A Confidence Interval Approach to Gene Chip Analysis

Recently, gene chip technology and data generated from this technology have given statisticians a new realm in which to develop statistical methodology. In gene chip analysis, mRNA from a tissue sample for a disease of interest is isolated and placed on the chip. From expression data generated from the chip, we then determine what genes are highly expressed for the particular disease of interest. The challenge is to find approximately 100 genes from 12,000 that warrant further investigation. To further complicate matters, gene chips are very expensive (~$1000 each) and sample sizes tend to be very small, with n<5 chips typically used. Using SAS 8.0 under Windows 2000 and programmed mostly within a DATA step, a confidence interval approach was developed that examines various thresholds for gene expression. mRNA from 8 normal and 8 apolipoprotine AI (apo AI, a gene known to play a pivotal role in HDL metabolism) knocked out mice were isolated and 16 gene chips were analyzed. Expression ratios (red/green) for each gene were determined within a gene chip (mouse) and 99% confidence intervals calculated for the mean expression ratio across the 8 mice within a treatment group (normal and apo AI). The lower limit of each confidence interval is compared to a series of expression ratio thresholds and those genes whose lower limit is greater than the threshold are flagged in both groups. The total number of genes flagged for each threshold is calculated for each group and plotted using SAS/GRAPH . Additionally, the genes flagged are compared between groups to determine how the genes are differentially expressed.