Reconstruction of transcriptional regulatory networks (TRNs) is a powerful approach to unravel the gene expression programs involved in healthy and disease states of a cell. However, these networks are usually reconstructed independent of the phenotypic properties of the samples and therefore cannot identify regulatory mechanisms that are related to a phenotypic outcome of interest. InPheRNo (Inference of Phenotype-relevant Regulatory Networks) is a computational tool to reconstruct phenotype-relevant transcriptional regulatory networks (TRNs) using transcriptomic data. Here we present InPheRNo, a novel computational tool to reconstruct ‘phenotype-relevant’ transcriptional regulatory networks. This method is based on a probabilistic graphical model (PGM) whose conditional probability distributions model the simultaneous effects of multiple transcription factors (TFs) on their target genes as well as the statistical relationship between target gene expression and phenotype.
The figure below depcits the method overview.
In order to run the code, you need to have Python 3.5 installed. In addition, the code uses the following python modules/libraries which need to be installed (the number in brackets show the version of the module used to generate the results in the manuscript):
- Numpy (version 1.13.0)
- Scipy (verison 0.19.1)
- Pandas (version 0.20.2)
- Sklearn (scikit-learn) (version 0.18.1)
- PyMC (version 2.3.6)
- StatsModels (version 0.8.0)
Instead of installing all these libraries independently, you can use prebulit Python distributions such as Anaconda, which provides a free academic subscription. If you are using Anaconda, you can easily install any specific version of the modules above using a command like:
conda install pymc=2.3.6
Running InPheRNo involves running three manuscripts (InPheRNo_step1.py, InPheRNo_step2.py and InPheRNo_step3.py) in a row. Since the intermediate results are used in the following steps, one needs to wait for the preceeding step to finish before running the next step.
This is a csv file in which rows contain the names of the regulators (e.g. TFs). The file should not have a header. As an example see the file "Data/TF_Ensemble.csv".
This is a (gene x phenotype) csv file (see "Data/Pvalue_gene_phenotype_interest.csv" as an example). The rows correspond to target genes of interest (this may be only a subset of all genes, or it may be all the genes). The p-value for TF-phenotype should not be included in this file. The value assigned to each gene represents the p-value of association between the expression of that gene and the variation in the phenotype across different samples obtained using a proper statistical test (e.g. a ttest for binary phenotype or Pearson's correlation for continuous, etc.). The genes should be sorted in an ascending order based on the p-value (smallest p-values appear first). The file is assumed to have a header.
Example:
| Pvalue | |
|---|---|
| gene1 | 1E-22 |
| gene2 | 5E-14 |
| gene3 | 3E-10 |
This is a (gene x samples) csv file containing the normalized gene (and TF) expression profiles across different samples. This file must contain expression of target genes provided in Input1.2 and TFs provided in Input1.1. The file has a header representing sample names. See "Data/expr_sample.csv" as a sample input. For best results, normalize the matrix such that the gene expresison values of each gene (approximately) follows a standard Normal distribution (mean=0, var=1), across all samples.
Example:
| sample1 | sample2 | sample3 | |
|---|---|---|---|
| TF1 | 0.1 | 0.9 | 0.5 |
| TF2 | -0.3 | 0.5 | -0.6 |
| gene1 | -1.1 | 0.6 | 1.4 |
| gene2 | 0.9 | -2.3 | -0.3 |
| gene3 | 0.4 | 0.8 | 1.5 |
The first step of InPheRNo generates two output files that by default will be located in a directory called "Results" placed in the current directory. These intermediate outupts will be used in the next step of InPheRNo.
If default parameters are used to run the first step, Output1.1 will be a file called "Pvalue_gene_phenotype_interest_tmp.csv" which is generated from Input1.2, properly sorted and cleaned up (if necessary). See folder "Results" for a sample.
If default parameters are used to run the first step, Output1.2 will be a file called "Pvalue_gene_tf_tmp.csv". This is a (gene x TF) csv file containing p-values of gene-tf association, sorted in an ascending order based on Output1.1 file. The file has a header. See folder "Results" for a sample.
To Run this step with default parameters, place all the three input files above in one folder. Then specify the following four arguments:
- input_directory: address of the data directory containing the three input files (e.g. "./Data")
- input_tf: name of Input1.1 file containing the name of regulators (e.g. "TF_Ensemble.csv")
- input_gene_phenotype_interest: name of Input1.2 containing p-value of gene-phenotype (e.g. "Pvalue_gene_phenotype_interest.csv")
- input_expression: name of Input1.3 containing the expression of genes and TFs (e.g. "expr_sample.csv")
The following line shows how to run InPheRNo using the sample files:
python3 InPheRNo_step1.py --input_directory ./Data --input_tf TF_Ensemble.csv --input_gene_phenotype_interest Pvalue_gene_phenotype_interest.csv --input_expression expr_sample.csv
By default, InPheRNo writes the intermediate outputs generated in this step into a directory called "Results" in the current directory. To change the location of the intermediate results, see advanced settings.
In addition to the arguments above, one can use the following optional arguments to change the default settings.
