BIOS Colloquium – Fall 2019

Fall 2019

Speaker: Chris Taylor
Affiliation: Department of Microbiology, Immunology and Parasitology, LSUHSC School of Medicine, New Orleans – LA
Title: Assessing the vaginal microbiome and interactions with sexually transmitted infections
Date and Time: August 26th, 2019, 3:00 pm
Room: LEC, Room 303
Abstract: Changes in the composition of the vaginal microbiome have been associated with sexually transmitted infections and reproductive health issues. I will present results from three clinical studies using 16S rRNA gene sequencing that examine sharing of microbiota among sexual partners, changes in the vaginal microbiota over time, and relationships with bacterial vaginosis and chlamydia trachomatis. 

Speaker: Yusha Liu
Affiliation: Statistics Department, Rice University, Houston – TX
Title: Function-on-Scalar Quantile Regression with Application to Mass Spectrometry Proteomics Data
Date and Time: September 16, 2019, 3:00 pm
Room: LEC, Room 303
Abstract: Mass spectrometry proteomics, characterized by spiky, spatially heterogeneous functional data, can be used to identify potential cancer biomarkers. Existing mass spectrometry analyses utilize mean regression to detect spectral regions that are differentially expressed across groups. However, given the inter-patient heterogeneity that is a key hallmark of cancer, many biomarkers are only present at aberrant levels for a subset of, not all, cancer samples. Differences in these biomarkers can easily be missed by mean regression, but might be more easily detected by quantile-based approaches. Thus, we propose a unified Bayesian framework to perform quantile regression on functional responses. Our approach utilizes an asymmetric Laplace working likelihood, represents the functional coefficients with basis representations which enable borrowing of strength from nearby locations, and places a global-local shrinkage prior on the basis coefficients to achieve adaptive regularization. Different types of basis transform and continuous shrinkage priors can be used in our framework. An efficient Gibbs sampler is developed to generate posterior samples that can be used to perform Bayesian estimation and inference while accounting for multiple testing. Our framework performs quantile regression and coefficient regularization in a unified manner, allowing them to inform each other and leading to improvement in performance over competing methods as demonstrated by simulation studies. We apply this model to identify proteomic biomarkers of pancreatic cancer missed by previous mean-regression based approaches. Supplementary materials for this article are available online. 

Speaker: Zach Belou
Affiliation: DXC Technology
Title: Enterprise Applications of Machine Learning in Fault Detection & Autonomous Vehicles
Date and Time: September 23, 3:00 pm
Room: LEC, Room 303
Abstract: As businesses collect more data than ever, sophisticated data pipelines and algorithms provide key advantages for innovative companies. From telecommunications providers that handle the world’s internet traffic to automotive manufacturers trying to perfect self-driving technology, every business is looking for the advantages machine learning and AI promise. In this talk, we will dive into two data science use cases that break down the challenges many enterprises face in the world of Big Data/machine learning, and how data scientists and engineers solve them. We will breakdown how businesses manage the data lifecycle, organize non-structured data, and develop machine learning algorithms in practice. We will also take a comprehensive look at Phased LSTM recurrent networks, Graph database processing, and image classification and their implementation in commercial environments.

Speaker: Zhuolin Qu, PhD
Affiliation: Mathematics Department, Tulane University, New Orleans – LA
Title: Network Modeling the Impact of Community-based Male-screening on the Chlamydia Trachomatis Prevalence in Women
Date and Time: October 7th, 2019, 3:00 pm
Room: LEC, Room 303
Abstract: We create and analyze a stochastic network-based model to understand the control of Chlamydia trachomatis (Ct) among young African American (AA) in New Orleans. Ct is the most commonly reported bacterial sexually transmitted infection in the United States and is a major cause of infertility, pelvic inflammatory disease, and ectopic pregnancy among women. Despite decades of screening women for Ct, the rates continue to increase among young AA compared to other groups. The community-based program “Check It” proposes that men are an important reservoir of infection for women and screening AA men could make an impact on the rates among women. To quantify the effectiveness of the male-screening strategy, we propose an agent-based network model to simulate a realistic sexual contact network for assortative mixing among the targeted population. We model both the existing intervention for women through the annual exam and the “Check It” male-screening based intervention through venue-based enrollment. The model accounts for various intervention strategies implemented in the program, including the expedited index treatment, expedited partner treatment, social network peer referral, and rescreening. We use sensitivity analysis to quantify the significance of each intervention component onto the prevalence in women.

 Speaker: Gustavo Didier
Affiliation: Mathematics Department, Tulane University, New Orleans – LA
Title: On scaling in high dimensions
Date and Time: October 14, 2019, 3:00 pm
Room: LEC, Room 303
Abstract: Scaling relationships have been found in a wide range of phenomena that includes coastal landscapes, hydrodynamic turbulence, the metabolic rates of animals, Internet traffic and finance. For scale invariant systems, also called fractals, a continuum of time scales contributes to the observed dynamics, and the analyst’s focus is on identifying mechanisms that relate the scales, often in the form of exponents. In this talk, we will look into the little explored topic of scale invariance in high dimensions, which is especially important in the modern era of “Big Data”. We will discuss the role played by wavelets in the analysis of self-similar stochastic processes and visit recent contributions to the wavelet modeling of high- and multidimensional scaling systems. This is joint work with P. Abry (CNRS and ENS-Lyon), B.C. Boniece (Washington University in St. Louis) and H. Wendt (CNRS and Université de Toulouse).

