Comparison of sequence-dependent tiling array normalization approaches

The detection of enriched DNA or RNA fragments by tiling microarrays has become more and more popular. These microarrays contain a high number of small probes covering genomic loci. However, to achieve high coverage the probe sequences cannot be selected for their hybridization properties. The affinity of the probes towards their targets varies in a sequence-dependent manner. In order to remove this bias a number of approaches have been developed and shown to increase the detection of enriched DNA or RNA fragments. However, these approaches also employ a peak detection algorithm that is different from the one used previously. Thus, it seems possible that the enhancement of detection is due to the peak detection algorithm rather than the sequence-dependent normalization.

We compared three different sequence-dependent probe level normalization procedures to a naïve sequence-independent normalization technique. In order to achieve maximal comparability, we used the normalized intensity values as input to a single peak detection algorithm. A so-called "spike-in" data set served as benchmark for the performance. We will show that the sequence-dependent normalization procedures do not perform better than the naïve approach, suggesting that the benefit of using these normalization approaches is limited. Furthermore, we will show that the naïve approach does well, because it effectively removes the sequence-dependent component of the measured intensities with the help of the control hybridization experiment.

Sequence-dependent normalization of microarray data hardly improves the detection of enriched DNA or RNA fragments. The "success" of the sequence-independent naïve approach is only possible due to the control experiment and requires proper scaling of the measured intensities.