Patterns: Predicting drug response through tumor deconvolution by cancer cell lines (Chen Lab)

Yu-Ching Hsu 1 2 3 4, Yu-Chiao Chiu 5 6, Tzu-Pin Lu 3, Tzu-Hung Hsiao 7, Yidong Chen 4 8 9


•Scaden-CA is a deep-learning model for tumor deconvolution
•Tumor deconvolution facilitates the use of a drug response prediction algorithm
•The model explores drug response mechanisms via DNA mutations and/or expression changes

The bigger picture

Drug repurposing involves utilizing approved or investigational drugs beyond the scope of the original medical application. This approach offers several advantages compared with developing entirely new drugs, including a lower risk of failure, a shorter development timeline, and lower investment costs. Using previous knowledge of the drug sensitivity of cancer cell lines, we created a tumor convolution model to predict how tumors will respond to anti-cancer drugs. In this approach, a deep learning model used single-cell gene expression profiles to deconvolute tumors into their constituent cancer cell lines; in other words, the model represented tumors as a mixture of different cancer-type-specific cell lines in varying proportions. Subsequently, the deconvoluted proportions facilitated the prediction of tumors’ drug responses. Ultimately, these observations highlight the potential for deep learning applications in the area of drug repurposing.


Large-scale cancer drug sensitivity data have become available for a collection of cancer cell lines, but only limited drug response data from patients are available. Bridging the gap in pharmacogenomics knowledge between in vitro and in vivo datasets remains challenging. In this study, we trained a deep learning model, Scaden-CA, for deconvoluting tumor data into proportions of cancer-type-specific cell lines. Then, we developed a drug response prediction method using the deconvoluted proportions and the drug sensitivity data from cell lines. The Scaden-CA model showed excellent performance in terms of concordance correlation coefficients (>0.9 for model testing) and the correctly deconvoluted rate (>70% across most cancers) for model validation using Cancer Cell Line Encyclopedia (CCLE) bulk RNA data. We applied the model to tumors in The Cancer Genome Atlas (TCGA) dataset and examined associations between predicted cell viability and mutation status or gene expression levels to understand underlying mechanisms of potential value for drug repurposing.

Read Full Text

Article Categories: Research Paper

Since 2004, UT Health San Antonio, Greehey Children’s Cancer Research Institute’s (Greehey CCRI) mission has been to advance scientific knowledge relevant to childhood cancer, contribute to understanding its causes, and accelerate the translation of knowledge into novel therapies. Greehey CCRI strives to have a national and global impact on childhood cancer by discovering, developing, and disseminating new scientific knowledge. Our mission consists of three key areas — research, clinical, and education.

Stay connected with the Greehey CCRI on Facebook, Twitter, LinkedIn, and Instagram.