jirihon / mg-feature-select

Metagenomic Feature Select

R/Bioconductor script for hierachical clustering by various DNA sequence features and measuring its similarity with reference clustering.

How to Run

1. Install the latest stable version of R (3.3.0) from CRAN.

2. Run the latest R and install packages Rcpp, inline, fastcluster and ggtree using Bioconductor installer.

   source("https://bioconductor.org/biocLite.R")
   biocLite(c("Rcpp", "inline", "fastcluster", "ggtree"))

3. Download the script feature_select.R from GitHub into current working folder and load it into R.

   source("feature_select.R")

4. Use functions load_features, eval_all_feature_comb, find_best_features and plot_tree to run the analysis and visualize the results. The most time consuming function eval_all_feature_comb can be run in parallel mode (set parallel option to TRUE). See the following example:

   features <- load_features("features.csv")
   res <- eval_all_feature_comb(features, min_features = 2, max_features = 6, parallel = TRUE)
   best_euc <- find_best_features(res, 'euc') # for euclidean distance
   best_cos <- find_best_features(res, 'cos') # for cosine dissimilarity
   plot_tree(best_euc$tree, features$Organism, "tree_euc.pdf")
   plot_tree(best_cos$tree, features$Organism, "tree_cos.pdf")

   

5. Feel free to inspect the structure of best_euc and best_cos variables. It is just a list of three items: features contains a vector of used features, k is a cutoff value (number of clusters) of the hierarchical clustering tree, that gave the best Variantion of Information (VI) with reference clustering and vi contains the best VI value.

6. You can also plot a 2D projection of the clustering for comparison with reference:

   ref_cls <- features$Organism
   plot_clustering_2d(features$HP1_aktivita, features$HP2_mobilita, ref_cls, "2d_ref.pdf")
   best_cls <- cutree(best_euc$tree, best_euc$k)
   plot_clustering_2d(features$HP1_aktivita, features$HP2_mobilita, best_cls, "2d_euc.pdf")