Predicting and characterizing a cancer dependency map of tumors with deep learning

Abstract

Genome-wide loss-of-function screens have revealed genes essential for cancer cell proliferation, called cancer dependencies. It remains challenging to link cancer dependencies to the molecular compositions of cancer cells or to unscreened cell lines and further to tumors. Here, we present DeepDEP, a deep learning model that predicts cancer dependencies using integrative genomic profiles. It uses a unique unsupervised pretraining that captures unlabeled tumor genomic representations to improve the learning of cancer dependencies. We demonstrated DeepDEP's improvement over conventional machine learning methods and validated the performance with three independent datasets. By systematic model interpretations, we extended the current dependency maps with functional characterizations of dependencies and a proof-of-concept in silico assay of synthetic essentiality. We applied DeepDEP to pan-cancer tumor genomics and built the first pan-cancer synthetic dependency map of 8000 tumors with clinical relevance. In summary, DeepDEP is a novel tool for investigating cancer dependency with rapidly growing genomic resources.

TCGADEPMAP– Mapping Translational Dependencies and Synthetic Lethalities within The Cancer Genome Atlas

“…Despite the abundance of annotated features within the DEPMAP, two recent studies estimated that expression-based predictive modeling of gene essentiality outperformed all other features (e.g., somatic mutations, copy number, methylation, etc.) ( 24 , 25 ). An added benefit of building predictive models using transcriptomics is that it is robust and broadly captured across TCGA and many other clinically studies ( 26 ), as well as PDX studies ( 27 , 28 ).…”

“…In addition to illuminating lineage and oncogenic dependencies, the DEPMAP has dramatically expanded the list of potential synthetic lethalities (i.e., the loss of a gene sensitizes tumor cells to inhibition of a functional redundant gene within the same pathway) ( 6 , 16 , 17 , 33 , 34 ). However, one of the current limitations of DEPMAP is that the available cancer cell models do not yet fully recapitulate the genetic and molecular diversity of TCGA patients ( 25 ). Thus, we assessed the landscape of predicted synthetic lethalities with loss-of-function (LOF) events (damaging mutations or deletions) in the TCGA DEPMAP using several multivariate analyses to predict lethal partners.…”

“…The translational dependency maps (i.e., TCGA DEPMAP , PDXE DEPMAP , and GTEX DEPMAP ) were built using expression-based predictive models of gene essentiality, as transcriptomic data reportedly outperforms other features (e.g., somatic mutations, copy number, etc.) in predicting dependencies ( 25 , 43 ). Another strength of expression-based predictive modeling is that it can be applied to the transcriptomic profiles of normal tissues (e.g., GTEX DEPMAP ) that do not have appreciable levels of the somatic alterations that are observed in malignant tissues ( 29 ).…”

“…Our rationale for using elastic-net ML is that it is suited to highly dimensional data with correlated variables ( 30 ) and expression-based predictive modeling of gene essentiality outperforms genomic features (e.g., somatic mutations, copy number, methylation, etc.) ( 25 , 43 ). Likewise, the DeepDEP authors also highlighted that a simplified DL model using expression only (Exp-DeepDEP) performed comparably well to DeepDEP ( 25 ), suggesting that ML and DL dependency models are both dominated by expression profiling ( 25 ).…”

“…( 25 , 43 ). Likewise, the DeepDEP authors also highlighted that a simplified DL model using expression only (Exp-DeepDEP) performed comparably well to DeepDEP ( 25 ), suggesting that ML and DL dependency models are both dominated by expression profiling ( 25 ). Intriguingly, despite the dominance of expression in the DeepDEP dependency models, it was unable to recapitulate the robust clustering of tumors and cell lines by t-SNE that was observed by expression alone ( 25 ).…”