The intensity difference filter is a simple filter which can be used to compare two or more data stores. It can assign a p-value to each individual probe without the need for replicates. It is designed to work with datasets which are largely similar, but which show an intensity dependent effect on the variability within the data, normally such that higher values show lower variability and vice versa.
The filter works by constructing a local distribution of differences from probes with similar average intensity to the probe being examined. The set of differences is then modelled with a normal distribution and this model is used to assign a probability of the probe being examined falling within that distribution. At low intensities the differences distribution will tend to show high variability, and at high intensities the variability will be lower, so the p-values calculated take this change in variability into account.
The number of probes in the local distribution is either 1% of the total probes examined, or 100 probes, whichever is the larger. Where multiple data stores are selected then all possible pairwise combinations of stores are examined, and the lowest p-value from that set is assigned to the probe.