vitkl/ParetoTI: R toolbox for Archetypal Analysis and Pareto Task Inference on single cell data

R toolbox for Archetypal Analysis and Pareto Task Inference on single cell data, partially based on ParTI described in Yuval Hart & Uri Alon paper in Nature Methods (2015): Inferring biological tasks using Pareto analysis of high-dimensional data. This package performs archetypal analysis using a python 2.7 implementation of PCHA algorithm. It was originally implemented in Matlab by Morten Mørup (http://www.mortenmorup.dk/MMhomepageUpdated_files/Page327.htm) and re-implemented in python by Ulf Aslak. Statistical significance of how well polytope defined by archetypes describes the data is determined using permutations of the dataset that disrupt relationships between variables but keep the distribution of each variable constant. \ We also implement a bootstraping-based procedure to identify the correct number of archetypes. In addition we have 2 methods for identifying genes and GO functions of each archetype. For convenience, we provide logistic regression model for classifying cells (or other objects) with logistic regression, and GeoSketch method for reducing the size of the data while preseving rare populations. Package is currently under development! Alternative methods for archetypal analysis will be considered such as AANet (https://github.com/KrishnaswamyLab/AAnet/) and SPAMS implemented in c++ via R interface (https://arxiv.org/pdf/1405.6472.pdf, http://spams-devel.gforge.inria.fr/downloads.html)

install.packages("remotes")
remotes::install_github("vitkl/ParetoTI")