This program takes in names of genes (in one of several ways, see options below), finds the annotations in the in the included databases or a user generate one. It will then sort the annotations it finds based on their frequency in the subset vs their frequency in the database. This is done using a hypergeometric distribution function from scipy.
Using the command line interface ($ gws) will output the results to a file in one of two formats, a human readable version (default), or a tsv (tab separate value) sheet for further prossesing.
This can also be run as a webapp for a GUI, that is why flask is required.
To run this, use the CLI script and run it from the project directory.
It takes in gene identifiers in some form and outputs the words that are associated with those names in order of least likely to occur randomly at the top. It also outputs which genes had which word, and the genes occurrence compared to the total appearances of that word. These are the options, they indicate what type of input and output, make sure to review carefully so your output is everything you hoped it would be.
--buildDB Indicates the input are database files, will start interactive DB builder. No analysis will be done. --correctedP Sorts results by Holm–Bonferroni corrected p values, compensating for the multiple hypothesis problem. --file Indicates the input is a file with genes in it. --folder Indicates the input are directorys and will process all files in the directory. --gene_list Prints list of genes associated with each word. --help Show help message (see for shorthand commands). --low_rep Prints the words that occur in relatively few of the genes inputed. --out Location to write the file that contains the results, default is out.txt in current folder. --prob_cut This option takes one argument and sets the probability cutoff, default is 0.05. --species REQUIRED. Indicates species to use, maize and ath included. More can be added with '--buildDB'. --tsv This will give a tsv output for machine readable purposes. Default is human readable output. --universe Takes one argument, file with list of genes to use as universe for enrichment query. One gene per line or split by comma. --webLinks Outputs associated web links with standard gene output.
Option currently not enabled, due to disrepair and limited functionality --network Indicates the input are starting points of a network, will first return list of genes in those networks, then traditional output of that list of genes (Only works with maize at the momment).
This code is hosted on GitHub and can be accessed here <https://github.com/monprin/geneWordSearch/>
_. It has a dependency of SciPy. Now on to some examples.
For full output samples, check out this <https://github.com/monprin/geneWordSearch/tree/master/tests>
_ folder.
Example 1 (Human Output)
Input:
::
$ gws --species maize --gene_list --webLinks --file tests/MaizeIons/Al27_candidates.txt
Truncated Output Sample:
::
Multiple Gene Words:
Word: put-169a-panicum
P-value: 0.000104078977059
Corrected P-value: 0.233345066566
Overlap: 370/28229
Genes Appeared In: grmzm2g000510 grmzm2g001255 grmzm2g001296 ...
Word: expressed
P-value: 0.000289040114643
Corrected P-value: 0.646293696342
Overlap: 258/19207
Genes Appeared In: grmzm2g001296 grmzm2g002023 grmzm2g004531 ...
Word: chromatin
P-value: 0.00103289214074
Corrected P-value: 2.30025079743
Overlap: 14/487
Genes Appeared In: grmzm2g003002 grmzm2g010342 grmzm2g023983 ...
...
Web Links associated with these genes:
ttp://bbc.botany.utoronto.ca/efp_maize/cgi-bin/efpweb.cgi?primarygene=grmzm2g164141&modeinput=absolute
http://bbc.botany.utoronto.ca/efp_maize/cgi-bin/efpweb.cgi?primarygene=grmzm2g132212&modeinput=absolute
...
Example 2 (Machine Output)
Input:
::
$ gws --species maize --gene_list --webLinks --tsv --file tests/MaizeIons/Al27_candidates.txt
Output Sample:
::
Word Pval CorPval Ocurrances in Sample Ocurrances in Database Genes Appeared In
Multiple Gene Words:
put-169a-panicum 0.000104078977059 0.233345066566 370 28229 grmzm2g000510 grmzm2g001255 grmzm2g001296 ...
expressed 0.000289040114643 0.646293696342 258 19207 grmzm2g001296 grmzm2g002023 grmzm2g004531 ...
chromatin 0.00103289214074 2.30025079743 14 487 grmzm2g003002 grmzm2g010342 grmzm2g023983 ...
...
http://bbc.botany.utoronto.ca/efp_maize/cgi-bin/efpweb.cgi?primarygene=grmzm2g164141&modeinput=absolute
http://bbc.botany.utoronto.ca/efp_maize/cgi-bin/efpweb.cgi?primarygene=grmzm2g132212&modeinput=absolute
Finding the Genes
First the command line interface parses the arguments, depending on the task at hand, it will call the geneWordSearch function that does all of the heavy lifting. This function first imports the necessary databases and classes. After that it finds each gene in the database (a pickled python dictionary) and gets their respective lists of annotations. To analyse these annotations, the words are stored as an object with a list of genes being one the attributes. Web links are stored now for later display if desired.
Finding Each Individual Word and Counting
There is a class for the word that contains all of the statistics within one object, in order to make keeping track of everything easy and clear in the code. The list we created above is now sorted into alphabetical order and instances of each word are created and counted, this avoids computationally costly (on lists) membership tests.
Computing Relative Probability
To do the probability we used the hypergeometric distribution functions in the SciPy library, due to the fact that large factorials are required and are very time consuming, but SciPy can handle this almost instantaneously through what I can only assume is black magic. To compute this, the total word count database is loaded in and searched for the necessary words. What this database is and how it is created is discussed more below. It also computes a corrected probability using the Holm-Bonferroni method to handle the multiple hypothesis test. This value is displayed, but not used to sort or cutoff by default. This can be enabled using the '--correctedP' option.
Printing
Finally the results are printed out using one of two relevant class methods depending on the options set. These are fairly self-explanatory since there are output samples above.
Web App
The web app is based on flask, jQuery, Bootstrap, and Jansy Bootstrap (for file selector plugin). It is run by navigating to the main folder and running views.py. It automatically starts the server on localhost:5000, which can be changed by changing by editing the views.py file in the webapp folder.
Databases
---------
This code requires three databases that are all included in the package in the databases folder.
Gene Annotations Database
This is the database that contains all of the annotations for the specified species genome. It is stored as a pickled dictionary of GeneNote objects corresponding to each gene with its annotations after the words are all split and standardized. There is also a text version of this in a tsv file for reference purposes and error checking.
The program that creates it is DBBuilder.geneWordBuilder. It is run by including the '--buildDB' flag with the main CLI. It also takes in the positional arguments as file paths to the databases you want to use to build. It will ask you a few questions about the file formatting and then process it. It will do this for each file you include and then will save the fresh new database into a folder in the databases folder named after the species name you provided inside the databases folder.
Word Count Database
This database is a dictionary that is pickled in the databases folder for the relevant species. There is also a textual representation of this information in a tsv sheet for references purposes.
This is created by the function DBBuilder.totalWordCounts that essentially uses the same algorithm as the main gene word search, but instead of for specific genes, it adds the annotations for all of the genes in the database and just doesn't store some of the superfluous information. It takes no arguments and creates two files in the databases folder. It runs automatically after a new database is created.
Networks Database (currently deprecated - may be resurrected)
This is a dictionary that contains the lists of genes that are connected. it is a simple mapping from supplied gene to a list of genes. This in the future will be able to be supplied by the end user, but at the moment it is also static. It is created by DBBuilder.networksBuilder which just takes in a tsv and splits the rows. At the moment input needs to be sorted, but that functionality will be added in the future, along with support for arbitrary species, it only works with maize right now.