In order to perform the binarization, we must con sider the nature of the data we are given. In particular, we are provided with an IC50 for each drug, and an EC50 value for each kinase target inhibited by the drug. Under the assumption that the primary mechanism http://www.selleckchem.com/products/Perifosine.html of tumor eradication is, in fact, the protein kinase inhibition enacted by these targeted drugs, a natural consequence would be the existence of a relationship between the IC50 and EC50 values. This rela tionship is explained as such suppose for a drug Si the IC50 value of Si and the EC50 of kinase target kj, are of similar value, then it can be reasonably assumed that kinase target kj is possibly a primary mechanism in the effectiveness of the drug.
In other words, Inhibitors,Modulators,Libraries if 50% inhibition of a kinase target directly correlates with 50% of the tumor cells losing viability, then inhibition of the kinase target is most likely one of the causes of cell death. Hence, the tar get that matches the drug IC50 is binarized as a target hit for the drug. The above assumption of direct correlation for all successful drugs is obviously an extremely restrictive assumption and will be unable to produce high accu racy predictions. Thus, the binarization scheme has to be modified to incorporate the following three factors First noises in varying Inhibitors,Modulators,Libraries magnitude will be present in the drug screen data generated by our collaborators. The noise is unavoidable, and as Inhibitors,Modulators,Libraries such, needs to be accounted for. In addition, despite the high accuracy of the drug protein interaction data procured from literature, we should still account for possible errors in the EC50 values for the numerous drugs.
Second the restrictive Inhibitors,Modulators,Libraries assumption considers that effective drugs operate on single points of failure within the patients signaling pathway. In reality, high sensitivity to a drug is often attributed to a family of related kinases or several independent kinases working synergistically over one or multiple pathways to induce tumor death. This cooperative multivariate Inhibitors,Modulators,Libraries behavior needs to be taken into account while binarizing a drug to its multiple possible targets. Third despite the high level of currently available knowledge on the biological effects of numerous targeted drugs, there remains the possibility of a drug having high sensitivity while having no known mechanisms explaining its sensitivity.
Therefore, we must consider the situation where there are latent mechanisms not considered within the dataset that are proving to be effective in some combination of treatment. This they point does not necessarily eliminate the possibility of kinase mechanisms being an important factor. We address all three concerns as follows By consid ering the log scaled EC50 values for each target and the log scaled IC50 value for each drug, we convert the mul tiplicative noise to additive noise.