In as such will have a sensitivity of 0. Either of these can be substituted with experimental sensitivity values that have the corresponding target combination. In numerous prac tical scenarios, the target combination of no inhibition has sensitivity 0. With the lower and upper bound of the target combi nation sensitivity fixed, we now must perform the infer ence step by predicting, based on the molarity calculator distance between the subset and superset target combinations. We per form this inference based on binarized inhibition, as the inference here is meant to predict the sensitivity of target combinations with non specific EC50 values.

Refining sensitivity predictions further based on actual drugs with specified EC50 values will be considered l With the inference function defined as above, we can create a prediction for the sensitivity of any binarized kinase target combination relative to the target set T, thus we can infer all of 2n ? c unknown sensitivities from the experimental sensitivities, creating a complete map of the sensitivities of all possible kinase target based therapies relevant for the patient. As noted previously, this complete set of sensitivity combinations constitutes the TIM. The TIM effectively captures the variations of target combina tion sensitivities across a large target set. However, we also plan to incorporate inference of the underlying nonlinear signaling tumor survival pathway that acts as the underly ing cause of tumor progression.

We address this using the TIM sensitivity values and the binarized representation of the drugs with respect to target set, Generation of TIM circuits In this subsection, we present algorithms for inference of blocks of targets whose inhibition can reduce tumor survival. The resulting combination of blocks can be rep resented as an abstract tumor survival pathway which will be termed as the TIM circuit. The inputs for this subsec tion are the inferred TIM from previous subsection and a binarization threshold for sensitivity. The output is a TIM circuit. Consider that we have generated a target set T for a sample cultured from a new patient. With the abil ity to predict the sensitivity of any target combination, we would like to use the available information to dis cern the underlying tumor survival network. Due to the nature of the functional data, which is a steady state snap shot and as such does not incorporate changes over time, we cannot infer models of a dynamic nature.

We con sider static Boolean relationships. In particular, we expect where n is a tunable inference discount parameter, where decreasing n increases Cilengitide yi and presents an optimistic estimate of sensitivity. We can extend the sensitivity inference to a non naive approach. Suppose for each target ti T, we have an asso ciated target score i.