Team J-SCAMP's project for BioDataHack. Repurposing drugs and reclassifying diseases through unsupervised machine learning.
Versions used for development:
- python (v3.4)
- numpy (v1.11.3)
- pandas (v0.19.2)
- matplotlib (v2.0.0)
- sklearn (v0.18.1)
- opentargets (v3.1.0) (http://opentargets.readthedocs.io/en/stable/)
Running main.py will pull down a disease-gene data from OpenTargets (unless that dataset is in the current working directory) and return two plots. Those include 'pca.png', which shows Principal Components 1 and 2 where diseases are red points and drugs are blue, and 'kmeans.png', where the four most likely clusters are colourised. The drug-gene (& disease-gene) dataset should present in the current working directory.
By default main.py will perform dimension reduction with Principal Component Analysis, but Singular-Value Decomposition and t-SNE can be performed if clusterType is set to 'gene-svd' or 'gene-tsne'. Nb. these options will require editing main.py. Similarly, if disease-gene information needs to be pulled from OpenTargets the script will pull down 500 diseases by default, with more or less being pulled by changing the maxDiseases variable.
