SADRAT - Smart Adverse Drug Reaction Assessment Tools

📕Contribution Guidelines for FOSSHACK 2020

1. History and Motivation

This project was started in Autumn Session 2019 by a research group at National Institute of Technology Rourkela under the guidance of Prof. B.P. Nayak(the same group that runs purplepotion) of Regenerative and Rehabilitative Medicine Laboratory, Department of Biomedical Engineering. The motivation behind this research is to improve the performance of the state-of-the art ML and DL models to detect Adverse Drug Reaction Signals(ADRS) from social media platforms like Twitter using the concept of Data Programming .
How it might look like:
Sample Tweet:
Tweet: "Having Headaches since morning! I need to get off of crocin!"
Sample Result:
ADR Probability: 88%
Detected Drug: Crocin
Detected ADR: Headache
This was quite an easy one to guess. But in real scenarios, sentence structures can be very complicated. One of the major improvements and novelty we want to bring in the pipeline of performing such tasks is to remove the human involvement in preparing a labelled dataset in order to perform supervised learning, instead, we want to make the process programming weakly supervised strategies and at the same time reducing the noise generated during the process to a minimum. The above classification might seem a simple enough task for a modern classifier. But, the bigger challenge that we are attempting to solve here are the following:

1. Creating a minimum noise, labelled training set for supervised classification.
2. Removing direct manual involvement in preparing the golden training and testing datasets.
3. Creating a highly scalable programmable alternative to maximize performance and minimize time required to generate labelled training sets.
   Finally, we would develop a web app around the models to interact with them from the client side.

2. Prerequisite Knowledge

1. A basic knowledge of Machine Learning(ML) and mathematics involved in ML.
2. An intermediate knowledge in Python (3.x) Programming Language. i.e. you should comfortable with OOP, Web development with flask, working with APIs, libraries and frameworks.
3. Project specific knowledge (discussed in 3.)

3. Project Specific Prerequisites:

1. Learn about Data Programming from here.
2. Learn about Snorkel from here
3. Learn about flask and dash - for contributing to the web app

What if I don't have any prerequisite knowledge? How can I contribute?

Each and every person reading this is a potential contributor for us. We have contribution opportunities for everyone irrespective of knowledge and experience. You can contribute.

1. For complete beginners having no previous programming knowledge: If you have spent some time reading about the project and what we are willing to achieve, you can contribute to the docs. You can help us spread the word about this project and help us get more people involved :)
2. People with Knowledge of Python but no ML knowledge: You can contribute to the web app. See (2.) above.
3. People with Knowledge of Python and ML: you should be able to contribute to each and everything we are working on here!
   

What else will I get?
Friends! We are people from diverse backgrounds and interests. Some are even working/incoming FTEs in well known software firms. This would be a great way to know each other and contribute to a single cause!

NOTE: Every contributor is valuable to us. Hence each and every contributor irrespective of the "type" of contribution, would be mentioned in the "contributors" section of the page.

COMMUNICATION

All communications regarding FOSSHACK 2020 would take place in the #fosshack2020 Slack Channel : LINK (expires in 30 days, i.e on 10th October 2020)

👨‍💻 Contributors 👩‍💻:

Author : Shaswat Lenka
Top Contributors: Debabrata Panigrahi, Ankit Samota, Vedant Raghuwanshi, Abhijeet Sahoo, Roshan Kumar Shaw

Research Overview

There is enough research evidence that social media can be an important source of indicating Adverse Drug Reactions and analyzing disease trends in a population. Although this signal is weak, many algorithms have been developed to extract the important signals that depict a valid ADR. However, the prediction of future disease trends from social media data in a population under study is a challenging task and breakthroughs have not been made in this direction. Also, another epidemiologic challenge that demands to be solved is predicting possible reason(s) behind the appearance of such trends for further verification and validation of the Early Warning System.

Hence, SADRAT would tackle the above-mentioned issues providing the pharmaceutical companies with the necessary parameters and predictive outcomes (i.e. Disease trends, probable reasons for disease etc. using predictive analytics) in a dashboard that would leverage their decision-making (related to marketing strategies, venturing into new market and the introduction of new and upgraded drugs) while targeting a particular population based on pivots such as season, age, gender, race, etc.

Installing Spacy

To correctly instal spacy, run these command in sequence in your Virtual environment to avoid getting runtime exceptions.
pip install spacy
python -m spacy download en

Installing TextBlob

pip install textblob
python -m textblob.download_corpora

Setting up the Jupyter Notebook Development Environment for Research:

pip install jupyter - In your virtual environment
ipython kernel install --user --name=yourvirtualenvname - To add a jupyter kernel

Build Instructions:

To build the project in your local computer:

1. Create a virtual environment for the project. (Highly Recommended)
2. clone the project, from your forked repository.
3. cd to the project root directory and install the dependencies: pip install requirements.txt

4. if you are using some IDE, make sure you have untracked all the hidden cache files in .gitignore.

   How to send a Pull Request(PR):

   We strongly recommend following a git-flow like workflow. If this looks too complicated to you, just follow good practices and naming conventions in your branches. Once you are done with your code in your branch, you can send a PR.
   Steps:
   (Asuming you are using a Linux/Unix system and already have git installed. For windows users, things might be a little different but the steps would mostly be the same.)

   1. fork this repository.
   2. Open terminal (in case of windows, open Git Bash), and clone your forked repository by this command: git clone <your_ssh_key> , you can get by clicking on the green "Code" button on your forked repository page and copy the text under the "clone with SSH". In my case, my SSH key looks something like this: git@github.com:purplepotion/sadrat.git. But before that you must have set up you ssh keys with GitHub.
   3. Next, set up the origin and upstream in of your cloned repository. Origin refers to where your code will get committed to i.e. your forked repository on GitHub. Your upstream would be this repository with which you would be keeping your cloned repository up to date.
      To add upstream: git remote add upstream git@github.com:purplepotion/sadrat.git Origin would automatically be set to your forked repository by default.
   4. Now create a branch to which you would be making changes, use a good naming convention so that you remember what you were doing in which branch.
      git checkout -b <your_branch_name> . After you make changes, commit to the branch and when done, push it to the origin by doing the following steps:
      git pull upstream master - Always do this before you commit as to stay updated with the upstream repository.
      git push origin <your_branch_name> - this will push your branch to your origin i.e. your forked repository.
   5. Now go this repository and send a PR!
   6. Wait for your PR to be reviewed and once done and no further modifications are requires, your PR will be accepted and merged successfully!