Title:Extracting more signal at high intensities in oligonucleotide arrays

Abstract: The majority of work in the analysis of microarray data has focused on improving the signal to noise ratio in the low concentration region. However, improvements are also possible at the opposite extreme: at high concentrations. This problem appears to have been neglected for two reasons: first, it is counter-intuitive to expect problems in what appears to be the cleanest region and second, the data necessary to identify and develop improvements has not been widely available. Fortunately, the situation has changed and precise calibration data produced by Affymetrix which enables the analysis of signals over a broad range of concentrations. Using this data, we address the question: ``Where on the concentration/intensity scale are we able (or unable) to detect fold changes of 2 and greater?'' First, we empirically demonstrate that while current techniques are capable of detecting changes in the low to middle concentration range, the situation is quite different for high concentrations. Specifically, in this regime, expression changes as large as 4 fold are severely compressed, and changes of 2 are often undetectable. However, a feature of {\sf GeneChips} is that transcripts are probed by a set of sequence fragments having a wide range of affinities. The key to improving the accuracy in a given intensity range, is choosing those probes that are most responsive in that range. Utilizing this property and borrowing ideas from the theory of local linear embeddings (LLE) we present a new approach capable of reducing the compression at high intensity.