dash - Dirichlet adaptive shrinkage for compositional data

• Authors: Kushal K Dey, Dongyue Xie, Matthew Stephens

Description

We present here a R package for fitting an adaptive method to shrink compositional counts data. We assume the input to be compositional counts with samples along columns and the composition categories along the rows. The method fits a multinomial model on each column, but we choose an adaptive prior - a mixture Dirichlet distribution with a given background probability as mean and concentration parameter ranging from very high values to very low values (even less than 1). The mixture proportions of the Dirichlet prior are estimated from the data.

For model formulation and the model fitting algorithm along with a discussion on choices of parameters, please check out our vignette

Example Applications of dash

We first provide an example application of dash in shrinking the base compositional probabilities in a Trasncription Factor Binding Site (TFBS) for a particular transcription factor.

library(Logolas)
library(grid)
library(gridBase)
library(ecostructure)
library(ggplot2)
library(Biobase)
library(dash)

We provide a simulation example of a position frequency matrix (PFM).

xmat <- cbind(c(5, 0, 2, 0),
              c(1, 1, 0, 1),
              c(100, 100, 50, 100),
              c(20, 50, 100, 10),
              c(10, 10, 200, 20),
              c(50, 54, 58, 53),
              c(1,1,1,3),
              c(2, 4, 1, 1))
rownames(xmat) <- c("A", "C", "G", "T")
colnames(xmat) <- paste0("pos-", 1:dim(xmat)[2])
xmat_norm <- apply(xmat, 2, function(x) return(x/sum(x)))

xmat

We fit the Dirichlet adaptive shrinkage (dash) model to the position frequency matrix generated above.

out <- dash(xmat, optmethod = "mixEM", verbose=FALSE, bf=TRUE)

We present the logo plot representations of the PWM matrix obtained by normalizing the sample PFM matrix and the one after applying dash. We use the R package Logolas to visually represent the logos.

grid.newpage()
layout.rows <- 1
layout.cols <- 2
top.vp <- viewport(layout=grid.layout(layout.rows, layout.cols,
                                      widths=unit(rep(6,layout.cols), rep("null", 2)),
                                      heights=unit(c(20,50), rep("lines", 2))))

plot_reg <- vpList()
l <- 1
for(i in 1:layout.rows){
  for(j in 1:layout.cols){
    plot_reg[[l]] <- viewport(layout.pos.col = j, layout.pos.row = i, name = paste0("plotlogo", l))
    l <- l+1
  }
}


plot_tree <- vpTree(top.vp, plot_reg)

color_profile = list("type" = "per_row", 
                     "col" = RColorBrewer::brewer.pal(4,name ="Spectral"))

pushViewport(plot_tree)
seekViewport(paste0("plotlogo", 1))
logomaker(xmat_norm,color_profile = color_profile,
          frame_width = 1,
          pop_name = "pre dash PWM",
          newpage = F)

seekViewport(paste0('plotlogo',2))
logomaker(out$posmean,color_profile = color_profile,
          frame_width = 1,
          pop_name = "post dash PWM",
          newpage = F)

Contact

For any queries or questions or bugs, please open an issue in this Github page or write to Kushal K Dey at [email protected].