React: Drug Safety in Your Pocket
Team: scikit_hammertime
==================================
Jake Beard: jake at minnow dot io
Anjney Midha: anjney at stanford dot edu
Ankit Kumar: ankitk at stanford dot edu
Jay Hack: jhack at stanford dot edu
Ross Lazerowitz: rosslazer at gmail dot com
Bayes Impact Hackathon 2014
==================================
The Problem:
- 100,000 Americans die each year due to known drug side effect
- Existing tools for users to search for adverse drug interactions are clunky, database level query interfaces
- Existing tools are limited to reported drug events - which are severely prone to underreporting
Solution:
- We use a distributed representation of the AERS ( Federal Drug Adverse Event Reporting System) dataset classified by the RxNorm hierarchy, using neural networks to predict novel interactions for pairs of drugs that do not have a historical interaction record
Results:
- We achieve 82% initial accuracy on novel unseen drug interactions
- Next steps are to include RxNorm features such as chemical composition.
==================================
acknowledgements:
Libraries used:
- word2vec : https://code.google.com/p/word2vec/
- scikit-learn: http://scikit-learn.org/stable/
- pandas: http://pandas.pydata.org/
Papers Referenced:
Log in or sign up for Devpost to join the conversation.