DeltaNeTS+

• DeltaNeTS+ is a major improvement to our previous method DeltaNet.
• DeltaNeTS+ is a method for inferring direct gene targets of drug compounds and diseases from steady-state and/or time-series transcriptional profiles.
• DeltaNeTS+ incorporates gene regulatory information (if avaialbe) during the inference.
• DeltaNeTS+ generates a perturbation score for each gene in every sample. The score magnitude reflects the confidence that the transcription process of this gene was directly affected by the external stimuli.
• The score sign indicates the nature of the perturbation: positive for gene induction, negative for gene repression.

Please refer to DeltaNeTS+ manuscript for more detailed information.

DeltaNeTS+ installation

To install DeltaNeTS+ directly from github repository, devtools R package is required.

1. Install and load devtools package in R using the follwoing commands:

## installing devtools from CRAN
install.packages("devtools")
## Loading devtools library
library(devtools)

2. Install the package called deltanetsPlus, using devtools::install_github("CABSEL/DeltaNeTSplus/deltanetsPlus"). When you are asked to install/update the dependencies, select 1 for 'All' (* deltanetsPlus dependencies: You can also manually install R packages of doParallel, doSNOW, glmnet, foreach, Matrix, parallel, and progress). Your package will be installed in R library directory.
3. Load the package, using library(deltanetsPlus), and now you're ready to use!

DeltaNeTS+ example codes using Caenorhabditis elegans expression data

deltanetsPlus package includes example data of lfc, pval, tp,and experiment, which were processed from C. elegans dataset of Baugh et al. 2005 (GSE2180), as well as the grn structure (tf-gene interactions) of C. elegans. This dataset consists of 30 time-series samples of three genetic perturbation experiments in C. elegans (10 time points in each experiment), and the gene knock-downs were mex-3 for Experiment A, pie-1 for Experiment B, and pie-1 and pal-1 for Experiment C.

• lfc: an nx30 matrix of log base2 Fold Change values of differential gene expressions. Rows are genes (n) and Columna are time-series samples from three different experiments (10 time points per each experiment).
• pval: an nx30 matrix of statistical p-values of the differential gene expressions.
• tp: a 1X30 vector of sample time points
• experiment: a 1x30 vector of group/experiment indication.
• grn: TF-gene interaction list.

1. Generate a deltanetsPlus object.

In this example, we are creating a DeltanetsPlus object using createDeltanetsPlusObj(), which will filter out unsignificant gene expressions and compute slopes of gene expression changes, given pval and tp, respectively.

d.obj = createDeltanetsPlusObj(lfc=lfc, pval=pval, tp=tp, experiment=experiment, p.thres=0.05)

One can use combin2() to combine two DeltanetsPlus objects (e.g. d.obj=cbind2(d.obj1,d.obj2)). Note that the genes should match between the two datasets.

2. Compute gene perturbation scores using deltanetPlus().

In this example, we will compute the perturbations for only a select set of genes.

gset = c("mex-3","pie-1","pal-1",sample(rownames(lfc),10))

In the above, gset includes the actual perturbation targets (mex-3, pie-1, and pal-1) as well as 10 random genes, which were not perturbed. In the manuscript describing DeltaNeTS+, we applied the method for the complete set of genes. If gset is not provided, deltanetsPlus() will compute the perturbation scores for the complete set of genes. The example below implements parallel computing (par=TRUE) with 2 nodes (numClusters=2). To turn off parallel option, users can set par=FALSE.

dts.res <- deltanetsPlus(d.obj, 
                         grn=grn, 
                         kfolds=10,
                         cv.method="cv",
                         perturbation="group",
                         gset=gset,
                         group=NULL,
                         lambda=10^seq(-2,5,length.out=100),
                         cv.opt = "lambda.1se",
                         par=TRUE, numClusters=2)

Finally, we can check the perturbation scores by DeltaNeTS+. An example of the result from the analysis above is shown below. The result expectedly shows large negative perturbation values for mex-3, pie-1, and pal-1/pie-1 in exp. A (1st Column), exp. B (2nd Column), and exp. C (3rd column), respectively, since these genes were knocked down in the experiments.

gset2 = intersect(gset, rownames(dts.res$P))
print(dts.res$P[gset2,])

Example results:

12 x 3 sparse Matrix of class "dgCMatrix"

genes	1	2	3
mex-3	-1.776621450	-0.59719729	-0.62503056
pie-1	-1.182894234	-2.05782894	-2.40294302
pal-1	.	-0.02866807	-3.01030870
ife-3	-0.002458885	-0.00233180	-0.00266895
frm-5.1	0.044570707	-0.08326815	.
dnj-17	-0.138966291	-0.30727911	-0.26977825
F29D10.2	-0.187006724	-0.17008810	-0.18143735
pgrn-1	-0.307675932	.	0.10154587
F25E5.3	-0.093054526	.	-0.12077273
Y69A2AR.23	0.058642704	.	.
ZK112.6	0.084273230	.	0.16039465
ncs-2	-0.104642724	.	.

Acknowledgements

This work was supported by the ETH Zurich Research Grant.