Speaker: Xueyan Sherry Liu, PhD
Affiliation: Mathematics Department, University of New Orleans, New Orleans – LA
Title: A novel statistical tool for analyzing colcoalization in super-resolution images.
Date and Time: October 21st, 2019, 3:00 pm
Room: LEC, Room 303
Abstract: Spatial data from high-resolution images abound in many scientific disciplines. For example, single-molecule localization microscopy, such as stochastic optical reconstruction microscopy (STORM), provides super-resolution images to help scientists investigate co-localization of proteins and hence their interactions inside cells, which are key events in living cells. However, there are few accurate methods for analyzing co-localization in super-resolution images. The current methods and software are prone to produce false positive errors, and are restricted to only 2-dimensional images. In this study we develop a novel statistical method, along with an R package and on-line apps, to effectively address the problems of unbiased and robust quantification and comparison of protein co-localization for multiple 2- and 3-dimensional image datasets. This method significantly improves the analysis of protein co-localization using super-resolution image data, as shown by its excellent performance in simulation studies and an analysis of LC3-LAMP1 protein co-localization in cell autophagy.

Speaker: Lin Zhu
Affiliation: Biostatistics Program – LSUHSC School of Public Health, New Orleans, LA
Title: Bayesian Adaptive Designs for Phase III Clinical Trials
Date and Time: October 23rd, 2019, 3:00 pm
Room: Medical Education Building (MEB) Seminar 5 (3217, formerly S14), LSUHSC downtown campus.
Abstract: There is an increasing interest in Bayesian group sequential designs and adaptive randomization designs in clinical trials because of its potential to improve efficiency, to shorten drug development time, and to enhance statistical inference precision without undermining the clinical trial’s integrity and validity. We propose three Bayesian adaptive designs, a Bayesian sequential design with adaptive randomization for continuous outcomes (BSDAR), a Bayesian sequential design for time-to-event outcomes (BSD4TEO), and a Bayesian sequential design for outcomes with a distribution that belongs to the exponential family (BSDEF). For all proposed designs, alpha spending functions are used to control the overall study-wide type I error rate. In the BSDAR, the randomization rate is adaptively changed to attribute newly recruited patients to different treatment arms more efficiently. For the BSD4TEO, Bayes factor is adapted for decision-making at interim analyses for survival outcomes. For the BSDEF, the previously proposed Bayesian sequential design is generalized to outcomes with a distribution belonging to the exponential family by setting priors based on its canonical form.

Algorithms are presented to calculate the optimal randomization rates, to make decision rules and to calculate power of the proposed tests for all three methods. Sensitivity analyses are implemented to evaluate the impact of different choices of prior parameters on choosing critical values. The power of tests, the expected sample size/event size of the proposed designs, and the quality of estimators are studied through simulations, and compared with the frequentist group sequential design or conventional Bayesian design. Simulations show that, when the total sample size is fixed, the proposed designs can obtain greater power and/or cost smaller expected sample size/event size than the traditional designs. The feasibility of each proposed design is further illustrated on real data sets.

Speaker: Karl Mahlburg
Affiliation: Mathematics Department, LSU, Baton Rouge – LA
Title: The extreme value variance estimator
Date and Time: October 28th, 2019, 3:00 pm
Room: LEC, Room 303

Abstract: If a random variable is sampled repeatedly some number of times, then its total range (i.e., the difference between its largest and smallest values) gives another random variable. The distribution of such a random variable is known as an “extreme-value distribution”, which are commonly used in applications such as financial modeling. I will discuss a simple application developed by Parkinson (1980) in which the extreme-value is used to estimate variances five times more efficiently than the commonly used unbiased sample variance. I will also discuss the role of extreme-value distributions in some interesting problems from enumerative combinatorics. 

Speaker: Christie Watters
Affiliation: Biostatistics Program, LSUHSC, New Orleans, LA
Title: Extension, investigation, and comparison of methods of inverse prediction to general models for heteroscedastic multivariate responses.
Date and Time: November 1st, 2019, 10:00 am
Room: LEC 306

Abstract: The object of inverse prediction is to infer the value of a condition t that caused an observed response y, based on a model relating responses to conditions fit to training data. Existing methods of inverse prediction were developed for univariate responses (r = 1), straight-line relations (SL) µ(t) = β0 + β1t to t, and constant variance. While other approaches have been described, three methods are most commonly used in practice; inverse regression (IR), inverse prediction (IP), and reverse regression (RR). The main objective of this research was to compare the performance characteristics of methods of inverse prediction, using both simulation experiments and re-sampling from data collections. In that process, I discovered or proved several new theoretical results and relations, as well as extended these methods beyond their original setting. The methods now have extensions into multivariate responses, linear splines models, and in the case of IP, a mixed model can be used to model the variance. Rejection rates, empirical representations of power are used as a measurement for comparison of these models in different settings.

Speaker: Luis Escobar
Affiliation: Experimental Statistics Department, LSU, Baton Rouge – LA
Title: Prediction of Failure Times and the Number of Field Failures in Reliability Studies
Date and Time: November 18th, 2019, 3:00 pm
Room: LEC, Room 303

Abstract: In this talk, we describe methods to construct two-sided prediction intervals or one-sided prediction bounds for future random quantities. Both new-sample prediction (using data from a previous sample to make predictions on a future unit or sample of units) and within-sample prediction problems (predicting future events in a failure process based on early data from the process) are considered. To illustrate new-sample prediction, we show how to construct a prediction interval for a single future observation from a previously sampled population/process (motivated by a customer’s request for an interval to contain the life of a purchased product).  To illustrate within-sample prediction, we show how to compute a prediction interval for the number of future failures in a specified period beyond the observation period (motivated by a warranty prediction problem).  A third example requires more general methods to deal with complicated censoring arising because units enter service at different points in time (staggered entry). This work is based on the forthcoming Second Edition of the book Statistical Methods for Reliability Data by W. Q. Meeker, L. A. Escobar, and F. G. Pascual.


Organizer: A. Chapple, PhD