CEPICS: a comparison and evaluation platform for integration methods in cancer subtyping.
1.0.0
Han Xu, Kuo Song, Ran Duan
Han Xu myxuxiaohan@outlook.com, Kuo Song uysk@foxmail.com
Download CEPICS_1.0.0.tar.gz.
Install R package locally from R studio.
We have prepared init.R to help install the packages that CEPICS depends on.
Download init.R.
Source init.R in R studio.
source(init.R)https://www.frontiersin.org/articles/10.3389/fgene.2019.00966/full
Just upload the data, datatype, methods and some kMax to get the report.
CEPICS(datalist, datatype = c("gaussian", "gaussian", "gaussian"),
functionList = list('PINS', 'LRA', 'SNF', 'iCluster', 'PFA'), kMax=5)Please make sure that MATLAB has been installed on your computer if you want to run PFA.
You can also set each parameter for each method. For more details, please see Supplementary Materials or the help in this R package.
Here, we present four reports generated by CEPICS for three scenarios.
Download CEPICS Reports for all scenarios and Supplementary Materials.
The reports generated by CEPICS consist of two parts.
The first part shows the comparison of subtyping results across different methods and different numbers of subtypes, including time consumption, Cox p-value, NMI, ARI, silhouette coefficient. Then, we present an overall samples similarity heatmap representing a robust prediction for sample pairwise similarities.
There are two comparison strategies for the NMI and ARI depending on the availability of true labels of patients. If you upload an empirical pre-determination of subtypes for samples by experts or clinicians based on clinical phenotypes, images or experience, CEPICS will take it as gold standard to compare. If you don’t have any subtype information, CEPICS will calculate NMI and ARI between the results of every two methods for each k, and then calculated the average NMI and ARI for each method at each k.
- True label available
- True label not available
When the true labels are available, CEPICS will find the best method based on the true labels considering Cox p-value, NMI, ARI, and silhouette coefficient in the frist part of the report.
The second part shows the performance of each method, including summary information of all metrics (when the true labels are available in Scenario 2 and 3.2, see Supplementary Materials for more details), KM survival curves, and patient similarity heatmaps for different subtype numbers of each method.
