Patient Anticancer Drug Response Prediction Based on Single-Cell Deconvolution

Predicting patient responses to anticancer drugs is essential for the development of effective treatment strategies. Tumors, intricate structures comprised of diverse cell types, exhibit considerable cellular heterogeneity. Leveraging single-cell data offers a promising avenue for deciphering this complexity and enhancing the accuracy of drug response prediction. In this study, we propose the Single-Cell Deconvolution Guided method for Patient Anticancer Drug Response Prediction (ScPDRP). ScPDRP utilizes single-cell gene expression data to deconvolve both cell line and patient gene expression profiles. Through the employment of several encoders and generative adversarial training, ScPDRP extracts domain-invariant features from cell line and patient data, facilitating downstream drug response prediction tasks. Evaluation of our model on a curated selection of drug datasets from the clinical TCGA dataset demonstrates its superior performance over existing state-of-the-art methods across nearly all drug datasets.