Experimental designs for optimisation of the image analysis process for cDNA microarrays

Image analysis is an important part of the process of generating good quality microarray data, prior to data mining. Most image analysis software packages have a large number of parameters, each with a range of settings, resulting in a plethora of potential permutations. In order to standardise the process, we have used Plackett–Burman and 26−1 fractional designs to select the optimal settings. Nineteen parameters were investigated, and optimal settings were successfully found by running only 42 experiments. The optimal parameter settings for our software, RoBioVision, were compared with expert-derived settings. The comparison results show that the optimal settings reduced within-slide variation. The results also indicate that the median is a good spot signal measurement compared to other studied alternative measurements since it has highest correlation coefficient between the left and right side spots.