Modeling cancer micro-environment using machine learning techniques

    We developed a novel computational pipeline namely ICAD (Inference of Cell types And Deconvolution) for an accurate inference of immune/stromal (I/S) cell types in a cancer tissue. We have validated our method by using bulk tissue RNA-seq data simulated by four sets of single cell RNA-seq data, and demonstrated ICAD can accurately identify of the I/S cell and sub-cell types, and predict their infiltration level. With applying the method to TCGA data sets, and integrated with independent single cell RNA-seq data, we have identified genes, miRNA and lncRNA expressed by cancer cells that may affect I/S cells’ infiltration and activity level, which have potential to be used as targets for the non-responding mechanisms of immunotherapy