A Strategy to Reduce Batch Effect in High-Throughput Experiments with Optimized Experimental Design

• Activity Number: 241
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Li Yan*+ and Song Liu and Changxing Ma
• Companies: Roswell Park Cancer Institute and Roswell Park Cancer Institute and SUNY at Buffalo
• Keywords: randomization ; design ; GWAS ; epigenetic ; mythylation ; batch effect

Abstract

In the past few years, drastic cost reduction and wide-spread availability makes possible for large scale genomic and epigenetic experiment, such as GWAS and mythylation studies. It had been observed that batch effect, due to non-biological experimental variation, can introduce profound heterogeneity that reduced the effectiveness of the experiment. Rescue efforts trying to adjust or filter the batch effect at the analysis stage had been proven difficult. It is highly desired to employ proper experimental design and optimization techniques to minimize the batch effect at the beginning of high-throughput experiments. We provide a general framework for this important practice. Our method employs block design concept followed by optimization which can handle both balanced and unbalanced subgroups. We demonstrated the proposed method is statistically more powerful than simple randomization through simulation studies. We had applied the principles and techniques to two Illumina methylation 450K experiments for breast cancer and bladder cancer, respectively, as well as a GWAS for Bone Marrow transplants, each involving hundreds of samples. The relevant software is freely available.