Prediction of cancer dependencies from expression data using deep learning

Itzhacky, Nitay; Sharan, Roded

doi:10.1039/d0mo00042f

Cited by 6 publications

(8 citation statements)

References 26 publications

(28 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Majority of these approaches did not systematically evaluate the effectiveness of different representation learning techniques from high dimensional data (e.g., qualitative evaluation of the learned embeddings) and assess their impact on the classification accuracy, even though they outperformed previous approaches. In a recent approach Itzhacky et al [37], devised a novel deep learning method for predicting gene dependencies and drug sensitivities from GE measurements. By combining dimensionality reduction strategies, they are able to learn accurate models that outperform shallow DNN or ML models.…”

Section: A Unimodal Approachesmentioning

confidence: 99%

Adversary-Aware Multimodal Neural Networks for Cancer Susceptibility Prediction From Multiomics Data

et al. 2022

View full text Add to dashboard Cite

Artificial intelligence (AI) systems are increasingly used in health and personalized care. However, the adoption of data-driven approaches in many clinical settings has been hampered due to their inability to perform in a reliable and safe manner to leverage accurate and trustworthy diagnoses. A critical and challenging usage scenario for AI is aiding the treatment of cancerous conditions. Providing accurate diagnosis for cancer is a challenging problem in precision oncology. Although machine learning (ML)-based approaches are very effective at cancer susceptibility prediction and subsequent treatment recommendations, ML models can be vulnerable to adversarial attacks. Since adversarially weak models can lead to wrong clinical recommendations, such vulnerabilities is more critical -especially when AI-guided systems are used to aid medical doctors. Therefore, it is indispensable that healthcare professionals employ trustworthy AI tools for predicting and assessing disease risks and progression. In this paper, we propose an adversaryaware multimodal convolutional autoencoder (MCAE) model for cancer susceptibility prediction from multi-omics data consisting of copy number variations (CNVs), miRNA expression, and gene expression (GE). Based on different representational learning techniques, the MCAE model learns multimodal feature representations from multi-omics data, followed by classifying the patient cohorts into different cancer types on multimodal embedding space that exhibit similar characteristics in end-to-end setting. To make the MCAE model robust to adversaries and to provide consistent diagnosis, we formulate robustness as a property, such that predictions remain stable with regard to small variations in the input. We study different adversarial attacks scenarios and take both proactive and reactive measures (e.g., adversarial retraining and identification of adversarial inputs). Experiment results show that the MCAE model based on latent representation concatenation (LRC) exhibits high confidence at predicting cancer types, giving an average precision and Matthews correlation coefficient (MCC) scores of 0.9625 and 0.8453, respectively and shows higher robustness when compared and tested with state-of-the-art approaches against different attack scenarios w.r.t. ERM and CLEVER scores. Overall, our study suggests that a well-fitted and adversarially robust model can provide consistent and reliable diagnosis for cancer.

show abstract

Section: A Unimodal Approachesmentioning

confidence: 99%

Adversary-Aware Multimodal Neural Networks for Cancer Susceptibility Prediction From Multiomics Data

et al. 2022

View full text Add to dashboard Cite

show abstract

“…Kim et al [ 10 ] developed REVEALER, a computational iterative approach that identifies groups of genomic features that together associate with a functional activation, context-specific essentiality, or drug response profile in order to understand the mechanisms which drive essentiality. More recently, Itzhacky et al [ 11 ] used a deep learning approach to simultaneously predict the essentiality of all genes, using transcriptomes from various cell lines. This full-blown approach suffers from a lack of interpretability, and cannot be used to define the set of modifier genes affecting condition-specific essentially, which was one of the main goals of the DepMap project, and is less accurate overall compared to the model presented here.…”

Section: Introductionmentioning

confidence: 99%

“…This full-blown approach suffers from a lack of interpretability, and cannot be used to define the set of modifier genes affecting condition-specific essentially, which was one of the main goals of the DepMap project, and is less accurate overall compared to the model presented here. Since its publication, multiple projects utilized the data and insights provided from the DepMap project [ 2 , 4 , 8 – 11 ]. These efforts, using open access genomic and epigenomic sequencing data, have been instrumental in characterizing gene essentiality, the contexts in which genes are essential, and the interaction with the expression of other genes.…”

Section: Introductionmentioning

confidence: 99%

Predicting gene knockout effects from expression data

2023

View full text Add to dashboard Cite

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.

show abstract

“…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).…”

Section: Introductionmentioning

confidence: 99%

“…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). However, each method uses different information sources and most are geared toward a single prediction task.…”

Section: Introductionmentioning

confidence: 99%

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

Gervits

Sharan

2022

Front. Bioinform.

Self Cite

View full text Add to dashboard Cite

Large scale cancer genomics data provide crucial information about the disease and reveal points of intervention. However, systematic data have been collected in specific cell lines and their collection is laborious and costly. Hence, there is a need to develop computational models that can predict such data for any genomic context of interest. Here we develop novel models that build on variational graph auto-encoders and can integrate diverse types of data to provide high quality predictions of genetic interactions, cell line dependencies and drug sensitivities, outperforming previous methods. Our models, data and implementation are available at: https://github.com/aijag/drugGraphNet.

show abstract

Prediction of cancer dependencies from expression data using deep learning

Abstract: Novel deep learning methods for predicting gene dependencies and drug sensitivities from gene expression measurements.

Cited by 6 publications

References 26 publications

Adversary-Aware Multimodal Neural Networks for Cancer Susceptibility Prediction From Multiomics Data

Adversary-Aware Multimodal Neural Networks for Cancer Susceptibility Prediction From Multiomics Data

Predicting gene knockout effects from expression data

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

Contact Info

Product

Resources

About