Robust evaluation of deep learning-based representation methods for survival and gene essentiality prediction on bulk RNA-seq data.

Deep learning (DL) has shown potential to provide powerful representations of bulk RNA-seq data in cancer research. However, there is no consensus regarding the impact of design choices of DL approaches on the performance of the learned representation, including the model architecture, the training methodology and the various hyperparameters. To address this problem, we evaluate the performance of various design choices of DL representation learning methods using TCGA and DepMap pan-cancer datasets and assess their predictive power for survival and gene essentiality predictions. We demonstrate that baseline methods achieve comparable or superior performance compared to more complex models on survival predictions tasks. DL representation methods, however, are the most efficient to predict the gene essentiality of cell lines. We show that auto-encoders (AE) are consistently improved by techniques such as masking and multi-head training. Our results suggest that the impact of DL representations and of pretraining are highly task- and architecture-dependent, highlighting the need for adopting rigorous evaluation guidelines. These guidelines for robust evaluation are implemented in a pipeline made available to the research community.

CHCHD4 regulates the expression of mitochondrial genes that are essential for tumour cell growth.

CHCHD4 (MIA40) is central to the functions of the mitochondrial disulfide relay system (DRS). CHCHD4 is essential and evolutionarily conserved. Previously, we have shown CHCHD4 to be a critical regulator of tumour cell growth. Here, we use integrated analysis of our genome-wide CRISPR/Cas9 and SILAC proteomic screening data to delineate mechanisms of CHCHD4 essentiality in cancer. We identify a shortlist of common essential genes/proteins regulated by CHCHD4, including subunits of complex I that are known DRS substrates, and genes/proteins involved in key metabolic pathways. Our study highlights a range of CHCHD4-regulated nuclear encoded mitochondrial genes/proteins essential for tumour cell growth.

Review and meta-analysis of the genetic Minimal Cut Set approach for gene essentiality prediction in cancer metabolism.

Cancer metabolism is a marvellously complex topic, in part, due to the reprogramming of its pathways to self-sustain the malignant phenotype in the disease, to the detriment of its healthy counterpart. Understanding these adjustments can provide novel targeted therapies that could disrupt and impair proliferation of cancerous cells. For this very purpose, genome-scale metabolic models (GEMs) have been developed, with Human1 being the most recent reconstruction of the human metabolism. Based on GEMs, we introduced the genetic Minimal Cut Set (gMCS) approach, an uncontextualized methodology that exploits the concepts of synthetic lethality to predict metabolic vulnerabilities in cancer. gMCSs define a set of genes whose knockout would render the cell unviable by disrupting an essential metabolic task in GEMs, thus, making cellular proliferation impossible. Here, we summarize the gMCS approach and review the current state of the methodology by performing a systematic meta-analysis based on two datasets of gene essentiality in cancer. First, we assess several thresholds and distinct methodologies for discerning highly and lowly expressed genes. Then, we address the premise that gMCSs of distinct length should have the same predictive power. Finally, we question the importance of a gene partaking in multiple gMCSs and analyze the importance of all the essential metabolic tasks defined in Human1. Our meta-analysis resulted in parameter evaluation to increase the predictive power for the gMCS approach, as well as a significant reduction of computation times by only selecting the crucial gMCS lengths, proposing the pertinency of particular parameters for the peak processing of gMCS.

An interactive web application for processing, correcting, and visualizing genome-wide pooled CRISPR-Cas9 screens.

A limitation of pooled CRISPR-Cas9 screens is the high false-positive rate in detecting essential genes arising from copy-number-amplified genomics regions. To solve this issue, we previously developed CRISPRcleanR: a computational method implemented as R/python package and in a dockerized version. CRISPRcleanR detects and corrects biased responses to CRISPR-Cas9 targeting in an unsupervised fashion, accurately reducing false-positive signals while maintaining sensitivity in identifying relevant genetic dependencies. Here, we present CRISPRcleanR WebApp , a web application enabling access to CRISPRcleanR through an intuitive interface. CRISPRcleanR WebApp removes the complexity of R/python language user interactions; provides user-friendly access to a complete analytical pipeline, not requiring any data pre-processing and generating gene-level summaries of essentiality with associated statistical scores; and offers a range of interactively explorable plots while supporting a more comprehensive range of CRISPR guide RNAs' libraries than the original package. CRISPRcleanR WebApp is available at https://crisprcleanr-webapp.fht.org/.

