Predicting tumor cell line response to drug pairs with deep learning

Xia, Fangfang; Shukla, Maulik; Brettin, Thomas; Garcia-Cardona, Cristina; Cohn, Judith D.; Allen, Jonathan; Maslov, Sergei; Holbeck, Susan L.; Doroshow, James H.; Evrard, Yvonne A.; Stahlberg, Eric; Stevens, Rick

doi:10.1186/s12859-018-2509-3

Cited by 111 publications

(112 citation statements)

References 26 publications

Supporting

Mentioning

109

Contrasting

Order By: Relevance

“…By analyzing feature weight parameters for the trained model, we found that predictions largely relied on specific compound substructural presence. This is in accordance with the previous research where Xia et al [14] reported that the compound descriptors provided the largest contribution to predictive ability of deeply-layered neural networks (deep learning). Figure 2 shows compounds containing such "high-importance" substructures, where substructures are highlighted in red.…”

Section: Comparison Of Descriptorssupporting

confidence: 93%

Prediction of Compound Cytotoxicity Based on Compound Structures and Cell Line Molecular Characteristics

Nakano

Brown

2020

J. Comput. Aided Chem.

View full text Add to dashboard Cite

In parallel to developments in Next-Generation Sequencing for cancer patient therapy decision making, personalized approaches to chemotherapy selection are also becoming desired. In an ideal situation, an individual's genomic, transcriptomic, and tumor-specific in-vitro response to chemical perturbation would be combined, and the US National Cancer Institute NCI-60 project has systematically screened a large chemical library against a variety of cell lines from various tumor types. Therefore, chemoinformatics approaches to make effective use of this data and identify the chemical and biological factors are of value. In this work, we investigate the impact of both chemical and biological descriptions of tumor response to chemical inhibition, and assess how well modeling approaches can predict tumor inhibition response on external datasets. We find that external datasets in both the classification and regression problems are reasonably well addressed, with the impact of chemical description outweighing the contribution from transcriptome or genome descriptions of tumors.

show abstract

Section: Comparison Of Descriptorssupporting

confidence: 93%

Prediction of Compound Cytotoxicity Based on Compound Structures and Cell Line Molecular Characteristics

Nakano

Brown

2020

J. Comput. Aided Chem.

View full text Add to dashboard Cite

show abstract

“…For example, Keine and colleagues [19] investigated the effect of polypharmacy on older adults with dementia or Alzheimer's disease and predicted the medication burden using ML. Other investigators [20][21][22] raised computational challenges with predicting the effects of drug combination (both synergism and antagonism) and suggested ML as an effective method for tackling the challenge when data are available. A recent review article [23] that assessed the applications of ML/AI in drug combinations revealed that most studies focus on predictive analytics (e.g., predicting the outcomes of drug combinations) as opposed to prescriptive analytics (e.g., optimizing the combinations of drugs).…”

Section: Related Workmentioning

confidence: 99%

Leveraging machine learning and big data for optimizing medication prescriptions in complex diseases: a case study in diabetes management

et al. 2020

View full text Add to dashboard Cite

This paper proposes a novel algorithm for optimizing decision variables with respect to an outcome variable of interest in complex problems, such as those arising from big data. The proposed algorithm builds on the notion of Markov blankets in Bayesian networks to alleviate the computational challenges associated with optimization tasks in complex datasets. Through a case study, we apply the algorithm to optimize medication prescriptions for diabetic patients, who have different characteristics, suffer from multiple comorbidities, and take multiple medications concurrently. In particular, we demonstrate how the optimal combination of diabetic medications can be found by examining the comparative effectiveness of the medications among similar patients. The case study is based on 5 years of data for 19,223 diabetic patients. Our results indicate that certain patient characteristics (e.g., clinical and demographic features) influence optimal treatment decisions. Among patients examined, monotherapy with metformin was the most common optimal medication decision. The results are consistent with the relevant clinical guidelines and reports in the medical literature. The proposed algorithm obviates the need for knowledge of the whole Bayesian network model, which can be very complex in big data problems. The procedure can be applied to any complex Bayesian network with numerous features, multiple decision variables, and a target variable.

