Optimal construction of a functional interaction network from pooled library CRISPR fitness screens

Abstract

Background Functional interaction networks, where edges connect genes likely to operate in the same biological process or pathway, can be inferred from CRISPR knockout screens in cancer cell lines. Genes with similar knockout fitness profiles across a sufficiently diverse set of cell line screens are likely to be co-functional, and these “coessentiality” networks are increasingly powerful predictors of gene function and biological modularity. While several such networks have been published, most use different algorithms for each step of the network construction process. Results In this study, we identify an optimal measure of functional interaction and test all combinations of options at each step—essentiality scoring, sample variance and covariance normalization, and similarity measurement—to identify best practices for generating a functional interaction network from CRISPR knockout data. We show that Bayes Factor and Ceres scores give the best results, that Ceres outperforms the newer Chronos scoring scheme, and that covariance normalization is a critical step in network construction. We further show that Pearson correlation, mathematically identical to ordinary least squares after covariance normalization, can be extended by using partial correlation to detect and amplify signals from “moonlighting” proteins which show context-dependent interaction with different partners. Conclusions We describe a systematic survey of methods for generating coessentiality networks from the Cancer Dependency Map data and provide a partial correlation-based approach for exploring context-dependent interactions.

Recharacterization of RSL3 reveals that the selenoproteome is a druggable target in colorectal cancer

“…CRISPR gene effect scores from the DepMap 22Q2 release were first corrected using Cholesky whitening as previously described 29 . A matrix of p-values corresponding to gene-wise Pearson correlations was calculated from the whitened data, then converted to FDR values using the p.adjust() function in R. A network was then constructed from all partners within 2 edge distances from a given gene of interest, using an FDR cutoff of 0.05 to define edges/partnership.…”

Dimensionality reduction methods for extracting functional networks from large‐scaleCRISPRscreens

“…In addition to directly identifying cancer‐specific genetic dependencies, co‐essentiality between genes can be measured and used to group genes into functional modules by measuring correlations between CERES scores in the DepMap—a type of analysis pioneered in the yeast genetic interaction research community (Baryshnikova et al , 2010 ; Costanzo et al , 2016 ). Indeed, this profile similarity analysis has been directly applied to the DepMap dataset to reveal functional similarities between human genes (Boyle et al , 2018 ; Pan et al , 2018 ; Kim et al , 2019 ; Buphamalai et al , 2021 ; Wainberg et al , 2021 ; Gheorghe & Hart, 2022 ). However, previous research has posited that profile similarities in the DepMap are confounded by technical variation unrelated to the cancer‐specific phenotypes of interest (Rahman et al , 2021 ).…”

“…In addition, we propose a novel method named “onion” normalization as a general‐purpose technique for integrating multiple layers of normalized data across different hyperparameter values into a single normalized network. The goal of the proposed onion normalization methods is to enable the construction of improved gene–gene similarity networks from the DepMap dataset, which has been a major recent focus of analyses of these data (Boyle et al , 2018 ; Wainberg et al , 2021 ; Gheorghe & Hart, 2022 ) but we note is distinct from other important applications of the DepMap goals such as direct clustering of the cell lines/genes (Pan et al , 2022 ), or more focused target/drug discovery‐oriented analyses (Chiu et al , 2021 ; Ma et al , 2021 ; Shimada et al , 2021 ). We apply onion normalization using either PCA‐normalized, RPCA‐normalized, or AE‐normalized data as input.…”

Dimensionality reduction methods for extracting functional networks from large-scale CRISPR screens

“…Indeed, this profile similarity analysis has been directly applied to the DepMap dataset to reveal functional similarities between human genes (Pan, et al, 2018;Boyle, Pritchard, & Greenleaf, 2018;Wainberg, et al, 2021;Kim, et al, 2019;Buphamalai, Kokotovic, Nagy, & Menche, 2021;Gheorghe & Hart, 2022). However, previous research has posited that profile similarities in the DepMap are confounded by technical variation unrelated to the cancer-specific phenotypes of interest (Rahman, et al, 2021).…”

“…In addition to directly identifying cancer-specific genetic dependencies, co-essentiality between genes can be measured and used to group genes into functional modules by measuring correlations between CERES scores in the DepMap - a type of analysis pioneered in the yeast genetic interaction research community (Baryshnikova, et al, 2010; Costanzo, et al, 2016). Indeed, this profile similarity analysis has been directly applied to the DepMap dataset to reveal functional similarities between human genes (Pan, et al, 2018; Boyle, Pritchard, & Greenleaf, 2018; Wainberg, et al, 2021; Kim, et al, 2019; Buphamalai, Kokotovic, Nagy, & Menche, 2021; Gheorghe & Hart, 2022). However, previous research has posited that profile similarities in the DepMap are confounded by technical variation unrelated to the cancer-specific phenotypes of interest (Rahman, et al, 2021).…”