In cit-bioinfo/BuildConsensus: Build a consensus classification from multiple classification methods

Overview

This package implements methods for building a consensus molecular classification using existing classification systems. Functions implemented in this package were used for the Consensus Classifications of Colorectal and Bladder Cancer.

Example

Three methods of reaching a consensus classification are implemented: COCA (Cluster of Clusters Analysis), MCL (Markov Cluster Algorithm) and CIT, which was used in Rebouiossou et. al.

All methods require the same input data, which is a data.frame with samples as observations (rows) and the different classification systems as variables (columns).

As an example, we can use the table for the six classification systems in muscle-invasive bladder cancer (MIBC).

library(BuildConsensus)
data(blca_class)

head(blca_class, 8)

library(BuildConsensus)
data(blca_class)
kableExtra::kable_styling(knitr::kable(head(blca_class, 8)), c("striped"))

MCL algorithm

To run the MCL algorithm using consensus.MCL, the mcl library has to be installed in your system. For installation instructions, please refer to Github README file: https://github.com/cit-bioinfo/BuildConsensus/ The MCL algorithm itself is outlined in the algorithm website: https://micans.org/mcl/

Once the data is provided in the data.frame format described above, we can run the MCL algorithm by also specifying the inflation factors and the output directory. Multiple inflation factors can be evaluated in one run.

A simple example can be run using:

mcl_res <- consensus.MCL(blca_class, I.values = 3, outdir = "MCL_res",n.iter = 500, sim.method = "CohenKappa")

This inflation value for this dataset renders a K=4 solution, as seen below.

knitr::include_graphics("MCL_res/Heatmap_I30.png")

When multiple inflation values are being compared, the silhouette vs inflation factor plot can guide the user towards selecting the optimal result.

knitr::include_graphics("MCL_res/silhouette_vs_InflationFactor.png")

Reproducing the MIBC BLCA Consensus

The Consensus Classes of MIBC were computed using the MCL algorithm and the parameters described below. Warning: this will take a long time to run.

mcl_res <- consensus.MCL(blca_class, I.values = seq(3, 15, 0.3), outdir = "../MCL_res", resamp = .8, n.iter = 500, sim.method = "CohenKappa", filter.ini = "fisher", ck.cut = 0)

Note that due to a random element, your solution may be differ slightly from the reported Consensus solution.

cit-bioinfo/BuildConsensus documentation built on Nov. 27, 2019, 11:29 a.m.