NEEP

Purpose

Null Empirically Estimated P-values (NEEP) is a non-parametric, high-throughput survival analysis meant for molecular expression data. The algorithm empirically finds the optimal cutoff (within a range) to separate expression values into low and high. The survival curves for this optimal cutoff are constructed. We calculate the MANTEL log-rank test for these optimally separated curves. The subsequent p-value distribution across all molecular objects will be skewed. To estimate the correct p-values, we sample 1M (or user specified) 'genes', which are a uniform permutation of the patient survival values from the provided clinical data. Then, p-values are estimated empirically and adjusted using the Benjamini-Hochberg approach.

Additional processing on NEEP output into protein interaction visualizations can be done using SINBAD: Survival-significant Isoform Networks By Altered Domain-inclusion (https://github.com/scwest/SINBAD). This project has been submitted to review for JOSS. Review status may be checked here:

Installation

Installation requires make, g++ and git. The g++ compiler must support the OpenMP API. GNU GCC (http://gcc.gnu.org/) will work, for example.

Run make check to confirm that OpenMP is supported by your compiler.

NEEP can be installed with the following commands.

git clone https://github.com/thecodingdoc/neep.git
cd ~/neep
make

The neep executable will be located in ~/neep. Add this to your path variable. This may be done by adding the following command to the .bashrc file.

export PATH=$PATH:~/neep

If git is not used but the repository zip is downloaded instead, the directory name should be changed to neep and chmod should be used to allow reading and writing.

The installation can be tested using the following commands.

cd ~/neep/test
sh test.sh

This command will execute NEEP on test data and produce test_output.txt. It will check that the survival calls made for genes with low p-value are in the same direction (high or low expression having increased survival) are the same between the baseline run (correct_ouput.txt) and the test run.

For preprocessing data before NEEP, any transformation which does not alter the expression order of patients for individual molecular objects will not affect NEEP results. For example, log-transformation is not required but not prohibited. The test data was generated and modified from the lung cancer (LUAD) dataset from the TCGA portal (https://portal.gdc.cancer.gov/). The test data should not be used for downstream biological analyses.

Usage

neep -c <clinical filename> -e <expression filename> -o <output filename> -n <number of bootstrap samples> -t <minimum threshold to check> -u

The minimum threshold to check refers to the lowest acceptable split between low and high expression when conducting KM tests. The maximum threshold will automatically be 1-t. The -u (optional) option is for to force a uniform null distribution. We were seeing some weird activity from c++ method random_shuffle() and we found that manually assigning a uniform distribution generator we obtained nulls which matched those from the Python and Julia programming languages (previous versions of NEEP).

Clinical file

Each row is a patient. There is no header. Days to event is a combination of the normal days to death and days to last followup used in clinical survival analysis. A 0 in event means that there are no death has been recorded (censored values).

<patient id>,<days to event>,<event>
...

Expression file Any type of expression can be used for NEEP: PPI abundance, gene expression, transcripts, transcript ratios, exon junction ratios, etc. This file is the normal expression matrix, where the rows are the molecular objects of interest and the columns are the patients. The patients must be the same set from the clinical file and they must be in identical order.

,<patient id 1>,<patient id 2>,...,<final patient id>
<molecular object 1>,<expression value>,<expression value>,...,<expression value>
...

Output file The output will be in tab-separated format with a header. Each line corresponds to one line from the expression matrix input. The molecular objects are ordered from lowest to highest adjusted NEEP p-value.

Column 1: molecular object ID; the same IDs used from the user defined expression file

Column 2: best log-rank statistic; the highest log-rank statistic generated across the threshold range

Column 3: the number of patients in the low expression group

Column 4: null empirically estimated p-value (neep)

Column 5: FDR adjusted neep; column 3 after FDR p-value correction

Column 6: direction (i.e. which survival curve is higher, low or high)

Column 7: hazard ratio. Please note that a high hazard ratio value indicates low probability of survival for patients expressing low values of a splice variant.

Column 8: 1 year mortality ratio

Column 9: 2 year mortality ratio

Column 10: 5 year mortality ratio

Contributing to NEEP

Making Contributions

Contributions to this software have been made by Dario Ghersi and Sean West. If you have any desired functionality or would like to contribute to NEEP in some way, please email Sean (seanchristianwest@gmail.com).

Reporting Issues or Problems with the Software

If any issues occur with downloading, installation, or running NEEP or if you find issues with the algorithm, please open a GitHub issue in the repository or email Sean (seanchristianwest@gmail.com).

How to Seek Support

For questions regarding running or the capabilities of NEEP, add a GitHub issue (https://github.com/thecodingdoc/neep/issues).