We will accomplish our overall goal of developing computational models that predict genome-wide essentiality based on the molecular characteristics of individual tumors via the completion of two Specific Aims:
Aim 1. Determine essential genes in lung cancer and melanoma. The goal here is to establish the ‘ground truth’ of essentiality in a sufficiently large number of cell lines so as to capture the genetic diversity of the disease that is observed in patients. This will involve large-scale RNAi data generation, and also the solving of a number of computational challenges inherent in RNAi screening. Specifically, we will:
- Generate genome-wide RNAi viability data in 200 non-small cell lung cancer cell lines.
- Establish mechanism of loss of viability (e.g., apoptosis vs. growth arrest)
- Develop computational methods that address inherently noisy RNAi data (e.g., off-target and partial knock-down effects).
- Develop principled analytical framework that establishes the optimal number of shRNAs per gene, number of cell lines and number of replicates for genome-wide RNAi screening.
- Determine the pattern of sensitivity of the cell lines to inhibitors of pathways known to be important in NSCLC (e.g., PI3K, EGFR inhibitors).
- Develop computational solutions to analytical challenges facing drug sensitivity determination.
- Apply the computational approaches developed above to conduct a similar screen of 100 melanoma cell lines.
A schematic of the technology that we have recently developed (Luo et al, PNAS, 2008) to conduct genome-wide RNAi screens is shown below:
Aim 2. Develop and test predictors of essentiality. The goal here is to develop computational models that accurately predict the patterns of essentiality discovered in Aim 1. This will involve i) performing extensive molecular characterization of the cell lines for which essentiality is known, ii) developing increasingly complex models that use the molecular features (including their network relationships and prior biological knowledge) to predict essentiality, iii) testing those models on an independent set of cell lines and refining the models accordingly, and iv) testing the refined models in physiologically relevant pre-clinical models. These activities will include:
- Profiling the 100 cell lines with respect to their mRNA, miRNA and long non-coding RNA (lncRNA) expression.
- Measuring the phospho-tyrosine protein and metabolite profiles of the cell lines.
- Profiling gene copy number and somatic mutation status.
- Developing sophisticated computational modeling approaches using the above data to predict gene essentiality (as determined in Aim 1).
- Developing predictive models that incorporate increasing complexity based on incorporation of network relationships and prior biological understanding.
- Testing the accuracy of these predictions experimentally and using the results to refine the models.
- Testing the refined models in physiologically relevant experimental systems of human cancer (primary human tumor slice short-term cultures).
- Extending these predictive approaches to the prediction of synergistic combinations of genetic (or pharmacologic) perturbation to be tested by the approaches of Aim 1.
At the conclusion of this project, we expect to have advanced the field in several dimensions. First, we will have established experimental approaches togenome-wide functional perturbation in the context of genetic diversity. Second, we will have addressed the major computational challenges that have been bottlenecks in interpreting such experimental results. Third, we will have developed, tested and refined computational models that a) provide insight into cancer biology, b) provide insight into therapeutic strategies for cancer, and c) establish a general framework for predicting other facets of cancer cell behavior. In addition, this project will serve as the vehicle for exploring the new frontiers of cancer systems biology generally, and for training a next generation of cancer system biologists.