Subnetwork State Functions Define Dysregulated Subnetworks in Cancer

Subnetwork State Functions Define Dysregulated Subnetworks in Cancer Salim A. Chowdhury, Rod K. Nibbe, Mark R. Chance, Mehmet Koyuturk JCB 2011

Previous Work

Mutual Information • Entropy H(X) high entropy low entropy • Mutual Information I(Y; X) = H(X)-H(X|Y) low mutual information high mutual information

Mutual Information – Dysregulation Measure C 0 1 1 0 1 0 0 1 1

Combinatorial Coordinate Dysregulation State function is considered informative if: • • There is no redundancy in the state function for every

Pruning the search space Where • Notice that Jbound() is not J() • This result will help us decide if we would like to extend subnetwork S. If we will not extend S.

KRANE Algorithm 0 0 1 1 01 • S is extensible if • From the possible extension we choose to further check only b extensions with the top J() value. • Stop extending S if |S|>d.

Complexity • Complexity is exponential in d. • To make sure we don’t miss subnetworks we should use • Using Jbound() we could prune the search space thus reducing running time without loosing results.

CRANE Runtime

Neural Network Classification • Rank the subnetwork according to I(FS; C) and take the top K ranked subnetworks that are not overlapping. • Use these Network as input for NN that predicts metastasis.

Effect of parameters

Conclusion Advantages: • Combinatorial dysregulation. • More sophisticated heuristics base on theoretical bound (“almost” exhaustive search). Shortcomings: • Runtime is exponential in d so we could check only relatively small networks. • Even for small values of d we have dimensionality problems. • No post-processing that tries to merge subnetworks. • Dismissing of overlapping subnetworks.