Predicting gene knockout effects from expression data.

BACKGROUND: The study of gene essentiality, which measures the importance of a gene for cell division and survival, is used for the identification of cancer drug targets and understanding of tissue-specific manifestation of genetic conditions. In this work, we analyze essentiality and gene expression data from over 900 cancer lines from the DepMap project to create predictive models of gene essentiality. METHODS: We developed machine learning algorithms to identify those genes whose essentiality levels are explained by the expression of a small set of "modifier genes". To identify these gene sets, we developed an ensemble of statistical tests capturing linear and non-linear dependencies. We trained several regression models predicting the essentiality of each target gene, and used an automated model selection procedure to identify the optimal model and hyperparameters. Overall, we examined linear models, gradient boosted trees, Gaussian process regression models, and deep learning networks. RESULTS: We identified nearly 3000 genes for which we accurately predict essentiality using gene expression data of a small set of modifier genes. We show that both in the number of genes we successfully make predictions for, as well as in the prediction accuracy, our model outperforms current state-of-the-art works. CONCLUSIONS: Our modeling framework avoids overfitting by identifying the small set of modifier genes, which are of clinical and genetic importance, and ignores the expression of noisy and irrelevant genes. Doing so improves the accuracy of essentiality prediction in various conditions and provides interpretable models. Overall, we present an accurate computational approach, as well as interpretable modeling of essentiality in a wide range of cellular conditions, thus contributing to a better understanding of the molecular mechanisms that govern tissue-specific effects of genetic disease and cancer.

Mutually exclusive genetic interactions and gene essentiality shape the genomic landscape of primary melanoma.

Melanoma is a heterogenous malignancy with an unpredictable clinical course. Most patients who present in the clinic are diagnosed with primary melanoma, yet large-scale sequencing efforts have focused primarily on metastatic disease. In this study we sequence-profiled 524 American Joint Committee on Cancer Stage I-III primary tumours. Our analysis of these data reveals recurrent driver mutations, mutually exclusive genetic interactions, where two genes were never or rarely co-mutated, and an absence of co-occurring genetic events. Further, we intersected copy number calls from our primary melanoma data with whole-genome CRISPR screening data to identify the transcription factor interferon regulatory factor 4 (IRF4) as a melanoma-associated dependency. © 2022 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

ELIMINATOR: essentiality analysis using multisystem networks and integer programming.

A gene is considered as essential when it is indispensable for cells to grow and replicate in a certain environment. However, gene essentiality is not a structural property but rather a contextual one, which depends on the specific biological conditions affecting the cell. This circumstantial essentiality of genes is what brings the attention of scientist since we can identify genes essential for cancer cells but not essential for healthy cells. This same contextuality makes their identification extremely challenging. Huge experimental efforts such as Project Achilles where the essentiality of thousands of genes is measured together with a plethora of molecular data (transcriptomics, copy number, mutations, etc.) in over one thousand cell lines can shed light on the causality behind the essentiality of a gene in a given environment. Here, we present an in-silico method for the identification of patient-specific essential genes using constraint-based modelling (CBM). Our method expands the ideas behind traditional CBM to accommodate multisystem networks. In essence, it first calculates the minimum number of lowly expressed genes required to be activated by the cell to sustain life as defined by a set of requirements; and second, it performs an exhaustive in-silico gene knockout to find those that lead to the need of activating additional lowly expressed genes. We validated the proposed methodology using a set of 452 cancer cell lines derived from the Cancer Cell Line Encyclopedia where an exhaustive experimental large-scale gene knockout study using CRISPR (Achilles Project) evaluates the impact of each removal. We also show that the integration of different essentiality predictions per gene, what we called Essentiality Congruity Score, reduces the number of false positives. Finally, we explored our method in a breast cancer patient dataset, and our results showed high concordance with previous publications. These findings suggest that identifying genes whose activity is fundamental to sustain cellular life in a patient-specific manner is feasible using in-silico methods. The patient-level gene essentiality predictions can pave the way for precision medicine by identifying potential drug targets whose deletion can induce death in tumour cells.

