Brown University Center for Statistical Sciences Seminar
Abstract: HIV and other sexually transmitted infections (STI), the potential for epidemic spread depends on connectivity in the underlying sexual network. Since most people have few partners, traditional STI theory and prevention strategies have focused on the role of "core groups" and "superspreaders" in establishing the connectivity needed to sustain transmission. This has been a successful approach for curable STI in low prevalence settings, but is not as effective in explaining or intervening in a generalized epidemic, as HIV has become in many sub-Saharan African countries. In generalized epidemics, transmission is well established in the broad group of less active persons. We show how network connectivity can emerge in such low risk networks. A central role is played by concurrent partnerships - partnerships that overlap in time - rather than high rates of partner change. This has direct implications for HIV and STI prevention. One is that a small change in behavior may be quite effective in this context, making behavior change a more viable prevention target than many believe. Another is that the right behavior change message is not to "have fewer partners," but, as Ugandan HIV messages put it 10 years ago, "zero grazing." We use a combination of empirical data (comparing Thailand, Uganda and the US, and racial groups in the US) and simulation to demonstrate these principles, and their relevance to the disparities in HIV prevalence, both worldwide and within the United States. Sponsored in part by the Herbert H. Goldberger Lectureships Fund And the Charles K. Colver Lectureship and Publication Fund
Applied Mathematics Colloquium
Abstract: "Compressed sensing" captures a new paradigm which connects sparse representations, high-dimensional geometry, probability, and algorithms. It suggests a new paradigm in information acquisition and processing of compressible signals. These signals can be approximated using an amount of information much smaller than the nominal dimension of the signal. Traditional approaches acquire the entire signal and process it to extract the information. The new approach acquires a small number of nonadaptive linear measurements of the signal and uses sophisticated algorithms to determine its information content. Emerging technologies can compute these general linear measurements of a signal at unit cost per measurement. I will discuss the connections to randomized algorithms and signal processing. In particular, I will focus on extremely fast algorithms and measurement designs for compressed sensing, including some prototype hardware designs.
Center for Statistical Sciences Seminar
Abstract: Motivation: Tumor gene expression profiles from microarray experiments, or "cancer signatures", provide valuable information in identifying tumor characteristics relevant to diagnosis and/or prognosis. Various statistical techniques have been developed for the discovery of cancer signatures. The profiles identified, however, are usually dependent on the techniques employed and difficult to replicate with different methods. Methods: Unsupervised analysis plays an important role in cancer signatures discovery by revealing "hidden" clusters that might be relevant to tumor risk structures. In this study, we first developed a method to quantify the statistical significance of stability measures in random projection, a data perturbation method that distinguishes in terms of both theoretical supporting and noise injection controlling. Then we established an approach, named "best representatives", to select the optimum clustering structures after data perturbation. Results: The results of clustering analyses on a tumor array dataset show that our method can not only detect unstable samples, but also correct mistakenly clustered samples. Furthermore, our method can also be used to detect outlier arrays and provide specific thresholds to QA/QC metrics for Affymetrix GeneChip data. Candidate for Assistant Professor (Research) in the Biostatistics Section of the Program in Public Health
Department of Mathematics Colloquium
Center for Statistical Sciences Seminar
Abstract: Copy number variation (CNV) has received increasing attention over the past few years, following findings published in 2004 showing significantly greater CNV content in humans than previously thought (Sebat et al, Iafrate et al). Comparative genomic hybridization microarrays ("CGH arrays") and high-density genotyping microarrays ("SNP arrays") can assay CNV on a genome-wide scale. This talk will describe (1) methods for inferring copy number state from this class of microarrays, (2) a discovery, using these technologies, of CNV between monozygotic twins as well as somatic copy number mosaicism (Bruder et al, 2008), and (3) remaining challenges in the analysis of CNV and CNV association. Candidate for Assistant Professor (Research) in the Biostatistics Section of the Program in Public Health
<--- 2008 Index