Correlation Cluster Filter
The correlation cluster filter will find sets of probes which have a similar
pattern of quantitation across a set of DataStores. This filter lets you
find connected sets of probes, and will discard probes which do not form a
cluster.
The actual clustering is fairly simplistic, and is optimised for speed
rather than completeness. A single pass is made through the set of probes.
Each probe is compared to the existing set of clusters. If the correlation
between this probe and all of the probes in an existing cluster is above the
threshold set then the probe is added to that cluster. If none of the clusters
want to claim a probe then it starts a new cluster. If two or more clusters
want to claim a probe then it is added to the one with the highest overall
correlation.
Options
- You can choose which data stores you want to use for your correlation. You
must choose at least 3 stores and there is no upper limit to the number you choose.
- You can set a minimum size of cluster you are interested in. You will find that
you will normally generate a small number of clusters containing a large number of
probes, but that you will also have a large number of clusters containing only a few
probes. Setting this value to a sensible number will stop you from being swamped
with large numbers of clusters.
- You can choose how correlated a set of probes need to be to form a cluster. This
value must be between 0 and 1. Setting it to a low value will form very loose and
noisy clusters with larger numbers of probes in them. Setting it higher will form
tighter clusters which contain fewer probes.