Semiparametric Theory for Missing Data Applied to Pooling Experiments for Chemical Compounds

• Activity Number: 44
• Type: Topic Contributed
• Date/Time: Sunday, August 3, 2003 : 4:00 PM to 5:50 PM
• Sponsor: WNAR
• Author(s): Bingming Yi*+ and Jacqueline M. Hughes-Oliver and S. Stanley Young
• Companies: Merck & Co., Inc. and North Carolina State University and Glaxo Wellcome Inc.
• Keywords: blocking ; group testing ; high throughput screening ; missing at random ; pooling experiment ; synergism

Abstract

In many pharmaceutical companies, pooling experiments have emerged as an attractive way to improve the screening efficiency of chemical compounds. To fully realize the potential benefits of pooling, it is important that inferences from pooling experiments are appropriately obtained. This work proposes a semiparametric model by viewing pooling experiments as a special missing data problem. Individual responses within a pool are missing when only the pooled response is obtained and no retesting is done on individuals within this pool. For each pool, pooled responses and all structural information for all compounds within the pool are always available. The strategy for determining which pools will be retested is based on observed data only, thus the pooling data can be regarded as missing at random. Application of the proposed semiparametric model is made to real data originating from existing compound libraries. We demonstrate increased efficiency of semiparametric estimators over estimators obtained from (incomplete) likelihood models.