Multiple Imputation for Microarray Missing Data with Slice Sampler

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Hui Xie*+ and Leann Myers and Steven Smith and Janet Rice and Michelle Lacey
• Companies: Pennington Biomedical Research Center and Tulane University and Pennington Biomedical Research Center and Tulane University and Tulane University
• Keywords: microarray ; missing data ; multiple imputation ; slice sampler ; truncated ; hyperretangle slices

Abstract

Missing data complicates microarray statistical analysis because many algorithms require complete gene data sets. One effective way to solve this problem is through imputation. Both single and multiple imputation methods can be applied to estimate missing points, but the latter provides superior performance because it estimates uncertainty of missing data. This project presents multiple imputation by using Slice Sampler which draws random samples from the joint uniform distribution introducing latent variables. By exploiting slice sampler, microarray missing data can be M (>1) times simulated within "hyperretangle" slices via Markov chains. Microarray data follows truncated and mixture multivariate normal distribution, respectively.Imputation accuracy is evaluated by RMSE. Across missing rates of 5-30% slice sampler reduces RMSE up to 20% as compared to KNN and 32% to row mean methods.