- -od, --output_directory (string, default='./Results'): Address of the output directory
- -mt, --max_num_tf (integer, default = 15): Maximum number of TFs recovered for each gene using Elastic Net
- -lr, --l1_ratio, (float, default = 0.5): l1 ratio of the Elastic Net model
- -ogp, --output_gene_phenotype (string, default = 'Pvalue_gene_phenotype_interest_tmp.csv'): Name of Output1.1 file
- -tgt, --output_gene_tf (string, default = 'Pvalue_gene_tf_tmp.csv'): Name of Output1.2 file
This step requires three input files. Two of these input files are the intermediate outputs generated in STEP1.
This is a (gene x phenotype) csv file (see "Data/Pvalue_gene_phenotype_all.csv" as an example) and is very similar to Input 1.2. The main difference is that this file needs to contain the gene-phenotype association p-values for all the genes and not just the genes of interest. This file is used to estimate the parameters of the PGM. The rows correspond to target genes of interest (this may be only a subset of all genes, or it may be all the genes). The p-value for TF-phenotype should not be included in this file. The value assigned to each gene represents the p-value of association between the expression of that gene and the variation in the phenotype across different samples obtained using a proper statistical test (e.g. a ttest for binary phenotype or Pearson's correlation for continuous, etc.). The genes should be sorted in an ascending order based on the p-value (smallest p-values appear first). The file is assumed to have a header.
This is the Output1.1 file generated in STEP1. If default parameters are used to run the first step, this will be a file called "Pvalue_gene_phenotype_interest_tmp.csv" which is generated from Input1.2, properly sorted and cleaned up (if necessary). See folder "Results" for a sample.
This is the Output1.2 file generated in STEP1. If default parameters are used to run the first step, Output1.2 will be a file called "Pvalue_gene_tf_tmp.csv". This is a (gene x TF) csv file containing p-values of gene-tf association, sorted in an ascending order based on Output1.1 file. The file has a header. See folder "Results" for a sample.
The second step of InPheRNo generates many output files. Each output file corresponds to one run of the PGM with random initialization. These results are aggregated in the third step to generate a final network.
If you have already used the default parameters in STEP1, you can easily run the second step by specifying the following arguments:
- input_dir: Address of directory containing gene-pheno association for ALL genes
- input_gene_pheno_all: Name of file containing gene-phenotype association p-values for all genes (Input2.1)
The following line shows how to run InPheRNo using the sample files:
python3 InPheRNo_step2.py --input_dir ./Data --input_gene_pheno_all Pvalue_gene_phenotype_all.csv
By default, InPheRNo writes the intermediate outputs generated in this step into a directory called "tmp" in the current directory. To change the location of the intermediate results, see advanced settings.
In addition to the arguments above, one can use the following optional arguments to change the default settings.
- -ido, --input_dir_step_one (string, default='./Results'): Address of directory containing outputs of InPheRNo_step1.py
- -igp, --input_gene_pheno (string, default='Pvalue_gene_phenotype_interest_tmp.csv'): Name of Input2.2 (i.e. Output1.1) file containing gene-phenotype association p-values generated by InPheRNo_step1.py
- -itg, --input_TF_gene (string, default='Pvalue_gene_tf_tmp.csv'): Name of Input2.3 (i.e. Output1.2) file containing TF-gene association pseudo p-values generated by InPheRNo_step1.py
- -od, --output_dir (string, default='./tmp'): Address of directory for the intermediate results of this step pertaining to different repeats
- -pt, --Prior_T: Prior probability assigned to variables "T_ij" as described in the manuscript.
- -mnt, --max_num_TF (integer, default=15):, Maximum number of TFs which was used as a parameter in InPheRNo_step1.py. If a value other than default was used in InPheRNo_step1.py, the same value should be specified here as well.
- -nr, --num_repeat (integer, default=100): Number of times a PGM is trained with random initialization. Repeats are used to ensure stability of the results. At least 100 repeats are recommended.
- -ni, --num_iteration, (integer, default=200): Number of iterations used for MCMC
- -nb, --num_burn (integer, default=100): Number of iterations of the MCMC to burn. This number must be smaller than number of the iterations.
- -nt, --num_thin (integer, default=1): Number of iterations to thin by
This step only uses the intermediate outputs generated in STEP2.
The third step of InPheRNo generates a single output containing the phenotype-relevant TRN and the confidence score for each edge. This file by default will be located in the directory called "Results" placed in the current directory.
This is the main final result. This is a (gene x TF) matrix containing the confidence score for each phenotype-relevant edge. The confidence scores should be thresholded (e.g. only values larger than 0.5 kept) to obtain a network.
If you have already used the default parameters in STEP1 and STEP2, you can easily run the third step:
python3 InPheRNo_step3.py
If you have used advanced settings in previous steps or if you want to change the default values, you can use the following optional arguments.
- -id, --input_dir (string, default='./tmp'): Address of directory for the intermediate results of STEP2
- -nr, --num_repeat (integer, default=100): Number of times a PGM is trained with random initialization in STEP2. This number should be identical to the number used in STEP2.
- -od, --output_dir (string, default='./Results'): Address of the output directory
- -on, --output_network, (string, default = 'Final_phenotype_relevant_TRN.csv'): Name of the final result file.
A set of sample inputs and sample outputs are provided in this repository in the directories "Data", "Results" and "tmp". The results are obtained by running InPheRNo using default parameters.