Hi Sebastian Land, and thank you for replying.
It was not exactly what I wanted to say. The point was to reuse the mathematical aspect of an existing approach (gene analysis) to fulfill data mining tasks, to say,
feature selection. You really can compute a graph between features, using one of these techniques, the choice will be the computational load :
Imagine now that you have an example set with 100 attributes :
row attr1 attr2 ...attr100
1 0.5 3 ....2
2 -0.7 4 ...1
3 0 4 ... 3
....
With such a technique, you should obtain a graph between features, with the following adjacency matrix :
attr1 attr2 attr3 ...
attr1 0 1 0 ...
attr2 1 0 0
attr3 1 1 0 ...
etc...
Thus, manipulating connected components (and others), you could detect which attributes are relevant. There "filter approach" would be powerful, both for data quality/profiling and for preprocessing.
About Cytoscape, actually two plugins would be needed, not exactly the whole software :
- Network Inference :
http://www.baderlab.org/Software/ExpressionCorrelation- Network clusters :
http://baderlab.org/Software/MCODE