A reference profile-free deconvolution method to infer cancer cell-intrinsic subtypes and tumor-type-specific stromal profiles

摘要

Flow chart of the DeClust algorithm ( ) and the simulation results ( – ). The accuracy of the estimated cell compartment fraction in was calculated via the correlation between simulated and estimated cell frequency profiles (average for the three components). The accuracy of the estimated expression profiles in was calculated by correlation between simulated and estimated expression profiles. Specifically, the average correlations over the cancer, immune, and stromal profiles are plotted at different noise levels and sample sizes. The noise levels represent the standard deviation of the noise added to the simulated mixed expression data under the log-normal distribution (see the “ ” section)