Analysis of GDSC Data

A brief video summary of this project can be found here on YouTube.

/assets

Contains relevant csv files

Raw data

• GDSC2_fitted_dose_response_25Feb20.csv - contains the raw IC₅₀ data for 135,242 drug / cell line combinations; data can be downloaded here
• cell_list.csv - contains GDSC-generated information about the included cell lines, including tissue and TCGA classification; data can be downloaded here
• drug_data.csv - contains GDSC-generated information about the included drugs, including drug pathways and targets; data can be found here

Processed data

• cell.csv - results of the analysis from GDSC_Project.ipynb for the cell data, including contains two dimensions from PCA and t-SNE for plotting, the cluster identities from k-nearest neighbors using full, PCA-transformed, and low rank approximations of the data, and the mean lnIC₅₀ per cell line
• cell_lrm.csv - low-rank approximation of the cell matrix


• drug.csv - results of the analysis from GDSC_Project.ipynb for the drug data, including contains two dimensions from PCA and t-SNE for plotting, the cluster identities from k-nearest neighbors using full, PCA-transformed, and low rank approximations of the data, and the mean lnIC₅₀ per compound
• drug_lrm.csv - low-rank approximation of the drug matrix


/src

Contains relevant scripts and notebooks

• GDSC_Project.R - contains
• GDSC_Project.ipynb - contains
• kmeans.py
  • find_kmeans: find an optimal number clusters via the elbow method and fit k-means with this many clusters
  • plot_kmeans: plot the SSE vs. clusters and elbow point for cell and drug data
• lowrank.py
  • fit_svd: fit SVD model iteratively for a given rank r
• pca.py
  • find_pc: returns the eigenvalues and eigenvectors of the covariance matrix
  • project_pca: transforms the original matrix via projection using a specified number of principal components
  • plot_pca: plot the variance by principal component and cumulativev variance by principal component
• utils.py
  • import_data: loads GDSC data and pre-process into a wide matrix
  • process_data: produces mean-centered data and masks

Summary