Biological network topology features predict gene dependencies in cancer cell lines

Abstract

In this paper we explore computational approaches that enable us to identify genes that have become essential in individual cancer cell lines. Using recently published experimental cancer cell line gene essentiality data, human protein-protein interaction (PPI) network data and individual cellline genomic alteration data we have built a range of machine learning classification models to predict cell line specific acquired essential genes. Genetic alterations found in each individual cell line were modelled by removing protein nodes to reflect loss of function mutations and changing the weights of edges in each PPI to reflect gain of function mutations and gene expression changes.We found that PPI networks can be used to successfully classify human cell line specific acquired essential genes within individual cell lines and between cell lines, even across tissue types with AUC ROC scores of between 0.75 and 0.85. Our novel perturbed PPI network models further improved prediction power compared to the base PPI model and are shown to be more sensitive to genes on which the cell becomes dependent as a result of other changes. These improvements offer opportunities for personalised therapy with each individual's cancer cell dependencies presenting a potential tailored drug target.The overriding motivation for predicting cancer cell line specific acquired essential genes is to provide a low-cost approach to identifying personalised cancer drug targets without the cost of exhaustive loss of function screening.

Prediction of cancer dependencies from expression data using deep learning

“…In order to be able to predict gene dependencies across cell lines, we need to solve a multivariate regression task, in which every cell line is a regression's dependent variable. We note that while a previous work approached this problem, it was focused on a very small number of specific cell lines and used an obsolete version of the dependency data, 8 hence we could not readily compare to it.…”

“…Previous work in this domain was mainly focused on small sets of target cell lines or drugs. [8][9][10][11] Related work for drug sensitivity prediction focused on inferring the sensitivity of new cell lines to a fixed set of drugs, leaving open the reverse prediction of cell line sensitivity to new, previously unobserved, drugs. [12][13][14][15] Here we aim to address this question of the predictive power of transcriptomics data in estimating gene dependencies and drug sensitivity.…”

A machine learning and directed network optimization approach to uncover TP53 regulatory patterns

“…Gene networks are a powerful tool to study complex diseases such as cancer 3 , 4 , 5 , 6 , 7 , 8 ; however, they have not been applied systematically to characterize different mutations. Different approaches can be chosen for gene networks analysis and inference, which provide different insights.…”

Predicting genetic interactions, cell line dependencies and drug sensitivities with variational graph auto-encoder

“…Genetic interactions have been predicted based on gene ontology information (Yu et al, 2016;Ma et al, 2018), mutation and expression data (Lee et al, 2018), protein-protein interaction (PPI) data (Hao et al, 2021) and by specifically-designed deep learning models (Ma et al, 2018;Cai et al, 2020). Gene dependencies have been predicted based on expression information (Itzhacky and Sharan, 2021;Lin and Lichtarge, 2021), pathway information (Lin and Lichtarge, 2021), genetic essentiality profiles (Wang et al, 2019), and PPI and genomic alteration information (Benstead-Hume et al, 2019). Drug sensitivity data have been predicted based drug structure information combined with gene ontology information (Kuenzi et al, 2020) or gene expression data (Wang et al, 2017;Zhang et al, 2018;Choi et al, 2020;Itzhacky and Sharan, 2021).…”