Table 2.

Representative drug sensitivity prediction models classified in terms of whether or not they implement also feature selection

	Prediction model	Example applications
Kernel-based	SVM1	Dong et al. (2015) used SVM classification model to predict drug sensitivity accurately for several drugs using baseline gene expression of cell line panels from preclinical studies (CCLE2 and CGP3) as features. Other applications of SVM for drug response prediction include, e.g., Costello et al. (2014), Jang et al. (2014), and Hejase and Chan (2015).
	BEMKL	Kernelized regression model for drug response prediction based on data integration across multiple omics profiles, through multi-task, multiple kernel learning (Costello et al. 2014; applications in breast cancer cell line panel). A particular emphasis was placed on the proteomic profiles in our follow-up work using NCI4-60 human tumor cell lines screen (Ali et al. 2018).
	cwKBMF5	Drug response prediction model (Ammad-ud-din et al. 2016) by utilizing cell line information along with the drug chemical properties as an additional information source through selective data integration. Applications in GDSC6 and CTRP7 cancer cell line panels, and wet-lab validations in AML cell lines conducted in-house.
	KRL	Kernelized rank learning (KRL; He et al. 2018) is a personalized drug recommendation method that selects the most promising drug based on its predicted effect per cell line. Applications shown in GDSC cell lines and TCGA breast cancer patients using one expression profile at a time.
Feature selection-based	Ridge Regression	Geeleher et al. (2014) and Geeleher et al. (2017) applied ridge regression model to predict drug responses in GDSC cell lines, and inferred marker panels for predicting comprehensive drug response profiles in patient tumors in the TCGA dataset (Geeleher et al. 2017).
	Elastic net	Jang et al. (2014) found elastic net regression as one of the best-performing modeling strategies for drug response prediction in CCLE and GDSC cancer cell lines. Similarly, Ding et al. (2018) applied elastic net regression to generate logistic models for drug sensitivity prediction through deep learning in CCLE and GDSC datasets.
	Random forests	Riddick et al. (2010) built an ensemble regression model using random forest (RF) for drug sensitivity prediction in NCI-60 cell line panel. The model was also used to create drug-specific gene expression signatures and identify core cell lines associated with each drug’s response. Other applications of RF include, e.g., Menden et al. (2013), Nguyen et al. (2016), and Rahman et al. (2017).
	MVLR8	Bayesian multi-view multi-task linear regression model (Ammad-ud-din et al. 2017) for drug response prediction by capitalizing on feature combinations that are most predictive of the drug’s response. This method also enables one to use functional-linked-networks (FLNs) as prior biological knowledge. Applications in GDSC and in-house TNBC9 cell line panels.

¹SVM, support vector machines

²CCLE, Cancer Cell Line Encyclopedia

³CGP, Cancer Genome Project

⁴NCI, National Cancer Institute

⁵cwKBMF, component-wise kernelized Bayesian matrix factorization

⁶GDSC, Genomics of Drug Sensitivity in Cancer project

⁷CTRP, Cancer Therapeutic Response Portal

⁸MVLR, multi-view linear regression

⁹TNBC, triple-negative breast cancer