BIOS Colloquium – Fall 2020

Fall 2020

Organizer: Huiyi Lin, PhD


Date and Time: Aug. 31 (Mon), 2020; 3-4 pm
Speaker: Erick Nguyen
Affiliation: MS student, Biostatistics Program, School of Public Health, LSUHSC
Title: Internship Application and Experience
Location: Zoom meeting

Abstract: The process for applying for summer internships, preparing for applications and interviews, and then sharing my USAA my summer internship experience will be presented and discussed.


Date and Time: Oct. 5 (Mon), 2020
Speaker: Andrew G. Chapple, Ph.D.
Affiliation: Assistant Professor, Biostatistics Program, School of Public Health, LSUHSC
Title: A novel Bayesian continuous piecewise linear log‐hazard model, with estimation and inference via reversible jump Markov chain Monte Carlo
Location: Zoom meeting

Abstract: We present a reversible jump Bayesian piecewise log‐linear hazard model that extends the Bayesian piecewise exponential hazard to a continuous function of piecewise linear log hazards. A simulation study encompassing several different hazard shapes, accrual rates, censoring proportion, and sample sizes showed that the Bayesian piecewise linear log‐hazard model estimated the true mean survival time and survival distributions better than the piecewsie exponential hazard. Survival data from Wake Forest Baptist Medical Center is analyzed by both methods and the posterior results are compared.


Date and Time: Oct. 19 (Mon), 2020; 3-4 pm
Speaker: Ann Chen, Ph.D.
Affiliation: Associate Member, Dept. of Biostatistics and Bioinformatics, Moffitt Cancer Center & Research Institute
Title: Single cell RNA-seq and metabolomics visual analytics
Location: Zoom meeting

Abstract: Single-cell technologies allow characterization of genomics, transcriptomes, and epigenomes for individual cells under different conditions and provide unprecedented resolution for researchers.  During the first part of the talk, we will first introduce an interactive toolbox SinCHet, which we develop to analyze single cell data for studying heterogeneity using Shannon Profile of at different resolutions.  We devise a D statistic, using area under the Profile of Shannon Differences, to detect heterogeneity differences between conditions.  Recently, we generalize this tool by implementing de-batching, semi-supervised clustering algorithm, and subpopulation-comparison modules in SinCHet-MS for analyzing single cell mass spectrometry (SCMS) metabolomics dat. These suites of tools provide insights into emerging or disappearing subpopulations between conditions, and enable the prioritization of biomarkers for follow-up experiments.  We applied these tools to several single datasets.  We will discuss the results and show how these approaches were used to improve melanoma treatments in animal experiments.   We demonstrate that our approach enables the scientific discovery and generates testable hypotheses that could be validated across different platforms and independent patient cohorts.


Date and Time: Nov. 2 (Mon), 2020; 3-4 pm
Speaker: Heng-Yuan (Henry) Tung
Affiliation: PhD student, Biostatistics Program, School of Public Health, LSUHSC
Title: New Lasso method for SNP-SNP interaction association study
Location: Zoom meeting

Abstract: A polygenic risk score (PRS), which collects various single nucleotide polymorphisms (SNPs), is a useful disease prediction and risk classification tool. Although the conventional PRS, the sum of multiple SNP individual effects, is helpful but not sufficient. Studies have shown that SNP-SNP interactions can improve prediction for certain complex diseases. The Additive-Additive 9 Interaction (AA9int) is a powerful method to measure the SNP-SNP interactions associated with an outcome by assessing the non-hierarchical structure and various directions of the additive SNP inheritance mode. Due to the consideration of all pairwise interactions of SNPs, the number of predictors dramatically increases in the SNP-SNP interaction analysis. In order to construct a multivariable model for PRS, an efficient screening method is therefore essential. For association studies with a small sample size, SNP pairs with a complicated interaction pattern tended to be neglected. For addressing these issues, we proposed AA9Lasso, a new Lasso approach to perform multivariable-based variable selection. This AA9Lasso is a modified version of group lasso with a new weighting in the penalty terms used for AA9int identified SNP pairs. The complicated interaction pattern could benefit from weighting in AA9Lasso during the screening process. Our simulation results showed that AA9Lasso performed better than the conventional univariate selection approach and standard Group Lasso in most of the conditions. Further, we developed the cluster-based two-stage AA9Lasso method to deal with the highly correlated issue of SNP pairs and reduce the computation burden. We then applied the AA9int for evaluating SNP-SNP interactions associated with prostate cancer aggressiveness for 1925 African American men. The AA9Lasso PRS selected distinct SNP pairs compared with the uni-pair p-value approach. Our results demonstrated that the integrated PRS of these two methods had achieved better than each of these two approaches.


Date and Time: Nov. 30 (Mon), 2020; 3-4 pm
Speaker: Cornelius L. Rosenbaum
Affiliation: MS student, Biostatistics Program, School of Public Health, LSUHSC
Title: Proxy variables reduce bias due to unmeasured confounders associated with dichotomous outcomes: A simulation study 
Location: Zoom meeting

Abstract: We present a simulation study and application that shows a proxy variable related to an unmeasured confounder improves the estimate of a related treatment effect in binary logistic regression. The simulation study included 1,000,000 randomly generated parameter scenarios of sample size 10,000. We assessed bias by comparing the probability of finding the expected treatment effect relative to the modeled treatment effect with and without the proxy variable. Inclusion of a proxy variable in the logistic regression model significantly reduced the bias of the treatment or exposure effect when compared to logistic regression without the proxy variable. Including proxy variables in the logistic regression model improves the estimation of the treatment effect at weak, moderate, and strong association with unmeasured confounders.