ANOVA-Based Clustering to Improve Potency Estimation in Quantitative High-Throughput Screening Data

• Activity Number: 451 - Current Trends in Statistical Genomics: Finding Needle in a Haystack?
• Type: Topic Contributed
• Date/Time: Wednesday, August 2, 2017 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Shyamal Peddada and Keith Shockley* and Abhishek Kaul and Shuva Gupta and Shawn Harris and Soumendra N Lahiri
• Companies: NIH/NIEHS and NIEHS and NIEHS and UC Davis and SSS and North Carolina State University
• Keywords: Estimation of AC50 ; High Throughput Screening ; Toxicology

Abstract

Quantitative high throughput screening (qHTS) experiments can generate thousands of concentration-response profiles. Therefore, qHTS technology provides an opportunity to test the large inventory of chemicals with unknown activity for potentially adverse effects. However, due to various factors such as experimental run, chemical supplier, impurities, potential signal flare due to chemicals in the neighboring wells, etc., the dose response patterns of different "replicates", within a given chemical, are sometimes non-homogeneous. It is practically impossible to evaluate several thousands of dose-response patterns one by one to identify chemicals whose replicates display non-homogeneous clusters. Since the potency of a chemical is estimated using dose-response curves, it is of considerable importance to develop an automated quality control process to flag and separate out anomalous chemicals where the replicates have distinct dose-response patterns. In this talk we introduce a novel ANOVA based methodology that identify such anomalous data. The proposed methodology is illustrated using a Tox21 Phase II estrogen receptor agonist assay data consisting of 30,000 concentration curves.