knnDemix

R package that provides a non-parametric hypothesis test for the mixture coefficient of a mixture with an unknown component.

Let $X_0\sim F_0$ and $X_m\sim \alpha X_0 + (1-\alpha) X_1$.
Then the sum of log p-values obtained by testing for $\hat{f}_0 \geq \alpha \hat{f}_m$ where $\hat{f}$ are the k nearest neighbour density estimates at the points in $X_0$ follows a $\chi^2$ distribution

This package provides a implementation of a corresponding hypothesis test for $\alpha$. Works. Needs proof. For example this might be applied to determine the knockout rate from fluorescence cytometry measurement of a biological sample given additionally only measurements from a negative control sample.

Installation

# install.packages("remotes")
remotes::install_github("jan-glx/knnDemix")