Pan-cancer proteomic map of 949 human cell lines.

The proteome provides unique insights into disease biology beyond the genome and transcriptome. A lack of large proteomic datasets has restricted the identification of new cancer biomarkers. Here, proteomes of 949 cancer cell lines across 28 tissue types are analyzed by mass spectrometry. Deploying a workflow to quantify 8,498 proteins, these data capture evidence of cell-type and post-transcriptional modifications. Integrating multi-omics, drug response, and CRISPR-Cas9 gene essentiality screens with a deep learning-based pipeline reveals thousands of protein biomarkers of cancer vulnerabilities that are not significant at the transcript level. The power of the proteome to predict drug response is very similar to that of the transcriptome. Further, random downsampling to only 1,500 proteins has limited impact on predictive power, consistent with protein networks being highly connected and co-regulated. This pan-cancer proteomic map (ProCan-DepMapSanger) is a comprehensive resource available at https://cellmodelpassports.sanger.ac.uk.

CRISPR/Cas9 Screening to Identify Conditionally Essential Genes in Human Cell Lines.

Forward genetic screens across hundreds of cancer cell lines have started to define the genetic dependencies of proliferating human cells. However, most such screens have been performed in vitro with little consideration into how medium composition might affect gene essentiality. This protocol describes a method to use CRISPR/Cas9-based loss-of-function screens to ask how gene essentiality in human cell lines varies with medium composition. First, a single-guide RNA (sgRNA) library is packaged into lentivirus, and an optimal infection titer is determined for the target cells. Following selection, genomic DNA (gDNA) is extracted from an aliquot of the transduced cells. The remaining transduced cells are then screened in at least two distinct cell culture media. At the conclusion of the screening period, gDNA is collected from each cell population. Next, high-throughput sequencing is used to determine sgRNA barcode abundances from the initial and each of the final populations. Finally, an analytical pipeline is used to identify medium-essential candidate genes from these screen results.

CRISPR screens in physiologic medium reveal conditionally essential genes in human cells.

Forward genetic screens across hundreds of cancer cell lines have started to define the genetic dependencies of proliferating human cells and how these vary by genotype and lineage. Most screens, however, have been carried out in culture media that poorly reflect metabolite availability in human blood. Here, we performed CRISPR-based screens in traditional versus human plasma-like medium (HPLM). Sets of conditionally essential genes in human cancer cell lines span several cellular processes and vary with both natural cell-intrinsic diversity and the combination of basal and serum components that comprise typical media. Notably, we traced the causes for each of three conditional CRISPR phenotypes to the availability of metabolites uniquely defined in HPLM versus conventional media. Our findings reveal the profound impact of medium composition on gene essentiality in human cells, and also suggest general strategies for using genetic screens in HPLM to uncover new cancer vulnerabilities and gene-nutrient interactions.

Systematic mapping of cancer cell target dependencies using high-throughput drug screening in triple-negative breast cancer.

While high-throughput drug screening offers possibilities to profile phenotypic responses of hundreds of compounds, elucidation of the cell context-specific mechanisms of drug action requires additional analyses. To that end, we developed a computational target deconvolution pipeline that identifies the key target dependencies based on collective drug response patterns in each cell line separately. The pipeline combines quantitative drug-cell line responses with drug-target interaction networks among both intended on- and potent off-targets to identify pharmaceutically actionable and selective therapeutic targets. To demonstrate its performance, the target deconvolution pipeline was applied to 310 small molecules tested on 20 genetically and phenotypically heterogeneous triple-negative breast cancer (TNBC) cell lines to identify cell line-specific target mechanisms in terms of cytotoxic and cytostatic drug target vulnerabilities. The functional essentiality of each protein target was quantified with a target addiction score (TAS), as a measure of dependency of the cell line on the therapeutic target. The target dependency profiling was shown to capture inhibitory information that is complementary to that obtained from the structure or sensitivity of the drugs. Comparison of the TAS profiles and gene essentiality scores from CRISPR-Cas9 knockout screens revealed that certain proteins with low gene essentiality showed high target addictions, suggesting that they might be functioning as protein groups, and therefore be resistant to single gene knock-out. The comparative analysis discovered protein groups of potential multi-target synthetic lethal interactions, for instance, among histone deacetylases (HDACs). Our integrated approach also recovered a number of well-established TNBC cell line-specific drivers and known TNBC therapeutic targets, such as HDACs and cyclin-dependent kinases (CDKs). The present work provides novel insights into druggable vulnerabilities for TNBC, and opportunities to identify multi-target synthetic lethal interactions for further studies.

