StableCox 0.3

The official implement of Stable Cox

System Requirements

Hardware requirements

`Stable Cox' package requires only a standard computer with enough RAM to support the in-memory operations.

Software requirements

OS requirements

This package is supported for Linux. The package has been tested on the following system:

• Linux: Ubuntu 18.04

Python Dependencies

'Stable Cox' mainly depends on the Python scientific stack.

lifelines=0.27.8
numpy=1.20.3
pandas=2.0.3
scikit-learn=1.3.0

Run demo

omics data

Select topN biomarker and build a predictor on the selected biomarker panel

from StableCox import StableCox import pandas as pd training_pd_data = pd.read_csv('./omics_data/HCC_cancer/train_median.csv', index_col=0)

test1_pd_data = pd.read_csv('./omics_data/HCC_cancer/test1_median.csv', index_col=0)

SC = StableCox(alpha=0.0005, hidden_layer_sizes = (98, 11), W_clip=(0.4, 4))

duration_col = "Survival.months"

event_col="Survival.status"

SC.fit(training_pd_data, duration_col, event_col)

cindex = SC.predict_with_topN(test1_pd_data, topN=10)

print("cindex", cindex)

clinical data

Make prediction directly without biomarker selection

import pandas as pd from StableCox import StableCox training_pd_data = pd.read_csv('./clinical_data/breast_cancer/breast_train_survival.csv', index_col=0)

test1_pd_data = pd.read_csv('./clinical_data/breast_cancer/breast_test1_survival.csv', index_col=0)

training_pd_data = training_pd_data.drop(['Recurr.months', 'Recurr.status', 'Cohort'], axis=1)

test1_pd_data = test1_pd_data.drop(['Recurr.months', 'Recurr.status', 'Cohort'], axis=1)

SC = StableCox(alpha=0.002, hidden_layer_sizes = (69, 15), W_clip=(0.02, 2))

duration_col = "Survival.months"

event_col="Survival.status"

SC.fit(training_pd_data, duration_col, event_col)

cindex = SC.predict(test1_pd_data)

• The expected running time is from several seconds to mins depends on the number of samples.

License

This project is licensed under the terms of the MIT license.