cca_cancer

Code accompanying our ISBI 2021 paper "Multimodal Fusion using Sparse CCA for Breast Cancer Survival Prediction" https://arxiv.org/abs/2103.05432 by Vaishnavi Subramanian, Tanveer Syeda-Mahmood, Minh N. Do

Code

The code for CCA is in MATLAB, and that for prediction is in Python. The code is structured as follows:

• code: Contains all the functions to run the CCA methods, the deflation methods, the K-SCCA and K-GCCA code, and code to simulate and evaluate metrics on simulated data.
• data: This directory needs to be created. Data can be downloaded from https://uofi.box.com/s/pvhs03t8u9wfjs6j31c6wbvec7237wfc and unzipped to subdirectories data/brca and data/simulated respectively.
• results: This directory contains all the results from our experiments reported in the conference paper.
• run: All code to re-run our experiments are provided in this directory.

Requirements

Python packages:

• numpy==1.18.5
• pandas==1.1.2
• scipy==1.5.2
• scikit-learn==0.23.2

Instructions to run code

Here, we provide the details to re-run the code provided in the run directory.

---

1. Simulated data

Code location: run/simulated/

Running SCCA on simulated data

• Run run_1_scca_on_simulated.m in MATLAB.
• Once done, run test_1_scca_on_simulated.m in MATLAB to accumulate all results.

Running GCCA on simulated data

• Run run_1_gcca_on_simulated.m in MATLAB. The code requires an input parameter indicated if the graph used in GCCA is based on prior knowledge run_1_gcca_on_simulated('prior1') or inferred from data run_1_gcca_on_simulated('no-prior').
• Once done, run test_1_gcca_on_simulated.m in MATLAB to accumulate all results. Here also, the code requires the same input parameter as above.

Summarizing simulation results

• Run the Jupyter notebook summarize_results_simulations.ipynb after completing the previous two steps.

---

2. 1-dimensional CCA on BRCA data

Code location: run/1_xcca_on_brca/

Running SCCA on BRCA data

• Run run_1_scca_on_brca.m in MATLAB.

Running GCCA on BRCA data

• Run run_1_gcca_on_brca.m in MATLAB. The code requires an input parameter indicated if the graph used in GCCA is based on prior knowledge run_1_gcca_on_brca('prior1') or inferred from data run_1_gcca_on_brca('no-prior').

Summarizing results

• Run the Jupyter notebook summarize_results_1_xcca_brca.ipynb after completing the previous two steps.

---

3. K-dimensional CCA on BRCA data

Code location: run/k_xcca_on_brca/

Running SCCA on BRCA data

• Run run_k_scca_on_brca.m in MATLAB.

Running GCCA on BRCA data

• Run run_k_gcca_on_brca.m in MATLAB. The code requires an input parameter indicated if the graph used in GCCA is based on prior knowledge run_k_gcca_on_brca('prior1') or inferred from data run_k_gcca_on_brca('no-prior').

Summarizing results

• Run the Jupyter notebook summarize_results_k_xcca_brca.ipynb after completing the previous two steps.

---

4. Survival prediction with K-dimensional CCA on BRCA data

Code location: run/k_xcca_on_brca/survival_prediction

Run the Jupyter notebook 1_predict_1yr_survival_classification.ipynb after generating the embeddings using the K-dimensional CCA above.

---