Essentiality and Transcriptome-Enriched Pathway Scores Predict Drug-Combination Synergy.

In the prediction of the synergy of drug combinations, systems pharmacology models expand the scope of experiment screening and overcome the limitations of current computational models posed by their lack of mechanical interpretation and integration of gene essentiality. We therefore investigated the synergy of drug combinations for cancer therapies utilizing records in NCI ALMANAC, and we employed logistic regression to test the statistical significance of gene and pathway features in that interaction. We trained our predictive models using 43 NCI-60 cell lines, 165 KEGG pathways, and 114 drug pairs. Scores of drug-combination synergies showed a stronger correlation with pathway than gene features in overall trend analysis and a significant association with both genes and pathways in genome-wide association analyses. However, we observed little overlap of significant gene expressions and essentialities and no significant evidence that associated target and non-target genes and their pathways. We were able to validate four drug-combination pathways between two drug combinations, Nelarabine-Exemestane and Docetaxel-Vermurafenib, and two signaling pathways, PI3K-AKT and AMPK, in 16 cell lines. In conclusion, pathways significantly outperformed genes in predicting drug-combination synergy, and because they have very different mechanisms, gene expression and essentiality should be considered in combination rather than individually to improve this prediction.

Systematic analysis of bypass suppression of essential genes.

Essential genes tend to be highly conserved across eukaryotes, but, in some cases, their critical roles can be bypassed through genetic rewiring. From a systematic analysis of 728 different essential yeast genes, we discovered that 124 (17%) were dispensable essential genes. Through whole-genome sequencing and detailed genetic analysis, we investigated the genetic interactions and genome alterations underlying bypass suppression. Dispensable essential genes often had paralogs, were enriched for genes encoding membrane-associated proteins, and were depleted for members of protein complexes. Functionally related genes frequently drove the bypass suppression interactions. These gene properties were predictive of essential gene dispensability and of specific suppressors among hundreds of genes on aneuploid chromosomes. Our findings identify yeast's core essential gene set and reveal that the properties of dispensable essential genes are conserved from yeast to human cells, correlating with human genes that display cell line-specific essentiality in the Cancer Dependency Map (DepMap) project.

Integrated regulatory models for inference of subtype-specific susceptibilities in glioblastoma.

Glioblastoma multiforme (GBM) is a highly malignant form of cancer that lacks effective treatment options or well-defined strategies for personalized cancer therapy. The disease has been stratified into distinct molecular subtypes; however, the underlying regulatory circuitry that gives rise to such heterogeneity and its implications for therapy remain unclear. We developed a modular computational pipeline, Integrative Modeling of Transcription Regulatory Interactions for Systematic Inference of Susceptibility in Cancer (inTRINSiC), to dissect subtype-specific regulatory programs and predict genetic dependencies in individual patient tumors. Using a multilayer network consisting of 518 transcription factors (TFs), 10,733 target genes, and a signaling layer of 3,132 proteins, we were able to accurately identify differential regulatory activity of TFs that shape subtype-specific expression landscapes. Our models also allowed inference of mechanisms for altered TF behavior in different GBM subtypes. Most importantly, we were able to use the multilayer models to perform an in silico perturbation analysis to infer differential genetic vulnerabilities across GBM subtypes and pinpoint the MYB family member MYBL2 as a drug target specific for the Proneural subtype.

DrugSniper, a Tool to Exploit Loss-Of-Function Screens, Identifies CREBBP as a Predictive Biomarker of VOLASERTIB in Small Cell Lung Carcinoma (SCLC).

