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

Abstract

CRISPR-Cas9 screens facilitate the discovery of gene functional relationships and phenotype-specific dependencies. The Cancer Dependency Map (DepMap) is the largest compendium of whole-genome CRISPR screens aimed at identifying cancer-specific genetic dependencies across human cell lines. A mitochondria-associated bias has been previously reported to mask signals for genes involved in other functions, and thus, methods for normalizing this dominant signal to improve co-essentiality networks are of interest. In this study, we explore three unsupervised dimensionality reduction methods - autoencoders, robust, and classical principal component analyses (PCA) - for normalizing the DepMap to improve functional networks extracted from these data. We propose a novel onion normalization technique to combine several normalized data layers into a single network. Benchmarking analyses reveal that robust PCA combined with onion normalization outperforms existing methods for normalizing the DepMap. Our work demonstrates the value of removing low-dimensional signals from the DepMap before constructing functional gene networks and provides generalizable dimensionality reduction-based normalization tools.

Variable efficiency of nonsense-mediated mRNA decay across human tissues, tumors and individuals

“…This revealed a prominent cluster of functional interactions (Fig. 4F) comprising NMD factor genes such as SMG5, SMG6, SMG7, UPF1 , and UPF3B , suggesting that the RPCO scores derived from CRISPR screens 67 are powerful for identifying NMD factors. Within this cluster were two candidate genes residing within 1q21.1-23.1, PMF1 and GON4L , showing notable codependency scores with SMG5 (0.06 and 0.0062, for PMF1 and GON4L , respectively) , SMG6 (0.02 and 0.006), and SMG7 (0.019 and 0.012).…”

“…To pinpoint the potential causal gene from the 30 identified candidates, we considered genetic interactions inferred from gene-level CRISPR screening data obtained from the Achilles Project 66 , allowing to infer functional links between genes by studies of codependency profiles across cell lines. Here, we used the de-biased data ( onion method, RPCO), suggested for its power for inferring gene function 67 (see Methods). Our hypothesis posits that if these candidate genes 1q21-23.1 are indeed implicated in NMD efficiency, their CRISPR knockout fitness effect should correlate with established NMD factor genes like UPF1 , UPF2 , UPF3B , SMG1 , and others; the codependency implies a functional link.…”

BaCoN (Balanced Correlation Network) improves prediction of gene buffering

“…However, improvements regarding all metrics were less pronounced than for the lineage-based corrections (Figure 2a-d). Finally, we turned to a method that we and others had independently confirmed to perform very well on co-essentiality networks -data whitening with subsequent computation of pairwise PCCs (Hassan et al, 2023;Gheorghe & Hart, 2022;Rahman et al, 2021). Data whitening can be done with different sphering methods.…”

“…Similarity between expression (co-expression) or genetic interaction signatures have been exploited to classify genes by their function in model organisms (Costanzo et al , 2016; Fischer et al , 2015; Collins et al , 2007; Eisen et al , 1998; Billmann et al , 2018; Costanzo et al , 2010). Following this concept, co-essentiality networks in human cells derived from genome-scale CRISPR screens from the Cancer Dependency Map (DepMap) enable systematic prediction of gene function as well (Wainberg et al , 2021; Hassan et al , 2023; Pan et al , 2022). The DepMap perturbs all approximately 18,000 genes in the human genome followed by the measurement of the effect of the perturbation on cell fitness.…”

“…A separate class of data normalization methods exploited the CRISPR screening-derived gene fitness effects to generate co-essentiality networks that predict similarity of gene functions (Hassan et al , 2023; Wainberg et al , 2021; Rahman et al , 2021; Boyle et al , 2018). A comparison of different normalization strategies showed that the perhaps most crucial part of the normalization strategy is the whitening of the fitness effect gene-by-cell line matrix prior to the computation of between-gene association indices (Gheorghe & Hart, 2022).…”