Approaches used by PREDICT consortium members have been designed to avoid or overcome the various pitfalls of high-throughput associative studies of gene expression datasets [3, 43] in order to develop the next generation of prognostic and predictive biomarkers. The potential to rapidly identify predictive biomarkers of drug response in tumor tissue to define sensitive and resistant patient cohorts has recently been accelerated through advances in functional genomics techniques that have been intensively developed by the PREDICT consortium using large scale RNAi screening approaches [44–47].
Through the use of this technology, the consortium has identified genes regulating response and resistance to common cytotoxic agents used in cancer medicine [40, 46, 48–50]. Through the integrative genomics analysis of these functional RNAi datasets in breast and ovarian cancer, we have identified regulators of mitotic arrest and ceramide metabolism as mediators of taxane resistance and confirmed their relevance in clinical trial cohorts [40, 46, 48–50]. For example, silencing of the ceramide transporter CERT was shown to confer sensitivity to paclitaxel across multiple cancer cell lines and follow-up analysis revealed that CERT was overexpressed in two separate paclitaxel-resistant cell lines. Analysis of microarray expression data from the OV-01 clinical trial revealed that over-expression of CERT occurred in ovarian cancers from patients with paclitaxel resistant disease, suggesting a role for this gene product in the regulation of response to paclitaxel in vivo .
Successful integration of RNAi functional genomics screening results with tumor gene expression data in order to identify a predictor of neoadjuvant paclitaxel response in breast cancer was dependent on the identification of gene coexpression modules representative of mitotic arrest and ceramide metabolic pathways relevant to drug response. The combination of these modules into a 'functional metagene' shows promise as a paclitaxel-specific predictive biomarker  that is predictive of pathological complete response to paclitaxel in breast cancer with a high sensitivity and specificity (area under the receiver operating characteristic curve (AUC) = 0.8) , outperforming any other clinical or molecular predictor of paclitaxel sensitivity identified to date.
Further supporting integrative genomics approaches to the identification of novel drug response mechanisms in vivo, we have integrated complex cancer datasets (gene expression and copy number data) to identify a particular chromosomal region that contributes to anthracycline resistance when amplified in breast cancer. Two causative genes, LAPTM4B and YWHAZ, were identified from this region: one is a known anti-apoptosis gene, and one is a novel gene affecting drug transport. These genes are strongly predictive of anthracycline resistance, and rigorous clinical evaluation is ongoing .
We have also demonstrated that molecular hypotheses can be utilized to predict drug response in vivo. We formed a rational hypothesis about drug mechanism to suggest a predictor of response to cisplatin. Briefly, we noticed links between BRCA1 mutations, cisplatin sensitivity, and DNA repair pathway competence. We developed a SNP array-based surrogate marker of DNA repair pathway competence and found that it strongly predicted for neoadjuvant cisplatin pathological complete response in a small cohort of estrogen receptor-negative/progesterone receptor-negative/ERBB2-negative breast cancer patients . We have derived a robust gene expression signature of chromosomal instability, which is prognostic in several types of solid tumor  and predictive of paclitaxel resistance in ovarian cancer . We have also identified a blood-based gene expression biomarker of early-stage Parkinson's disease , which is currently being validated in a larger study, and have integrated diverse genomic data sets to generate an atlas of disease-associated protein complexes, several of which were novel .
These studies highlight the power of comprehensive functional genomics datasets combined with monotherapy clinical trial tumor genomics datasets to illuminate the clinical relevance of specific genes to individual patient drug sensitivity. Furthermore, the studies provide robust and efficient methodological tools to accelerate predictive biomarker development and identify mechanisms of drug resistance that will be applied to biomarker discovery in RCC in this proposal.