developed.

The director of the Center is Mark Craven and the associate director is Michael Newton, both of whom are
professors in the Department of Biostatistics & Medical Informatics.

BMI student and faculty share in computer security award

Privacy in Pharmacogenetics: An End-to-End Case Study of Personalized Warfarin Dosing

BMI Student Eric Lantz, his advisor Professor David Page, and their colleagues Matthew Fredrikson, Somesh Jha, Simon Lin, and Thomas Ristenpart have won the best paper award at USENIX Security, a major computer security conference held August 20th - 22nd, 2014, in San Diego. The paper, “Privacy in Pharmacogenetics: An End-to-End Case Study of Personalized Warfarin Dosing,” Fredrikson, Jha and Ristenpart are in Computer Sciences. Lin is at Nationwide Children's Hospital in Columbus, Ohio.

The paper demonstrates that publishing a predictive model for personalized medicine or pharmacogenetics could in some cases reveal private information about patients on whose data the model was trained, via a "model inversion attack." It also shows that while differential privacy can protect against such an attack, for current typical data set sizes if the privacy level is elevated high enough to provide meaningful privacy protection then the trained predictive model is of little or no utility.


Sushmita Roy obtains NSF CAREER Award for Regulatory Networks

Reconstructing the evolution of regulatory networks

Assistant Professor Sushmita Roy has obtained a prestigious NSF CAREER award to pursue an innovative project entitled "Comparative Network Biology to Study the Evolution of Regulatory Networks".

Central to how living cells accomplish diverse biological functions are regulatory networks that control what genes need to be activated under different environmental conditions. As evolution is the ultimate tinkerer of living systems systematic comparisons of how regulatory networks evolve to drive species-specific differences are critical to understand cellular functions. Through advances in genomics, it is now possible to measure the activity levels of almost all genes for many species. This provides a unique opportunity to systematically compare these gene activity levels across multiple organisms and link changes in activity to changes in the networks of individual species. However, this is challenging because, first, such comparisons require the regulatory networks to be known in not one but multiple species including those that are poorly characterized, and second, computational methods to compare molecular datasets across species other than DNA sequence are in their infancy. This project will address these challenges by developing novel computational methods to identify and compare regulatory networks across multiple species, and correlate regulatory network patterns of divergence to phenotypic changes.

Yajuan Si has joined BMI and PHS

The Departments of Biostatistics & Medical Informatics and of Population Health Sciences are pleased to welcome new Assistant Professor Yajuan Si, PhD. Yajuan is joining us after completing her doctoral work in Statistical Science at Duke University and a post-doctoral fellowship at Columbia University under the direction of Professor Andrew Gelman. Her research interests are centered around Bayesian statistics and include latent variable models, complex survey inference methods, causal inference and post-stratification, and data confidentiality protection. Welcome Yajuan!

It has historically been a challenge to perform Bayesian inference in a design-based survey context. Yet, classical analysis is not robust especially for subgroup estimation. Model-based estimates are subject to bias under misspecified model assumptions. To address these problems, Yajuan develops a nonparametric Gaussian process regression of model-based survey inference and a unified framework under multilevel regression and post-stratification in realistic settings with survey weights.

Lisa Gress, MS, joins BMI.

The BMI Department, together with the UW NIH CTSA-funded Institute for Clinical and Translational Research, is pleased to welcome Lisa Gress, MS, to the ICTR Biomedical Informatics core. Lisa will work with Biomedical Informatics core faculty--in particular Professors Mark Craven, Eneida Mendonca, and David Page--and other UW investigators on exploiting the UW electronic health record in order to advance clinical research. Lisa completed her BA degree at Kenyon College, and her MS Degree in Computational Linguistics at University of Washington, Seattle, in 2013. She brings her expertise in natural language processing to problems of exploiting free text in the EHR.

Anthony Gitter joins BMI and MIR

These six medulloblastoma tumor samples all belong to the WNT subtype and exhibit functional similarities, but their mutations are quite heterogeneous. The multi-sample prize-collecting Steiner forest (Multi-PCSF) algorithm infers pathways that may be commonly perturbed by the diverse mutations. It searches a protein-protein interaction network to discover hidden, unmutated genes (circles) that connect the patient-specific mutated genes (squares) through physical interactions (figure by Tobias Ehrenberger).

The Department, together with the Morgridge Institute for Research (MIR), is very pleased to welcome Anthony Gitter, PhD, as a new Assistant Professor at UW and Investigator at MIR. Anthony completed his PhD in Computer Science at Carnegie Mellon University and a post-doctoral fellowship at Microsoft Research and MIT in Cambridge, Massachusetts. His research interests are centered in computational biology, and include probabilistic graphical models, host response to viral infection, and cancer genomics, among others. He will be joining Prof. Paul Ahlquist's virology research group at MIR.

In a current project, large-scale cancer genomics studies have revealed the diversity of mutations across tumors, and shown that many mutations are unique to a single tumor. Anthony develops network algorithms to discover how heterogeneous mutations may disrupt the same signaling pathways to produce similar effects on cellular state.

Associate Professor Sijian Wang has obtained new R01 methodology funding

Illustration of the performance of proposed regularized mixture Cox regression on TCGA ovarian cancer dataset. The motivation of the regularized mixture Cox regression is to address the possible heterogeneity in data.  Specifically, it automatically divide the whole population into several clusters. Simultaneously, it fits a (different) Cox model in each cluster respectively.  As shown in the figure, our regularized mixture Cox regression divides patients into three clusters: High-Risk” cluster, “Medium-

We are pleased to announce that Associate Professor Sijian Wang has obtained new R01 funding for his project entitled "Heterogeneous and Robust Survival Analysis in Genomic Studies". The four year project aims to develop, evaluate, and disseminate powerful and computationally-efficient statistical methods to model the heterogeneity in both patients and biomarkers in genomic studies, and to enhance the ability of biomarker detection and prediction power. It is funded by the National Human Genome Research Institute (NHGRI).

About the figure: Illustration of the performance of our proposed regularized mixture Cox regression on TCGA ovarian cancer dataset. The motivation of the regularized mixture Cox regression is to address the possible heterogeneity in data. Specifically, it automatically divide the whole population into several clusters. Simultaneously, it fits a (different) Cox model in each cluster respectively. As shown in the figure, our regularized mixture Cox regression divides patients into three clusters: High-Risk” cluster, “Medium-Risk” cluster and “Low-Risk” cluster. The left panel shows the estimated smoothed hazard function for each cluster. The middle panel shows the Kaplan-Meier curves of three clusters, which are well separated. The right panel shows the estimated regression coefficients of twelve patient-specific index scores (PSRPs) derived from gene expression data. Clearly, the estimated coefficients vary much among three clusters, indicating our hypothesized heterogeneity that different genes play roles in different clusters (patients).