The development of predictive biomarkers of response to targeted therapies is an unmet clinical need for many antitumoral agents. Recent genome-wide loss-of-function screens, such as RNA interference (RNAi) and CRISPR-Cas9 libraries, are an unprecedented resource to identify novel drug targets, reposition drugs and associate predictive biomarkers in the context of precision oncology. In this work, we have developed and validated a large-scale bioinformatics tool named DrugSniper, which exploits loss-of-function experiments to model the sensitivity of 6237 inhibitors and predict their corresponding biomarkers of sensitivity in 30 tumor types. Applying DrugSniper to small cell lung cancer (SCLC), we identified genes extensively explored in SCLC, such as Aurora kinases or epigenetic agents. Interestingly, the analysis suggested a remarkable vulnerability to polo-like kinase 1 (PLK1) inhibition in CREBBP-mutant SCLC cells. We validated this association in vitro using four mutated and four wild-type SCLC cell lines and two PLK1 inhibitors (Volasertib and BI2536), confirming that the effect of PLK1 inhibitors depended on the mutational status of CREBBP. Besides, DrugSniper was validated in-silico with several known clinically-used treatments, including the sensitivity of Tyrosine Kinase Inhibitors (TKIs) and Vemurafenib to FLT3 and BRAF mutant cells, respectively. These findings show the potential of genome-wide loss-of-function screens to identify new personalized therapeutic hypotheses in SCLC and potentially in other tumors, which is a valuable starting point for further drug development and drug repositioning projects.

Combined gene essentiality scoring improves the prediction of cancer dependency maps.

BACKGROUND: Probing genetic dependencies of cancer cells can improve our understanding of tumour development and progression, as well as identify potential drug targets. CRISPR-Cas9-based and shRNA-based genetic screening are commonly utilized to identify essential genes that affect cancer growth. However, systematic methods leveraging these genetic screening techniques to derive consensus cancer dependency maps for individual cancer cell lines are lacking. FINDING: In this work, we first explored the CRISPR-Cas9 and shRNA gene essentiality profiles in 42 cancer cell lines representing 10 cancer types. We observed limited consistency between the essentiality profiles of these two screens at the genome scale. To improve consensus on the cancer dependence map, we developed a computational model called combined essentiality score (CES) to integrate the genetic essentiality profiles from CRISPR-Cas9 and shRNA screens, while accounting for the molecular features of the genes. We found that the CES method outperformed the existing gene essentiality scoring approaches in terms of ability to detect cancer essential genes. We further demonstrated the power of the CES method in adjusting for screen-specific biases and predicting genetic dependencies in individual cancer cell lines. INTERPRETATION: Systematic comparison of the CRISPR-Cas9 and shRNA gene essentiality profiles showed the limitation of relying on a single technique to identify cancer essential genes. The CES method provides an integrated framework to leverage both genetic screening techniques as well as molecular feature data to determine gene essentiality more accurately for cancer cells.

An Evaluation of Machine Learning Approaches for the Prediction of Essential Genes in Eukaryotes Using Protein Sequence-Derived Features.

The availability of whole-genome sequences and associated multi-omics data sets, combined with advances in gene knockout and knockdown methods, has enabled large-scale annotation and exploration of gene and protein functions in eukaryotes. Knowing which genes are essential for the survival of eukaryotic organisms is paramount for an understanding of the basic mechanisms of life, and could assist in identifying intervention targets in eukaryotic pathogens and cancer. Here, we studied essential gene orthologs among selected species of eukaryotes, and then employed a systematic machine-learning approach, using protein sequence-derived features and selection procedures, to investigate essential gene predictions within and among species. We showed that the numbers of essential gene orthologs comprise small fractions when compared with the total number of orthologs among the eukaryotic species studied. In addition, we demonstrated that machine-learning models trained with subsets of essentiality-related data performed better than random guessing of gene essentiality for a particular species. Consistent with our gene ortholog analysis, the predictions of essential genes among multiple (including distantly-related) species is possible, yet challenging, suggesting that most essential genes are unique to a species. The present work provides a foundation for the expansion of genome-wide essentiality investigations in eukaryotes using machine learning approaches.

Drug Repurposing for Paracoccidioidomycosis Through a Computational Chemogenomics Framework.

