By incorporating the drug target interaction data and sensitivities of training drugs with genomic signatures, we were able to achieve a cor relation coefficient of 0. 79 for prediction of Erlotinib sensi tivity using 10 fold cross validation. The result illustrates the fundamental concept of the importance of drug target interaction and functional selleck compound data under which we develop the sensitivity prediction method presented in this paper. By developing a framework around the functional and tar get information extracted from the primary tumor drug screen performed by our collaborators, we seek to develop a cohesive approach to sensitivity prediction and com bination therapy design. This necessitates the generation of the tumor pathway structure for individual patients to decide on the target inhibitors for therapy based on the personalized patient pathways.

We envision that the overall schematic of the design of personalized pathways and personalized therapy will be similar to the workflow shown in Figure 1. The explanations of the various steps in the design process are as follows, The primary contributions of this paper are, methods for extraction of numerically relevant drug targets from single run drug screens, design of the personalized TIM circuit based on drug perturbation data, algo rithms for sensitivity prediction of a new drug or drug cocktail, validation over canine osteosarcoma primary tumors and pathway flow inference using sequen tial protein expression measurements. The scope of the present article is concentrated around steps B, C and D of Figure 1.

The perturbation data required for our proposed method originates from a drug screen consisting of 60 small molecule inhibitors with quantified kinase interac tion behaviors. This drug screen, denoted Drug Screen Version 1. 0, consists of two sets of data, The first set is the experimentally generated drug sensitivities provided as 50% inhibitory concentration values. The IC50 values denote the amount of a drug required to reduce the population of cancerous cells in vitro by half. The sen sitivity values are expected to change during each new cell line tumor culture experiment. The generation of the sensitivities in step C can be done within 72 hours of ini tial biopsy using drug sensitivity assays which is a period of limited cell divisions for most primary cultures.

Thus, the estimated personalized maps may be closer to real time circuits in cancer cells akin to the signaling found in an untreated patient within a day or two after biopsy, and not the evolving Brefeldin_A consensus pattern of signaling for grow ing and dividing tumor cells as subpopulations emerge with increased fitness in vitro. In addition, the drug screen contains experimentally derived half maximal con centration values for the interaction of each drug and each kinase target.