bitor PLX4720 and MEK inhibitors in CCLE-SkinGlioma; ERBB2 as a predictor for sensitivity to Lapatinib in CCLE-BreatOvary; ABL1 for sensitivity to ABL1 inhibitors in CCLE-Blood. This highlights CHER’s ability to derive models that not only are predictive of drug sensitivity but also helps elucidate mechanism of action. A Case Study of Sensitivity to Paclitaxel in Melanoma and Glioma Cell Lines For the AEB 071 supplier CCLE-SkinGlioma dataset, we pooled samples of melanoma and glioma together because melanocytes and neuroglia are both embryologically derived from ectoderm. PCA analysis also PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/19753314 shows the similarity between the two types of samples. Among the drugs screened, we found paclitaxel especially interesting because both melanoma and glioma cell lines show a wide range of sensitivity. Paclitaxel is a compound that targets tubulin and stabilizes the microtubule, leading to a defect in cell division. It 
has been used to treat various cancers, including melanoma, breast and ovarian cancers. 10 / 22 Context Sensitive Modeling of Cancer Drug Sensitivity Fig 6. Example of predictive model for melanoma and glioma samples. A. CHER’s model for drug sensitivity to paclitaxel. Each vertical bar represents a data of a sample. All features are gene expression profiles except PTEN, which is a mutation profile feature. AKT1 and WT1 are predictive for both melanoma and glioma. PTEN-mut, DUSP6 and USP6 are predictive features specific for melanoma whereas DUSP14 is PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/1975321/ specific for glioma. The greyed out heatmaps represents those features are not predictive for the samples. The predictions are obtained from leave-one-out procedure with the final selected features. B. Expression of WT1 is predictive of the cytotoxic drugs paclitaxel, irinotecan and topotecan, which likely due to IGF1-R activity.In other cancers, sensitivity to paclitaxel has been associated with PI3K, MAPK and NF-B pathways. CHER selects both shared and cancer-type specific features for the ACT phenotype of paclitaxel. Expression of both AKT1 and WT1 are selected as shared predictors. Interestingly, mutation of PTEN, and expression of DUSP6 and USP6 are selected only for melanoma samples. CHER selection of AKT1 and PTEN suggests the PI3K/AKT pathway is predictive of resistance to paclitaxel in melanoma cells. In our model high expression of DUSP6 and DUSP14 is predictive of resistance to paclitaxel in melanoma and glioma. DUSP6 and DUSP14 suggest high levels of MAPK activity may be involved in paclitaxel resistance, since transcription of DUSP6 is regulated by ERK and DUSP14 modulates ERK activity and negatively regulates proliferation. CHER’s prediction is supported by several studies that have shown that inhibition of ERK activity could increase sensitivity to paclitaxel in many cancers including colon, lung, and cervical cancers. Note that the elastic net fails to select any features for paclitaxel’s ACT phenotype. Resistance Predictor WT1 May Suggest Combined Therapy Because several genes in our models seem to serve as proxies to pathway activity, we ask if the association between expression of genes and drug resistance may suggest combinatorial treatments. The assumption is, if a gene is predictive of hyper-activation of an oncogenic pathway and is associated with resistance to a drug, inhibition of the pathway may overcome the resistance and sensitize individuals to the original therapy. In the CCLE-SkinGlioma dataset, CHER selects WT1 expression as a predictor for paclitaxel and topotecan. Addition
Posted inUncategorized