syssynbio/prob

pseudotemporal progression-based Bayesian method for inferring GRNs from cross-sectional clinical transcriptomic data

PROB

Latent-temporal progression-based Bayesian method for inferring GRNs from cross-sectional clinical transcriptomic data

PROB contains two main functions, ‘Progression_Inferrence’ (for inferring pseudotemporal progression) and ‘ODE_BayesianLasso’ (for inferring ODE network using Bayessian Lasso). The visualization module of PROB enables the users to plot latent-temporal progression trajectory and to plot temporal expression of the selected genes. In addition, the prior network information can be incorporated into function ‘ODE_BayesianLasso’ when PROB is applied to large scale network reconstruction.

Functions

The input and output of each function are described below.

1. Pseudotemporal progression inference

[Data_ordered,PPD, TimeSampled]=Progression_Inferrence(Data)

This function (.m) is designed to infer pseudotemporal progression from the clinical cross-sectional gene expression data.

Input

Data: a n×m matrix containing gene expression profiles (the first n-1 rows) and grade information of patients (the last row). m is sample size, i.e., the number of patients.

Output

Data_ordered: a matrix containing the ordered (and smoothed) gene expression data.

PPD: a vector containing pseudotemporal progression distance for each patient.

TimeSampled: a vector containing standardized time-points sampled for Data_ordered.

2. GRN inference

[Para_Post_pdf,S,AM]=ODE_BayesianLasso(Data_ordered, TimeSampled)

This function (.m) is designed to reconstruct causal GRN based on the results of the above pseudotemporal progression inference.

Input

Data_ordered: matrix for the ordered (and smoothed) gene expression data (i.e., a subset from the output of the first function).

TimeSampled: vector for the standardized time-points associated with Data_ordered (i.e., the output of the first function).

Output

Para_Post_pdf: cell format saving the posterior distribution over the regulatory coefficients of each gene in the GRN model.

S: a matrix saving the presence probability.

AM: Adjacent matrix of the inferred GRN. (aij) for the regulatory strength from gene j to gene i.

3. Visualization

Progression_Plot(Data)

This function (.m) is designed to plot pseudotemporal progression trajectory.

Input

Data: a n×m matrix containing single cell gene expression profiles (the first n-1 rows) and grade information of patients (the last row). m is sample size, i.e., the number of patients.

Output

Two figures of pseudotemporal progression trajectory and pseudotemporal progression score along the ordering of patients.

TemporalGene_Plot(geneID,TimeSampled,Data_ordered)

This function (.m) is designed to plot pseudotemporal expression of the selected genes.

Input

geneID: the ID of genes selected for visualization.

Data_ordered: matrix for the ordered and smoothed gene expression data (i.e., the output of the first function).

TimeSampled: vector for the standardized time-points associated with Data_ordered (i.e., the output of the first function).

Output

A figure of the expression dynamics of the selected genes over the inferred pseudotemporal progression.

Cytoscape_Reformat(AM,NodeID)

This function (.R) is designed to reformat the AM to a suitable format for input as cytoscape software.

Input

AM: The adjacent matrix of the inferred GRN resulted from the above function ‘ODE_BayesianLasso’.

NodeID: The ID or the symbol of the genes in the AM.

Output

A matrix containing 3 columns: source nodes, interaction coefficients and target nodes.

4. Incorporating prior network information

[Para_Post_pdf,S,AM]=ODE_BayesianLasso_PriorNet(Data_ordered,TimeSampled,PriorNet)

This function (.m) is designed to incorporate prior network information into function ‘ODE_BayesianLasso’, when PROB is applied to large scale network reconstruction.

Input

Data_ordered: matrix for the ordered and ordereded gene expression data (i.e., the output of the first function).

TimeSampled: vector for the standardized time-points associated with Data_ordered (i.e., the output of the first function).

PriorNet: a matrix for the prior network information. (Pij) represents the regulatory strength from gene j to gene i.

Output

Para_Post_pdf: cell format saving the posterior distribution over the regulatory coefficients of each gene in the GRN model.

S: a matrix saving the presence probability.

AM: Adjacent matrix of the inferred GRN. (aij) for the regulatory strength from gene j to gene i.

Example demonstration

The users need to first download all the files in this repository and save them into the working directory.

Load data

load ExampleData.mat  % Firstly load example data;
Data = ExampleData; % The ExampleData is a dataset containing expression levels of 6 genes (the first 6 rows) and  clinical information (the last row).

Pseudotemporal progression inference

[Data_ordered,PPD,TimeSampled]=Progression_Inference(Data);

Bayesian Lasso for GRN parameter estimation

[Para_Post_pdf,S,AM]=ODE_BayesianLasso(Data_ordered,TimeSampled);
csvwrite('AdjacentMatrix.csv', AM); % save Adjacent matrix for visualization (can be reformed for cytoscape visualization)

Visualization

Progression_Plot(X_stage);  % plot pseudotemporal progression trajectory
geneID=[1:6]; % specify the ID of genes you want to plot
TemporalGene_Plot(geneID,TimeSampled,Data_ordered); % plot pseudotemporal gene expression

Other information

Please cite oue paper if you used codes here.

Xiaoqiang Sun, Ji Zhang, Qing Nie. Inferring latent temporal progression and regulatory networks from cross-sectional transcriptomic data of cancer samples. (link) (preprint has posted on bioRxiv: https://biorxiv.org/cgi/content/short/2020.10.07.329417v1)

Any questions please contact [email protected]