Detecting Peaks in a High-resolution Map of Active Promoters in the Human Genome Using Double-Regression Model

• Activity Number: 62
• Type: Contributed
• Date/Time: Sunday, August 7, 2005 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Ming Zheng*+ and Yingnian Wu and Tae Hoon Kim and Leah Barrera and Chunxu Qu and Michael Singer and Todd Richmand and Roland Green and Bing Ren
• Companies: University of California, Los Angeles and University of California, Los Angeles and Ludwig Institute for Cancer Research, UCSD and University of California, San Diego and Ludwig Institute for Cancer Research, UCSD and Nimblegen Systems, Inc. and Nimblegen Systems, Inc. and Nimblegen Systems, Inc. and University of California, San Diego
• Keywords: human genome ; promoter ; peak finder ; model-based

Abstract

We develop a model-based detection method for detecting the peaks in biological data. The detection of peaks is important in a biological sense as peaks correspond to the beginning of genes and such detection can be used to verify existing genes and identify new genes. From the observation of the raw data, local fitting of a double-regression model, which is to fit two regression models simultaneously under some consistence restriction, is proposed for detection. A detailed scientific justification of the double-regression model from the point of view of how the data are generated and approximated by Poisson process are provided, and the model is proved to be equivalent to a single regression model with multiple predictors well-designed for detecting peaks. Statistical inference (e.g., confidence intervals of the positions of peaks, the significance level of each peak) are calculated for the user's reference. Multiple testing methods, such as FDR-controlling method by Benjamini and Hochberg (1995) and Benjamini and Yekutieli (2001) are proposed for the problem of testing the identified peaks simultaneously.