GGID

Protein annotation via Information Diffusion over GO-derived protein Graphs. The deployed app lives here.

Table of Content

1. Graphical Abstract
2. Overview
3. Installation
4. Deployment
5. License

Graphical Abstract

Overview

A detailed description of the method is available as part of the deployed application (THEORY tab). However, here is a brief overview:

• We can annotate proteins by propagating/diffusing labels through protein networks

• I created a new type of network for the human kinases, one based purely on GO term annotations

• To diffuse labels through this network, I also created a web-app interface using the Dash framework combined with the Cytoscape library

• In this repo are both the code to create the network, and code for the Dash web app

Installation

Assuming you have python3.6+ installed:

1. Clone repo to local machine with

   $ git clone https://github.com/ily123/ggid

2. Create a virtual env

3. Inside the env, install dependencies with

   $ pip install -r requirements.txt

Usage

1. The network needed to run the front-end is pre-computed (netowrk/kinase_matrix.pkl), so to run the front-end locally issue the following command in your terminal:

   $ python application.py

   and then open the app in your browser at localhost:8050.

2. To regenerate/modify the network, run the create_go_similarity_kinase_network.ipynb notebook and follow it cell by cell.

   • Note: the Gene Ontology files (annotations and the term ontology) that the network is constructed from are included under data/. You can manually download the latest versions of these files from the Gene Ontology website. The code expects gaf-2 format for the annotation corpus, and obo for the term ontology.

3. There are a couple of other notebooks included, one has some exploratory analysis (_explore_human_annotations.ipynb) and the other scrapes human kinase names from uniprot (_get_human_kinases_from_uniprot.ipynb). The kinases are already included under data/, but you can re-run the notebook if you want to update the list.

Deployment

The front-end is a Flask app, so it can be deployed seemlessly with most cloud providers. I deployed it with Heroku, and it was rather painless. Note that if you plan to plug in a proteome-scale network into the tool, you'll have to account for additional space/time requirements. The tool, as is, is designed for a small network with near-instant diffusion times.

License

I don't know how to license this repo.

Parts of this work have been published & publicly presented before, so I imagine they are part of public domain now. So feel free to fork.

That being said, I am not an IP lawyer, so on the tiny off-chance you plan to use this code commercially, message me on github/linkedin/etc and I can put you in touch with people who know the answer.