Fig.1 - The amount of chocolate consumption is correlated with the prospect of winning a Nobel Prize

Causal Reasoning Analysis

A collection of tools, which, when given a readout of a perturbation experiment, runs tools such as CARNIVAL/causalR to infer the causal factors that leads to the observed outcome

Background

See slides for some links to background concepts such as Prize Collecting Steiner Trees and some papers that have formed the foundation of tools such as CARNIVAL/causalR.

Dependencies

CARNIVAL

R libraries

library(CARNIVAL)
library(progeny)
library(dorothea)
library(OmnipathR)
library(data.table)
library(igraph) 

CBC solver (for integer linear programming) optimisation

conda install -c conda-forge coincbc

CausalR

library(OmnipathR)
library(CausalR)
library(data.table)

To Run Carnival

For more info on CARNIVAL, see https://www.bioconductor.org/packages/release/bioc/html/CARNIVAL.html

# import CarnivalClass
source('./src/carnival.R')

# instantiate the class; imports perturbation data
# calculates pathway and TF activity scores; processes omnipath data
carnival <- CarnivalClass$new(scores_file_path = 'data/lincs_troglitazone_PC3.csv',
                              organism = 'Human',
                              datasets = 'dorothea',
                              cbc_solver_path = '~/.conda/envs/causal_reasoning/bin/cbc')
# run carnival
carnival$runCarnival(threads = 32, absTFactivityThreshold = 2)

# detect sub modules in the resulting subnetworks
comms <- carnival$detect_communities(carnival$carnival_result$sifAll[[1]])

# plot subgraphs (plot the subgraph corresponding to the largest community)
carnival$plot_subgraph(carnival$get_subgraph(carnival$network_sif, #carnival$carnival_result$sifAll[[1]],
                                             comms[[1]]),
                       max_size = 50, margin = -0.5)


# access the results
carnival$carnival_result

# access other scores (pathway scores and TF activity scores)
carnival$progeny_scores
carnival$tf_activity_scores

To Run CausalR

For more info, see: https://www.bioconductor.org/packages/release/bioc/html/CausalR.html

# import CausalrClass
source('./src/causalR.R')

# instantiate the class
# it is important to define what is a good threshold for up/down regulation 
# based on the scores found in scores_file_path 
# datasets: which network to import (use dorothea network for testing, 
# causalR can take long to run on the whole omnipath network)
caus <- CausalrClass$new(scores_file_path = 'data/lincs_troglitazone_PC3.csv', 
                         organism = 'Human', 
                         upregulationThreshold = 1,
                         downregulationThreshold = -1,
                         datasets = 'dorothea') 

# Run "Sequential Causal Analysis of Networks" 
caus$run_scanr(numberOfDeltaToScan = 5, topNumGenes = 100)

# Access the results
caus$result

# Top genes who rank among the top genes most consistently:
caus$result$commonNodes