ciphelab/tviblindi

Topological and Geometrical Tools for Single-cell Data

tviblindi

Topological and Geometrical Tools for Single-cell Data

This package is under development and depends on several libraries - issues during installation are expected.

We recommend to pull the docker container provided with all dependencies and Rstudio server

parts of code from following libraries are used:
RcppAnnoy
Gudhi
Phat

Single-cell trajectory inference toolkit

---

tviblindi puts concepts from graph theory and algebraic topology to use for trajectory inference (TI) in high-dimensional biological data (cytometry, scRNA-seq, CITE-seq).

We provide easy-to-use tools for identifying potential developmental trajectories and grouping them in a classification tree. This includes a graphical user interface that enables the user to

• inspect trajectories in different classes by topological relationships,

• view the trajectories drawn on a 2-dimensional layout of input data,

• track progression of expression levels for markers of interest at different stages of development,

• check the composition of tracked populations in terms of labelled cell populations and

• export enhanced FCS files for viewing results of the TI analysis in FlowJo or other gating software.

Docker container

Apple silicon (M1) is not supported for the time being

port=7777
data_path="path to data folder to mount"
rpassword="password for rstudio server (user=rstudio)"
docker run -it -d -p $port:8787 --name tviblindi_container -v $data_path:/data -e PASSWORD=$rpassword stuchly/tviblindi

then in your webrowser visit localhost:7777 (user: rstudio, password: rpassword; localhost may depends on your docker daemon setting)

Installation

devtools::install_github("stuchly/tviblindi")

Current version of tviblindi depends on CGAL library version 4.14

Macos:

brew tap-new CGAL/legacy

brew extract --version=4.14 CGAL CGAL/legacy

brew install CGAL/legacy/[email protected]

Usage

library(tviblindi)
sn<-make_snowman3d(2000) #requires package TDA


colnames(sn)<-1:3
lab<-rep(1,nrow(sn))
lab[which.max(sn[,2])]<-0
lab<-as.factor(lab)


tv1<-tviblindi(data=sn,labels=lab)
tv1$origin$default<-which.max(sn[,2]) #fix the origin on the top of snowman's head

KNN(tv1,50,method="balltree") #balltree more faster for small data
#Denoise(tv1) #reduce noise before witness complex construction; for real world data 
Som(tv1, xdim=15, ydim=15) #kmeans clustering by default - 15*15 clusters
Filtration(tv1) #default setting is too conservative, less simplices could be created with same resolution (e.g. Filtration(tv1,alpha2=0.1))
DimRed(tv1)
Pseudotime(tv1,sym = "min") #sym="min" - experimental asymmetric model, looks better in this example
Walks(tv1,N=1000)

launch_shiny(tv1) #note the "Min trajectory count % per leaf" slider to see all classes & "Point size" selector

To inspect the connectedness of different populations in your dataset based on Louvain clustering, use the fully automated 'Connectome' functionality - creates a png file.

Connectome(tv1)

Reference

The article is in process

The shiny app was developed as a part of master's thesis of co-author David Novak: Studying lymphocyte development using mass cytometry