shfong / naga

Network Assisted Genomic Analysis Package

Network Assisted Genomic Analysis (NAGA)

Network Assisted Genomic Analysis(NAGA) re-prioritizes significant single nucleotide polymorphisms (SNPs) to genes using network diffusion methods including random walk and heat diffusion.

A companion website and REST API can be found at http://nbgwas.ucsd.edu/.

Documentation

A readthedocs page will be coming soon! In the mean time, you can view the documentations by building the sphinx documentation in the docs directory. Simply run the following in the docs folder

make docs

and open the index.html in the docs/build/html directory.

Installation

It is recommended that NAGA be run under Anaconda with python-igraph manually installed using conda and to create a new conda environment

To create a new conda environment and activate it:

conda create -n nagaenv
source activate nagaenv

If you would like to use naga in a Jupyter Notebook, you will need to add the Jupyter kernel. To do so:

# Make sure to activate the environment first!
conda install ipykernel # or pip install ipykernel
python -m ipykernel install --user --name nagaenv --display-name "Python (Naga)"

To install python-igraph via conda:

conda install -c conda-forge python-igraph

To install NAGA via pip:

pip install naga-gwas

Tutorial

notebooks/tutorial.ipynb demonstrates how to use this package once it is installed.

Citing NAGA

Carlin, Daniel E., Samson H. Fong, Yue Qin, Tongqiu Jia, Justin K. Huang, Bokan Bao, Chao Zhang, and Trey Ideker. "A Fast and Flexible Framework for Network-Assisted Genomic Association." iScience 16 (2019): 155.

naga's Issues

In tutorial Jupyter notebook refers to file under data/ directory that does not exist

The tutorial.ipynb notebook refers to the file ..data/scz/gene_level_summary_status_pmid_25056061.txt that does not exist.

Running

Notebook:

Need to revisit how to handle column names and column name defaults in the init vs read functions

Right now, the read function default kwargs actually overwrite the kwargs form init.

More broadly, how should we handle column names?

• when to validate (i.e. when to fail)
• what defaults are (integers vs names)
• Should init even be allowed to have kwargs
• Do we need a function dedicated to changing the column name designations?

Handle multiple transcripts in assign snp to genes

Generally, protein_coding_file would contain multiple lines per gene. Should improve the naive way simply taking the first line; instead, we should consider take the union of all the start and end position (longest possible region).

Protein coding table gene name should be documented

Currently, it's not documented that the gene names for the protein coding table needs to be in the dataframe's index. This should be corrected.

Update `neg_log_val()` in `utils.py` to optionally replace infinite values

Either in neg_log_val() in utils.py or in convert_to_heat() there needs to be a way to replace any infinite values with the maximum value found in array.

pandas >= 0.23.0 required

The sort-argument was added to pandas.concat in version 0.23.0. The pandas dependency should be replaced with pandas>=0.23.0 in setup.py.

https://pandas.pydata.org/pandas-docs/stable/generated/pandas.concat.html

https://github.com/shfong/nbgwas/blob/00e75fc4154b3504d790b253bb96f0f369204194/nbgwas/tables.py#L406

Handle exception: strong gene association(P-value=0) in convert_to_heat

As I was working with a GWAS summary stats, I ran into this case where I have a gene which its assigned p-value is 0. As it is in the current code, using neg_log would result in "inf" for the initial heat of this gene. Further, if I chose to normalize, there would be no heat for all genes across the network since dividing by the sum (which is inf) would result in 0 for all network genes.

How do we want to handle this exception?

Review and modify function names

Such as function diffuse, better include the word "kernel" in it

enable a quick access to gene rank prior to diffusion

Especially for binarization case, since it is trivial for -log(p-val)

parameter check

Not priority, but we should add parameter check to throw error when not-supported parameters is used in function.

Improve _validate_dataframe error messages

Currently, it just says these attributes are not in the column. It should say attribute x is not in dataframe header

Cytoscape lacks gene names

Thanks for getting this package together. We are using it for analysis of screen data. Anyway, everything works well except when Cytoscape (3.7.1) is evoked with CyRest the Gene name is not shown. Index number is present instead.

Validate heat vector prior to propagation

Should check for p-value of 0 before converting to heat via "negative_log" option. Ideally, there would be a flag to set ceiling of p-value/heat(negative_log).

tutorial Juptyer notebook refers to file SZ_hugo_genes_only.txt that does not exist

Under the Evaluating Performance section of notebooks/tutorial.ipynb the notebook attemps to load ../../nbgwas_analysis/data/gold_standards/SZ_hugo_genes_only.txt file which does not exist.

Screen shot:

Import network to a Nbgwas object

In nbgwas.py function random_walk(self, alpha=0.5), line 656: it assumes that network.node() are integer and has the field "name". This assumption breaks if I assign Nbgwas object with a networkx that already has nodes labeled with gene name (i.e. network.nodes() gives you a list like ['GeneA', 'GeneB']).

Consider:

1. Is function random_walk() a proper place to modify network properties?
2. How to handle the discrepancy in different ways of importing network to Nbgwas object?

Add enrichment score module and hypergeometic test function