Paracoccidioidomycosis (PCM) is the most prevalent endemic mycosis in Latin America. The disease is caused by fungi of the genus Paracoccidioides and mainly affects low-income rural workers after inhalation of fungal conidia suspended in the air. The current arsenal of chemotherapeutic agents requires long-term administration protocols. In addition, chemotherapy is related to a significantly increased frequency of disease relapse, high toxicity, and incomplete elimination of the fungus. Due to the limitations of current anti-PCM drugs, we developed a computational drug repurposing-chemogenomics approach to identify approved drugs or drug candidates in clinical trials with anti-PCM activity. In contrast to the one-drug-one-target paradigm, our chemogenomics approach attempts to predict interactions between drugs, and Paracoccidioides protein targets. To achieve this goal, we designed a workflow with the following steps: (a) compilation and preparation of Paracoccidioides spp. genome data; (b) identification of orthologous proteins among the isolates; (c) identification of homologous proteins in publicly available drug-target databases; (d) selection of Paracoccidioides essential targets using validated genes from Saccharomyces cerevisiae; (e) homology modeling and molecular docking studies; and (f) experimental validation of selected candidates. We prioritized 14 compounds. Two antineoplastic drug candidates (vistusertib and BGT-226) predicted to be inhibitors of phosphatidylinositol 3-kinase TOR2 showed antifungal activity at low micromolar concentrations (<10 µM). Four antifungal azole drugs (bifonazole, luliconazole, butoconazole, and sertaconazole) showed antifungal activity at low nanomolar concentrations, validating our methodology. The results suggest our strategy for predicting new anti-PCM drugs is useful. Finally, we could recommend hit-to-lead optimization studies to improve potency and selectivity, as well as pharmaceutical formulations to improve oral bioavailability of the antifungal azoles identified.

TranscriptAchilles: a genome-wide platform to predict isoform biomarkers of gene essentiality in cancer.

BACKGROUND: Aberrant alternative splicing plays a key role in cancer development. In recent years, alternative splicing has been used as a prognosis biomarker, a therapy response biomarker, and even as a therapeutic target. Next-generation RNA sequencing has an unprecedented potential to measure the transcriptome. However, due to the complexity of dealing with isoforms, the scientific community has not sufficiently exploited this valuable resource in precision medicine. FINDINGS: We present TranscriptAchilles, the first large-scale tool to predict transcript biomarkers associated with gene essentiality in cancer. This application integrates 412 loss-of-function RNA interference screens of >17,000 genes, together with their corresponding whole-transcriptome expression profiling. Using this tool, we have studied which are the cancer subtypes for which alternative splicing plays a significant role to state gene essentiality. In addition, we include a case study of renal cell carcinoma that shows the biological soundness of the results. The databases, the source code, and a guide to build the platform within a Docker container are available at GitLab. The application is also available online. CONCLUSIONS: TranscriptAchilles provides a user-friendly web interface to identify transcript or gene biomarkers of gene essentiality, which could be used as a starting point for a drug development project. This approach opens a wide range of translational applications in cancer.

Complex modifier landscape underlying genetic background effects.

The phenotypic consequence of a given mutation can be influenced by the genetic background. For example, conditional gene essentiality occurs when the loss of function of a gene causes lethality in one genetic background but not another. Between two individual Saccharomyces cerevisiae strains, S288c and Σ1278b, ∼1% of yeast genes were previously identified as "conditional essential." Here, in addition to confirming that some conditional essential genes are modified by a nonchromosomal element, we show that most cases involve a complex set of genomic modifiers. From tetrad analysis of S288C/Σ1278b hybrid strains and whole-genome sequencing of viable hybrid spore progeny, we identified complex sets of multiple genomic regions underlying conditional essentiality. For a smaller subset of genes, including CYS3 and CYS4, each of which encodes components of the cysteine biosynthesis pathway, we observed a segregation pattern consistent with a single modifier associated with conditional essentiality. In natural yeast isolates, we found that the CYS3/CYS4 conditional essentiality can be caused by variation in two independent modifiers, MET1 and OPT1, each with roles associated with cellular cysteine physiology. Interestingly, the OPT1 allelic variation appears to have arisen independently from separate lineages, with rare allele frequencies below 0.5%. Thus, while conditional gene essentiality is usually driven by genetic interactions associated with complex modifier architectures, our analysis also highlights the role of functionally related, genetically independent, and rare variants.