High-resolution mapping of cancer cell networks using co-functional interactions

Abstract

Powerful new technologies for perturbing genetic elements have expanded the study of genetic interactions in model systems ranging from yeast to human cell lines. However, technical artifacts can confound signal across genetic screens and limit the immense potential of parallel screening approaches. To address this problem, we devised a novel PCA-based method for eliminating these artifacts and bolstering sensitivity and specificity for detection of genetic interactions. Applying this strategy to a set of >300 whole genome CRISPR screens, we report ~1 million pairs of correlated "co-functional" genes that provide finer-scale information about cell compartments, biological pathways, and protein complexes than traditional gene sets. Lastly, we employed a gene community detection approach to implicate core genes for cancer growth and compress signal from functionally related genes in the same community into a single score. This work establishes new algorithms for probing cancer cell networks and motivates the acquisition of further CRISPR screen data across diverse genotypes and cell types to further resolve the complexity of cell signaling processes.

Making data-driven hypotheses for gene functions by integrating dependency, expression, and literature data

“…Furthermore, we found clusters of genes that have no strongly correlated genes, but high median correlation values. The small range of essentiality values driving the correlations in this set of genes weakens the predictive capacity, as has been described previously for olfactory receptors (Boyle et al, 2018).…”

Convergent mutations in tissue-specific regulatory regions reveal novel cancer drivers

“…To further characterize the role of IPO9 in cancer progression, we correlated the gene-level growth effects for IPO9 with all other genes ( Figure 5G-H) following normalization, as previously described (Boyle et al, 2018) (Methods, Supplemental Table 16). Gene ontology (GO) analysis of the 168 genes for which proliferation across cell lines had a correlation greater than 0.3 with IPO9 revealed the striking enrichment of non-coding RNA metabolic processes (7.29-fold, FDR adjusted q = 7.55e-16, Supplemental Table 6) and catalytic activity on RNA (5.32-fold, q = 1.02e-4).…”

“…These values represent the normalized effect on cell growth for knockout of the given gene, such that negative values are associated with more lethal knockout. However, potential batch effects are present in the reported essentiality scores (Boyle et al, 2018) , and we sought to adjust for these in our aggregated analyses. In brief, for the co-essentiality testing with IPO9 and driver gene list analysis, the whole gene effect score matrix was normalized using a strategy to remove batch effects (Boyle et al, 2018) .…”

Integrating genetic dependencies and genomic alterations across pathways and cancer types

“…Recently, several studies have attempted to systematically identify associations between differential dependencies and genomic alterations using data from large-scale RNAi and CRISPR datasets [14,16,21,22,23,24,25,26]. The computational approaches used in these studies can be grouped into two classes.…”

“…The computational approaches used in these studies can be grouped into two classes. The first class of approaches attempt to identify associations between differential dependencies and single genomic biomarkers [14,16,21,22,23,24]. These approaches recapitulate many of the classic oncogene addiction and synthetic lethal interactions, as well as a few additional associations.…”