Computational analysis of omics data towards the identification of essential genes in cancer cells

Abstract

Essential genes are often associated with fundamental functions in diverse biological processes across species. Previous studies demonstrated that both environmental and genetic factors could determine the essentiality of genes. Profiling the essential genes allows us to discover the genes responsible for phenotypes of interest. Therefore, such a genetic/environment – essentiality – phenotypes axis is particularly useful in clinical applications such as antibiotics and anti-cancer drug development. Recent technological development in CRISPR based gene editing enables accurate and large-scale profiling of gene essentiality in different cellular contexts, particular in cancer cells. Due to efficacy problem in sgRNA library design in genome-wide CRISPR gene knockout experiment, accurate detection of essential genes from such assay would be crucial for downstream analysis. I developed a two-component mixture model to identify the essential genes from genome-wide gene knockout assays. The method could be applied to various cellular contexts without the requirement of prior knowledge of essential genes. By its application to cancer cell lines, the newly developed method was demonstrated with its performance in stably and accurately identifying essential genes. Moreover, applying the method to hundreds of cancer cell lines provides a glimpse of differentially required putative cancer-specific essential genes, which could be useful in guidance of targeting therapy in human cancers. Since differentially required essential genes are potentially associated with the genetic background of each cancer cell, I was interested in exploring whether differential essentiality of genes could be predicted by a small number of functionally active genes. With matched data between copy number alteration and gene expression, I utilized machine learning algorithms to select possible predictive feature genes to predict the essentiality of target genes in cancer cell lines. The results show that the essentiality of certain genes could be readily predicted using a handful of other genes. However, more feature genes and other information might be required to make robust predictions for other target genes. This study reveals that functional relations of feature genes to target genes substantially contributed to the predictability of essentiality for many genes. In cancer cells, the essentiality of genes often involves gene interactions which may not be obvious in normal cells. Synthetic lethal (SL) interaction provides a framework to identify synthetic essentiality of genetic interactions of two genes. Functional disruption of genes in SL relationships is the prerequisites for such compensatory interactions to occur, which is the hallmark of cancer. Here, I developed an algorithm to predict putative gene pairs in SL interactions. The predicted SL interactions were then verified by other studies. And biological annotation of these putative SL interactions, I found that most of the interactions were enriched within pathways. The analysis of SL interactions could be used to identify potential drug targets in cancer cells with specific gene loss.