A Statistical Model-Based Quantile Normalization Method for Hit Detection in High-Throughput Screening Experiment (HTS)

• Activity Number: 30
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Quality and Productivity
• Author(s): Xin Wei*+
• Companies: Roche Pharmaceutical
• Keywords: random effects ; quantile normalization ; high throughput screening ; quanlity control ; variance component

Abstract

The objective of a high throughput screening (HTS) experiment in pharmaceutical industry is to identify active compounds (hit) against specific drug targets/biological entities from an enterprise-level compound inventory that typically contains millions of compounds. A HTS experiment usually contains multiple positive/negative controls for each test plate and all the compound of interest being tested only once. Despite the great technological advancements in HTS field, major obstacles for HTS experiment quality control and hit detection still remain such as the systemic bias from unknown sources and the lack of technical replications for test compounds. In this paper, we propose to use random effects model to quantitatively determine the sources of systemic variation in HTS experiment based on control data alone. Furthermore, we design a modified Quantile-Quantile normalization across all the levels for the major variation source to minimize the systemic bias identified by the random model. Both in silico prediction and wet lab confirmation reveal that this random model based QQ normalization out-perform the popular normalization method such as z-score by recovering more hits.