Unifying probe effect and array effect to detect transcribed fragments in tiling arrays

In this paper, we present a model to elaborate the sources of randomness in multiple tiling array data. We also propose a new probe score system which integrate the intensity information of a probe and its neighbors. This new score system is obtained by using slide window and median polish strategies. These strategies firstly unify probe affect across probes and the array effect across arrays based on the region-specific fixed values interpreting the transcription levels of regions in probes, then unite these unified information into the region-specific fixed values. And this united information become a new standard to evaluate that probes are or aren't in transcriptional regions. Similar to median method, our method is a way that integrate the information from multiple arrays in the analysis stage rather than the results stage. A classical hidden Markov model (HMM) is used to model the distribution of tilling array probe scores in transcribed and non-transcribed regions and then to predict the transcribed fragments. The priority of the proposed score system is illustrated based on Affymetrix's RNA tiling array data.