Comparison of Two Ways of Handling Replicates in the Context of Distance Measure of Time Course Microarray Gene Expression Data

• Activity Number: 359
• Type: Contributed
• Date/Time: Wednesday, August 10, 2005 : 8:30 AM to 10:20 AM
• Sponsor: General Methodology
• Author(s): Yu Guo*+ and Cheng Li
• Companies: Harvard School of Public Health and Harvard School of Public Health
• Keywords: microarray data ; time course data ; distance measure ; correlation coefficient ; replicate

Abstract

In a time course gene expression microarray experiment, a commonly used distance measure for clustering analysis is correlation-based distance. But with replicated data, treating replicates and time points as independent in calculating correlation may not be appropriate. We propose that taking the average of the replicates before calculating correlation may give a more accurate measure of gene-relatedness. We used the inner ear cells of a mouse dataset to answer two questions: How do we treat replicates in time course microarray data? How many replicates are economically feasible and powerful enough for a reliable conclusion? We simulated experiments based on the assumption that replicates of each measurement follows a normal distribution and used different ways to estimate the distributions of correlation coefficient estimates. Results show that taking average over replicates gives less biased estimates of correlation by average improvement of 15%. Going from one to two replicates gives 41% less variance in correlation estimates, while going from two to three gives only 18%. So, using two replicates in our setting would be enough to correct bias and remain economical.