COMPREHENSIVE IN SILICO ASSESSMENT OF CANCER CELL LINE FIDELITY

Since 1971, when US president Richard Nixon declared a “war against cancer”, an estimated $200 billion has been spend on research and drug development for cancer. Globally, the cancer incidence and the mortality rates have declined over the past decade, owing to progress made in healthcare and biological research. In order to study the biological properties of cancers and to enable novel drug discovery, in-vitro model systems are used and the most common amongst these are cancer cell lines. These are primary-tumor derived cells that are propagated in the lab, and are used to study both the biological and pharmacological aspects of cancer. The use of cell lines in research, in the field of cancer diagnostics, has resulted in a number of drugs that are now available in the market to combat cancers. Since the isolation of the first cancer cell line, HeLa, in 1951, a number of cell lines are available for various types of cancers and these have been around for decades now. Cell lines in culture not only lose the properties of their cells of origin gradually, they also acquire mutations over each passage that causes the lines to behave very differently as compared to primary tumors. Additionally, there is competition among populations of cells in culture and it is widely believed that cell lines that are presently available are largely clonal populations of those cells that grow faster and survive better as compared to others. This results in loss of heterogeneity that is characteristic of in-vivo cancers. In order to quantitatively assess the fidelity of these cell lines, we use a modified version of the network biology platform CellNet to construct tumor type specific gene regulatory networks for twelve common primary cancers and train random forest classifiers. Using expression data from 917 cell lines as the query and comparing their GRNs (Gene Regulatory Networks) against the tumor type specific classifiers, the probability scores for each of these cell lines to resemble primary tumors are calculated. Those cell lines that show a high classification score to their tumor type of origin are considered ideal in vitro models for that cancer. The GRN status and network influence scores were also calculated for cell lines and using these statistics and CellNet predictions, in-vitro models for cancer research are suggested. Additionally, we propose the use of Cancer CellNet as a clinical diagnostic platform to resolve the tumor identity in unknown cases and demonstrate its ability to obtain this information using tumor samples where the cells of origin of cancer remain unknown