Screenshots of Tilescope. (a) The applet of Tilescope, the graphic user interface of the pipeline. (b) An example of the data analysis result web page.

The ROC curves of the three feature identification methods implemented in Tilescope. The comparison of the performance of these methods was based on a well-studied STAT1 ChIP-chip data set and a list of experimentally tested STAT1 binding sites.

Tiling array data processing by Tilescope. (a) Flow chart of major data processing steps. Yellow icons represent data in user-accessible files, and blue ones data in the pipeline program memory. See main text for details. (b) Log-intensity scatter plots of a tiling array from the STAT1 experiment set before and after normalization by four different methods. The first panel is the log₂T verses log₂R plot before normalization, where T and R are test intensity and reference intensity, respectively. The gray line represents where these two log-intensities are equal. The second panel is log₂(T/R) verses log₂(T×R) plot (the MA plot) before normalization. The dependency of the log-ratio on the intensity, evinced by a fitted loess curve, is prominent in the data. The other panels are the MA plots of array data after mean, median, loess, or quantile normalization. They clearly show that the distribution of log-ratios is centered at zero by all normalization methods, but the intensity-specific artifacts in the log-ratio measurements are removed by only loess or quantile normalization and not by the mean- or median-based method. (c) Signal and P value maps of all tiles in the ENCODE ENm002 region. In this region, the tiles near the transcription start site of IRF1, a transcription factor known to be regulated by STAT1, give the strongest signals. (d) Tilescope-identified STAT1 binding sites at the 5'-end of IRF1 are shown on the custom track in the UCSC genome browser.