2015 Informs Annual Meeting

MC33

INFORMS Philadelphia – 2015

MC32 32-Room 409, Marriott Statistical Innovations in Computational Biology and Genomics Cluster: Big Data Analytics in Computational Biology/Medicine Invited Session Chair: Mingyao Li, Statistical Innovations in Computational Biology and Genomics, University of Pennsylvania, 213 Blockley Hall, 423 Guardian Drive, Philadelphia, PA, 19104, United States of America, mingyao@mail.med.upenn.edu 1 - Prediction using Multinomial Inverse Regression in Microbiome Studies Hongzhe Li, Professor of Biostatistics, University of Pennsylvania, 215 Blockley Hall, Philadelphia, PA, United States of America, hongzhe@mail.med.upenn.edu Next-generation sequencing technologies allow 16S rRNA gene surveys or metagenome shotgun sequencing in order to characterize taxonomic composition. The data can be summarized as k-mer counts. We consider the regression problem for for such high dimensional in order to build a model for predicting the clinical outcomes based on microbiome data and demonstrate its applications. 2 - Computational Validation of NGS Variant Calls using Genotype Data Margaret Taub, Assistant Scientist, Johns Hopkins Department of Biostatistics, 615 N Wolfe St, E3527, Baltimore, MD, United States of America, mtaub@jhsph.edu, Suyash Shringarpure, Ingo Ruczinski, Rasika Mathias, Kathleen Barnes We performed a comparison of different variant calling algorithms on 642 samples whole-genome sequenced to an average depth of 30x, focusing on characteristics of variants called by different subsets of callers. We developed a classifier which uses genotyping array data, often collected for all sequenced individuals, as a gold standard to improve calibration of variant calls. We found little difference in quality between single- and multi-sample calling methods at 30x coverage. Chair: Douglas King, University of Illinois at Urbana-Champaign, 117 Transportation Bldg., 104 S. Mathews Ave., MC-238, Urbana, IL, 61801, United States of America, dmking@illinois.edu 1 - Methods in Treatment Planning with Continuous Dose Delivery Kimia Ghobadi, Massachusetts Institute of Technology, 77 Massachusetts Ave, Cambridge, MA, United States of America, kimiag@mit.edu, David Jaffray, Dionne Aleman, Caroline Chung In this work we investigate continuous dose delivery models and algorithms for head-and-neck patients. We discuss the necessary changes and considerations in the optimization models and algorithms for different tumour sites. We will also present the clinical realization of the plans and compare the obtained clinical results with the simulated treatment plans. 2 - Optimal Timing of Living-donor Liver Transplantation under Risk-aversion Ozlem Cavus, Bilkent University, Department of Industrial Engineering, Ankara, 06800, Turkey, ozlem.cavus@bilkent.edu.tr, Umit Emre Kose, Oguzhan Alagoz, Andrew J. Schaefer The timing of liver transplantation is crucial as it affects the quality and the length of the patients’ life. Previous studies used risk-neutral Markov Decision Processes to optimize the timing of the transplantation. In this study, we model the risk- averse behavior of the patients using coherent dynamic measures of risk. We obtain optimal policies for different patients and donated organs. We also derive the structural properties of the optimal policy. Supported by TUBITAK [Grant 213M442]. MC33 33-Room 410, Marriott Topics in Health Systems Sponsor: Health Applications Sponsored Session

3 - Engineering Technician Location Analytics Dwayne Otis, Air Products and Chemicals, 7201 Hamilton Boulevard, Allentown, PA, 18195, United States of America, otisdk@airproducts.com In this talk we present a model whose objective is to determine, within a given physical region, the near-optimal required number of technicians and their assigned home base locations in order to satisfy defined customer service levels. Ideally, the workload for each technician would be balanced and the total cost to meet the desired service level is minimized. Furthermore, we examine additional areas where this general technique could be applied.

MC31 31-Room 408, Marriott Data Mining in Healthcare

Sponsor: Data Mining Sponsored Session

Chair: Ramin Moghaddass, Assistant Professor, University of Miami, 1251 Memorial Dr, Coral Gables, FL, 33146, United States of America, ramin@miami.edu 1 - Influential Neighbor Analysis with a Hierarchical Bayesian Model Ramin Moghaddass, Assistant Professor, University of Miami, 1251 Memorial Dr, Coral Gables, FL, 33146, United States of America, ramin@miami.edu, Cynthia Rudin For a doctor with enough experience, almost every patient would have similarities to key cases seen in the past and each new patient could be viewed as a mixture of important pieces of these key past cases. Because doctors often tend to reason this way, an efficient computationally aided diagnostic tool might be helpful in locating key past cases of interest that could assist with diagnosis. We develop a model to mimic the type of logical thinking that physicians use when Fulton Wang, MIT, 5 Cambridge Center #792, Cambridge, MA, 02142, United States of America, fultonwang@gmail.com, Cynthia Rudin Falling rule lists are classification models consisting of an ordered list of if-then rules, where (i) the order of rules determines which example should be classified by each rule, and (ii) the estimated probability of success decreases monotonically down the list. These kinds of rule lists are inspired by healthcare applications where patients would be stratified into risk sets and the highest at-risk patients should be considered first. 3 - Machine Learning for Clinical Decision Support for Heart Failure(HF) Readmission Wei Jiang, PhD Student, Johns Hopkins University, 6606 Copper Ridge Drive Apt. 201, Baltimore, MD, 21209, United States of America, wjiang1990@gmail.com, Sean Barnes, Matthew Toerper, Scott Levin, Eric Hamrock, Sauleh Siddiqui, Stephanie Cabral Predicting risk of HF readmission have gained increasing attention.Previous studies mainly used administrative data. We will focus on using clinical data from EMR for predicting HF readmission by creating structured data from unstructured clinical data and combining it with administrative data. Then we use classification models such as random forest and support vector machine for predicting purpose. In the end, we will demonstrate the value of clinical data in predicting HF readmission. 4 - Or’s of and’s for Interpretable Machine Learning: Prediction and Explanation Tong Wang, Graduate Student, MIT, 70 Pacific Street, apt 242A, Cambridge, MA, 02139, United States of America, tongwang@mit.edu, Cynthia Rudin We present a form of interpretable models for data prediction and explanation. The model is comprised of a small number of disjunctions of conjunctions (or’s of and’s). We apply OA models to solve two machine learning tasks, classification and causal effect analysis. We formed two integer linear programs to construct the optimal OA models that can achieve good performance and interpretability, by incorporating regularizations on the sparseness. We show that regularizations reduce computation. considering past cases. 2 - Falling Rule Lists

215