Microarray Normalisation and CGH Profiling Software

BACKGROUND: Ideally, a normal vs. normal microarray CGH experiment should allow one to define the experimental baseline noise level to use as thresholds for defining copy number change. However, intraexperimental variation (introduced by differences in hybridization, labeling, and scanning) precludes one from using baseline noise levels determined from one experiment, and applying to another experiment. To circumvent this, a sliding window algorithm was developed.

HYPOTHESIS: When performing microarray CGH on a sample with no copy number changes, the baseline noise level would simply be calculated as the standard deviation (i.e. variability) of all the spots across the genome. However, in samples with copy number alterations (eg. tumour samples), determination of baseline noise by this method would be biased, as the calculated standard deviation would be larger than expected (due to contribution of spots in regions of copy number alteration). Instead, by using a method of sampling many areas of the genome, determining the variability in those samplings, and identifying the minimum calculated variability of all those samplings, one may obtain a more representative value of the baseline noise. Finally, rather than randomly sampling across the genome, a sliding window method (of user-defined window size) may be used to systematically calculate the variability across all possible samplings in the genome. This derived minimum variability may thus be used as the noise threshold for determining copy number changes.

ALGORITHM: The window is defined as a number of contiguous linear array spots along each chromosome (default window size = 80). Starting at the pter of a chromosome (usually, chromosome 1), the window is populated with normalized intensity ratios. Subsequently, the standard deviation (i.e. variability) of the window is determined. The window subsequently slides a single spot position toward the chromosome qter, and repeats the variability determination for the new window. This process is reiterated until the end of the chromosome is reached. The window then moves to the next chromosome and repeats this process. When the entire genome has been examined, the baseline noise threshold is determined to be the minimum calculated variability sampled from across the entire genome. This threshold is then used in determining which spots have copy number imbalance. Note chromosomes with number of contiguous features less than the window size are currently excluded.

Please note: The default size of the sliding window was determined by comparisons between metaphase CGH and array CGH results performed in this laboratory using NUB-7, prostate, and other cell lines with established genomic imbalance data to derive empirical threshold values for a given array experiment. Users of the software should empirically derive their own optimal window size. In the future, the software will allow for automatic determination of optimal window size based on statistical analysis of the raw data.