Abstract 2094: A user-friendly R Shiny app for predicting drug response of cancer using deep learning

Abstract Realizing the full potential of precision oncology requires accurate prediction of treatment response. We previously developed a deep learning model, namely DeepDR, to predict drug sensitivity by integrating baseline profiles of mutations and gene expression of a cancer sample (Chiu et al.; DOI: 10.1186/s12920-018-0460-9). The model features a transfer learning design that captures two types of features: i) tumor relevant representations of mutation and expression data learned from tumor samples, and ii) predictive model for drug response (measured by IC50 values) learned from high-throughput drug screens of pan-cancer cell lines. DeepDR was applicable to both cell lines and tumors and achieved superior performance over conventional methods. To make the DeepDR model more accessible to biomedical researchers with limited programming skills, here we present a user-friendly platform implemented by using an R Shiny framework. The R Shiny app is a web interface that allows users to upload mutation and/or gene expression profiles of a cancer sample (cell or tumor model) and predict the sample’s response to 265 FDA-approved and investigational compounds covered by the screening library of the Genomics of Drug Sensitivity in Cancer (GDSC) project. The app provides an intuitive user interface to interactively visualize, search, and filter prediction results among 265 compounds. It also enables downstream analyses including statistical tests and provides links to external compound databases, such as PubChem. In the original publication of DeepDR, we have validated its performance across cell lines and tumors using well-known pharmacogenomic associations, such as estrogen receptor antagonist (tamoxifen) and EGFR-targeting drugs (afatinib and gefitinib), as well as a novel agent, CX-5461, in treating hematopoietic malignancies. Here we utilized the proposed Shiny app to further predict drug sensitivity of patients with hepatocellular carcinoma (HCC; n=356) of The Cancer Genome Atlas (TCGA). We investigated CTNNB1 as a demonstrating example for it is one of the most prevalent, yet undruggable, trunk mutations of HCC. A systematic search yielded nine compounds that were predicted to be significantly more effective in HCC tumors harboring CTNNB1 mutations (n=92) compared to others (n=264) (with one-tailed t-test P < 0.0001). These top compounds target critical cancer pathways, such as cell apoptosis (p53 activator and Bcl-2 inhibitor), the PI3K/AKT and MAPK signaling pathways, and DNA-dependent protein kinases, suggesting candidates for further in vitro and in vivo investigations. In summary, the present R Shiny app is a user-friendly platform that enables in silico drug screening using deep learning with no requirement of programming experience. We expect the tool to foster accessibility of our deep learning prediction machine and facilitate the process of drug development in cancer. Citation Format: Li-Ju Wang, Michael Ning, Tapsya Nayak, Satdarshan P. Monga, Yufei Huang, Yidong Chen, Yu-Chiao Chiu. A user-friendly R Shiny app for predicting drug response of cancer using deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 2094.