ACSNI

Automatic context-specific network inference

Determining tissue- and disease-specific circuit of biological pathways remains a fundamental goal of molecular biology. Many components of these biological pathways still remain unknown, hindering the full and accurate characterisation of biological processes of interest. ACSNI leverages artificial intelligence for the reconstruction of a biological pathway, aids the discovery of pathway components and classification of the crosstalk between pathways in specific tissues.

This tool is built in python3.8 with tensorflow backend and keras functional API.

Installation and running the tool

The best way to get ACSNI along with all the dependencies is to install the release from python package installer (pip)

pip install ACSNI This will add four command line scripts:

Script	Context	Usage
ACSNI-run	Gene set analysis	ACSNI-run -h
ACSNI-derive	Single gene analysis	ACSNI-derive -h
ACSNI-get	Link pathway trait	ACSNI-get -h
ACSNI-split	Split expression data	ACSNI-split -h

Utility functions can be imported using conventional python system like from ACSNI.dbs import ACSNIResults

Input ACSNI-run

Expression Matrix - The expression file (.csv), specified by -i, where columns are samples and rows are genes. The expression values should be normalised (eg. TPM, CPM, RSEM). Make sure the column name of the 1st column is "gene".

gene	Sample1	Sample2	Sample3
Foxp1	123.2	274.1	852.6
PD1	324.2	494.1	452.6
CD8	523.6	624.1	252.6

This input should not be transformed in any way (e.g. log, z-scale)

Gene set matrix - The prior matrix (.csv) file, specified by -t, where rows are genes and column is a binary pathway membership. Where "1" means that a gene is in the pathway and "0" means that the gene is not know a priori. The standard prior looks like below. Make sure the column name of the 1st column is "gene".

gene	Pathway
Foxp1	0
PD1	0
CD8	1

You can also supply gene IDs instead of gene symbols.

The tool can handle multiple pathway columns in the -t file as below.

gene	Pathway1	Pathway2	Pathway3
Foxp1	0	0	0
PD1	0	1	0
CD8	1	0	1

Note: Each pathway above is analysed independently, and the outputs have no in-built relationship. The tool is designed to get a granular view of a single pathway at a time.

Output ACSNI-run

Database (.ptl)

Content	Information
co	Pathway Code
w	Subprocess space
n	Interaction scores
p	Score classification
d	Interaction direction
run_info	Run parameters
methods	Extractor functions

Predicted Network (.csv)

Content	Meaning
name	Gene
sub	Subprocess
direction	Direction of interactions with subprocess

Null (.csv) {Shuffled expression matrix}

Input ACSNI-derive

Expression Matrix - See ``-i``` description above.

Note - We recommend removing any un-desirable genes (eg. MT, RPL) from the expression matrix prior to running ACSNI-derive as they usually interfere during initial prior matrix generation steps. For TCR/BCR genes, counts of alpha, beta and gamma chains can be combined into a single count.

Biotype file (Optional) - The biotype file (.csv) specified by -f, given if the generation of gene set should be based on a particular biotype specified by -b.

gene	biotype
Foxp1	protein_coding
PD1	protein_coding
MALAT1	lncRNA
SNHG12	lncRNA
RNU1-114P	snRNA

Correlation file (Optional) - The correlation file (.csv) specified by -u, given if the user wishes to replace "some" specific genes with other genes to be used as a prior for the first iteration of ACSNI-run (internally).

gene	cor
Foxp1	0.9
PD1	0.89
MALAT1	0.85
SNHG12	0.80
RNU1-114P	0.72

#Example run using the automatic estimation of optimal dimensions (p=0)

ACSNI-derive -i <expression.csv> -g <"gene"> -m 1.2 --ct 0.60 --pc 5 -p 0

#Example run using fixed dimensions (16%)

ACSNI-derive -i <expression.csv> -g <"gene"> -m 1.2 --ct 0.60 --pc 5

Output ACSNI-derive

Database (.ptl)

Content	Information
co	Pathway Code
n	Interaction scores
d	Interaction direction
ac	Correlation and T test results
fd	Unfiltered prediction data
run_info	Run parameters
methods	Extractor functions

Predicted (.csv)

Content	Meaning
name	Gene
predict	Classification of genes

Null (.csv) {Shuffled expression matrix}

Input ACSNI-get

ACSNI database - Output of ACSNI-run (.ptl) specified by -r.

Target phenotype - Biological phenotype file (.csv) to link ACSNI subprocesses, specified by -v. The sample IDs should match the IDs in the -i analysed by ACSNI-run.

Variable type - The type of phenotype i.e "numeric" or "character", specified by -c.

Outputs the strength of the associations across the subprocesses (.csv).

Input ACSNI-split

Expression Matrix - See ``-i``` description above.

Number of splits - The number of independent cohorts to generate from `-i```.

Outputs the data splits in the current working directory.

Extras

Example files representing the datasets analysed in the paper are included inside the folder "Resources".

R functions to reproduce the downstream analyses reported in the paper are inside the folder "R".

Example runs are inside the folder "sh".

To clone the source repository

git clone https://github.com/caanene1/ACSNI

Citation

ACSNI: An unsupervised machine-learning tool for prediction of tissue-specific pathway components using gene expression profiles Chinedu Anthony Anene, Faraz Khan, Findlay Bewicke-Copley, Eleni Maniati and Jun Wang