Abstract Introduction: In clinical oncology, with breast cancer in the front line, - cancer genome sequencing is entering clinical diagnostics; - novel drugs target signaling pathways that drive tumor growth; - metastasized cancer is expected to change into a chronic disease, requiring therapy response monitoring and repeated personalized choice of targeted therapy. Hence, there is a need to clinically interpret available genomic data in order to diagnose which signaling pathway is driving tumor growth in a patient, including the causative mutation, such that appropriate targeted drugs can be chosen. Currently available pathway analysis tools and databases are not suitable for this. Companion diagnostics: To address the above need, we are developing computational models of signaling pathways, which can be used as companion diagnostic tools to translate genomic data into meaningful clinical information to enable selection of targeted drugs. The aim is to develop such models for the 10 to 20 most relevant oncogenic pathways, and use them to predict which one is most likely to drive tumor growth in an individual patient, including the probable underlying genomic defect. A model of the Wnt pathway: As a proof of concept, we have built a computational model of the Wnt pathway, which is active in colon adenoma and a main player in colon cancer, but also relevant in breast cancer. Our first generation model of the Wnt pathway covers its transcriptional program, for which we carefully selected a list of Wnt target genes. We have modeled this pathway by a Bayesian network, which interprets the expression levels of the target genes (from Affymetrix U133Plus2.0 arrays), and infers a probability that the Wnt pathway is active in a certain patient tumor sample. The parameters of the model are partly derived from literature and partly fitted on experimental data. Results: We first fitted a model on data from six pairs of Wnt knockdown experiments on colon cancer cell line LS174T, and tested it on a public set of 32 normal colon samples plus 32 colon adenomas from patients (GSE8671). This model perfectly predicted no Wnt activity in the normal samples and an active Wnt pathway in the adenomas. Next, we fitted the model on these 64 patient samples, and tested it on a set of 145 normal colon and colon cancer samples from patients (GSE20916). The model again predicted no Wnt activity in all 44 normal samples, and an active Wnt pathway in all cancer samples, except for four of the 36 samples from surgically removed colon carcinomas, which most likely are the most heterogeneous. Thirdly, we took the latter model, trained on colon samples, and tested it on three breast cancer data sets, one with 51 breast cancer cell lines (GSE12777) and two sets of breast cancer samples from patients (GSE12276, n = 204; GSE21653, n = 266). Also on this different tissue type, our model correctly predicted Wnt activity for the few breast cancer cell lines that are known to have an active Wnt pathway. For the two patient studies, a significantly higher number of samples of the basal subtype were predicted to have an active Wnt pathway compared to the other subtypes (Fisher's exact test: p = 0.021 for GSE12276; p = 2.7e-5 for GSE21653), in line with increasing evidence for Wnt activation in the basal subtype. Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr P5-04-03.