An iteractive consensus clustering framework for multi-platform data analysis
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
R
data
imgs
inst/shiny
man
tests
vignettes
.Rbuildignore
.gitignore
CrossICC.Rproj
DESCRIPTION
LICENSE
NAMESPACE
README.Rmd
README.md

README.md

CrossICC

Table of the content

Overview


Unsupervised clustering of high-throughput molecular profiling data is widely adopted for discovering cancer subtypes. However, cancer subtypes derived from a single dataset are not usually applicable across multiple datasets from different platforms. We previously published an iterative clustering algorithm to address the issue (see this paper), but its use was hampered due to lack of implementation. In this work, we presented CrossICC that was an R package implementation of this method. Moreover, many new features were added to improve the performance of the algorithm. Briefly, CrossICC utilizes an iterative strategy to derive the optimal gene set and cluster number from consensus similarity matrix generated by consensus clustering. CrossICC is able to deal with multiple cross platform datasets so that requires no between-dataset normalizations. This package also provides abundant functions to help users visualize the identified subtypes and evaluate the subtyping performance. Specially, many cancer-related analysis methods are embedded to facilitate the clinical translation of the identified cancer subtypes.

There are two modes for the integration of clusters derived cross-platform datasets: cluster mode and sample mode. For cluster mode, samples from each platform are clustered separately and centroids of each sub cluster derived from ConsensusClusterPlus were further clustered to generate super cluster. This process avoided removing batch effect across platforms. The details step by step illustration of this algorithm can be found in our previous published paper and our recent submitted paper[coming soon]. For sample mode, sub clusters were firstly derived from ConsensusClusterPlus in each platform. We then calculated correlation coefficient between samples and centroids of clusters to get a new feature vector of each samples. Based on this new matrix, samples were divided into new clusters.

Installation

Via GitHub (latest)

install.packages("devtools")
devtools::install_github("bioinformatist/CrossICC")

Usage

CrossICC has the ability to automatically process arbitrary numbers of expression datasets, no matter which platform they came from (Even you can use sequencing and microarray data together). What you only need is a list of matrices in R, without any type of pre-processing (never need manipulation like filtering or normalization).

library(CrossICC)
CrossICC.obj <- CrossICC(demo.platforms, skip.mfs = TRUE, max.iter = 100, 
                         cross = "cluster", fdr.cutoff = 0.1, 
                         ebayes.cutoff = 0.1, filter.cutoff = 0.1)

CrossICC will automatically iterate your data until it reaches convergence. By default, CrossICC will generate an .rds formatted object in your home path (~/, a.k.a $HOME in Linux), followed by an shiny app as shown below that is opened in your default browser, which provides you a very intuitive way to view the results.

FAQ

  • Question 1: NA values involved in our data set, how to go through them?

A: Users may encounter other unexpected error due to NA values in raw dataset. Therefore, we strong recommanded that users checked the NA valus in thire data set before loading it into CrossICC. To check the completed cases in matrix, completed.cases can be a good option to do that. Here, we also present an example for users to impute there data in case they don't want to remove case in the dataset. The imputation method shown here are KNN method from impute package.

# for a individual matrix, plz do imputation using the following r code
tempdata.impute=impute.knn(as.matrix(tempdata) ,k = 10, rowmax = 0.5, colmax = 0.8)
normalize.Data=as.data.frame(tempdata.impute$data)

Contribution

Qi Zhao @likelet and Yu Sun @bioinformatist implemented the packages. Zhixiang Zuo @zhixiang supervise the project. Zekun Liu performed the test and helped run an example of the package. For more information or questions, plz contact either of the authors above.