show abstract

“…In particular, Chiu et al [5] and Li et al [15] combined auto-encoder with deep neural networks to predict drug responses of genomically profiled cell lines and tumors. Xia et al [6] utilized deep neural encoders to combine genomic features together with drug profiles to predict cell lines' responses to drug pairs. Wang et al [16] applied DNN model to predict chemically induced liver toxicity endpoints from transcriptomic responses.…”

Section: Related Work Drug Response Prediction Based On Genomic Featuresmentioning

confidence: 99%

“…We describe them in details in the following sections. As mentioned in many previous studies [14,15,6], including drug descriptor features will overshadow the impact of genomic features, in this study, we focus on the impact of genomic features and only use one-hot encoding features for drugs and do not utilize any drug feature extractors. For virtual drug screening, our approach could be easily extended to include more features.…”

Section: Datasetsmentioning

confidence: 99%

“…On the other hand, many machine learning approaches have been applied in this field and have achieved remarkable improvements. For example, Nguyen et al [4] exploited random forest model in predicting in vivo drug responses whereas Chiu et al [5] and Xia et al [6] employed deep neural network to predict drug responses of cell lines, which have achieved state-of-the-art results. However, all methods mentioned so far treat the genomic features in a flat fashion, neglecting the fact that proteins within a tissue rarely act alone [7] and instead they often team up as "molecular machines" characterized by intricate physicochemical dynamic connections to perform biological functions at both cellular and systems levels.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Integrating Protein-Protein Interaction Information into Drug Response Prediction by Graph Neural Encoding

Xie

Peng

Zhou

2019

Preprint

View full text Add to dashboard Cite

Predicting tumor cells' reactions to medical treatments has enormous significance for both medical and clinical studies. Many machine learning approaches have been previously applied to predict the effectiveness of anti-cancer drugs by analyzing genomic profiles. However, most of the current methods neglect the fact that proteins rarely act alone and often team up as "molecular machines" characterized by intricate dynamic connections to perform biological functions. Other studies indicate that subnetworks of protein-protein interaction (PPI) networks could be crucial for the development of cancer and could be used as new therapeutic targets. Recent advances in graphical representation learning inspired by deep neural networks have provided a potential solution to include PPI information into drug response prediction to improve performance. We present a graph neural network (GNN) based approach, GNNDR, to incorporate PPI information into drug sensitivity prediction based on genomic profiles. By communicating information among neighborhoods on the PPI network, our model fuses genomic features into a structural PPI graph which encodes information relevant to drug responses. We compare our approach with state-of-the-art baselines on three publicly available datasets which include both genomic features and drug responses of cell lines and Patient Derived Xenografts (PDX). We investigate the impact of different types of interactions as well as the effectiveness of different graph pooling strategies. Experimental results demonstrate the superior performance of our model over the baselines. In particular, on the NIBR-PDXE dataset, our model outperforms the baselines in 21 out of 26 cancer-treatment pairs and achieves a MSE of 1.46 on the CCLE-GDSC dataset and a MSE of 0.11 on the NCI-ALMANAC dataset. In addition, our model provides substantial model interpretability in terms of gene similarity and gene importance via the graph pooling procedures. We propose GNNDR, a GNN based approach which not only possesses the large learning capacity, but provides a solution to combine PPI information with genomic features for drug response prediction. Empirical evaluation demonstrates the effectiveness of adding protein information. Our approach, as far as we know, the first GNN based model for drug response prediction, provides a promising perspective for the identification of anti-cancer therapies.

show abstract

Predicting tumor cell line response to drug pairs with deep learning

Cited by 111 publications

References 26 publications

Prediction of Compound Cytotoxicity Based on Compound Structures and Cell Line Molecular Characteristics

Prediction of Compound Cytotoxicity Based on Compound Structures and Cell Line Molecular Characteristics

Leveraging machine learning and big data for optimizing medication prescriptions in complex diseases: a case study in diabetes management

Integrating Protein-Protein Interaction Information into Drug Response Prediction by Graph Neural Encoding

Contact Info

Product

Resources

About