Breakpoint and Outliers Detection (Fig 2)

A breakpoint is defined when two consecutive $\hat{\theta}_i$ are different. The AWS procedure may in some cases identify breakpoints which correspond to small shifts and not to a real genomic alteration. This is probably due to specific local effects on the chromosome, unrelated to the biological variation we want to investigate. Thus, a procedure which is very similar to the JOIN step of the GLSo algorithm proposed by Jong et al. (2003) based on penalysed maximum likelihood, allows the undesirable breakpoints to be removed.

The AWS procedure may fail to detect very fine structures such as a clone located in a region for which the signal $Y_i$ differs significantly from the expected values of this region. To overcome this limitation in the detection of outliers, we have designed a special procedure based on median-absolute-deviation for detecting the remaining outliers. It should be noted that when an outlier presents a large deviation, it is already detected at the breakpoint detection step.

Figure 2: Breakpoints and Outliers