view ks_distribution.R @ 2:2c8e564adede draft

Uploaded
author greg
date Tue, 01 Aug 2017 14:27:08 -0400
parents 56f42cc1dd58
children
line wrap: on
line source

#!/usr/bin/env Rscript

suppressPackageStartupMessages(library("optparse"))

option_list <- list(
    make_option(c("-c", "--components_input"), action="store", dest="components_input", help="Ks significant components input dataset"),
    make_option(c("-k", "--kaks_input"), action="store", dest="kaks_input", help="KaKs analysis input dataset"),
    make_option(c("-n", "--number_comp"), action="store", dest="number_comp", type="integer", help="Number of significant components in the Ks distribution"),
    make_option(c("-o", "--output"), action="store", dest="output", help="Output dataset"),
    make_option(c("-s", "--specified_colors"), action="store", dest="specified_colors", default=NULL, help="List of component colors")
)

parser <- OptionParser(usage="%prog [options] file", option_list=option_list)
args <- parse_args(parser, positional_arguments=TRUE)
opt <- args$options

set_component_colors = function(colors, number_comp) {
    # Handle colors for components.
    if (is.null(colors)) {
        # Randomly specify colors for components.
        component_colors <- c("red", "yellow", "green", "black", "blue", "darkorange")
    } else {
        # Handle selected colors for components.
        component_colors <- c()
        colors <- as.character(colors) 
        items <- strsplit(colors, ",") 
        for (item in items) { 
            component_colors <- c(component_colors, item)
        }
        num_colors_specified <- length(component_colors)
        if (num_colors_specified < number_comp) {
            # The number of selected colors is less than the number of
            # components, so we'll add random colors that were not
            # selected to the set of component colors until we have a
            # color for each component.
            loop_count <- number_comp - num_colors_specified
            for (i in 1:loop_count) {
                if (!(is.element("red", component_colors))) {
                    component_colors <- c(component_colors, "red")
                } else if (!(is.element("yellow", component_colors))) {
                    component_colors <- c(component_colors, "yellow")
                } else if (!(is.element("green", component_colors))) {
                    component_colors <- c(component_colors, "green")
                } else if (!(is.element("black", component_colors))) {
                    component_colors <- c(component_colors, "black")
                } else if (!(is.element("blue", component_colors))) {
                    component_colors <- c(component_colors, "blue")
                } else if (!(is.element("darkorange", component_colors))) {
                    component_colors <- c(component_colors, "darkorange")
                }
            }
        }
    }
    return(component_colors)
}

get_pi_mu_var = function(components_data, number_comp) {
    if (number_comp == 1) {
        pi <- c(components_data[1, 9])
        mu <- c(components_data[1, 7])
        var <- c(components_data[1, 8])
    } else if (number_comp == 2) {
        pi <- c(components_data[2, 9], components_data[3, 9])
        mu <- c(components_data[2, 7], components_data[3, 7])
        var <- c(components_data[2, 8], components_data[3, 8])
    } else if (number_comp == 3) {
      pi <- c(components_data[4, 9], components_data[5, 9], components_data[6, 9])
      mu <- c(components_data[4, 7], components_data[5, 7], components_data[6, 7])
      var <- c(components_data[4, 8], components_data[5, 8], components_data[6, 8])
    } else if (number_comp == 4) {
        pi <- c(components_data[7, 9], components_data[8, 9], components_data[9, 9], components_data[10, 9])
        mu <- c(components_data[7, 7], components_data[8, 7], components_data[9, 7], components_data[10, 7])
        var <- c(components_data[7, 8], components_data[8, 8], components_data[9, 8], components_data[10, 8])
    } else if (number_comp == 5) {
        pi <- c(components_data[11, 9], components_data[12, 9], components_data[13, 9], components_data[14, 9], components_data[15, 9])
        mu <- c(components_data[11, 7], components_data[12, 7], components_data[13, 7], components_data[14, 7], components_data[15, 7])
        var <- c(components_data[11, 8], components_data[12, 8], components_data[13, 8], components_data[14, 8], components_data[15, 8])
    } else if (number_comp == 6) {
        pi <- c(components_data[16, 9], components_data[17, 9], components_data[18, 9], components_data[19, 9], components_data[20, 9], components_data[21, 9])
        mu <- c(components_data[16, 7], components_data[17, 7], components_data[18, 7], components_data[19, 7], components_data[20, 7], components_data[21, 7])
        var <- c(components_data[16, 8], components_data[17, 8], components_data[18, 8], components_data[19, 8], components_data[20, 8], components_data[21, 8])
    }
    results <- c(pi, mu, var)
    return(results)
}

plot_ks<-function(kaks_input, component_colors, output, pi, mu, var) {
    # Start PDF device driver to save charts to output.
    pdf(file=output, bg="white")
    kaks <- read.table(file=kaks_input, header=T)
    max_ks <- max(kaks$Ks, na.rm=TRUE)
    # Change bin width
    max_bin_range <- as.integer(max_ks / 0.05)
    bin <- 0.05 * seq(0, (max_bin_range + 1 ))
    kaks <- kaks[kaks$Ks<max_ks,]
    h.kst <- hist(kaks$Ks, breaks=bin, plot=F)
    nc <- h.kst$counts
    vx <- h.kst$mids
    ntot <- sum(nc)
    # Set margin for plot bottom, left top, right.
    par(mai=c(0.5, 0.5, 0, 0))
    # Plot dimension in inches.
    par(pin=c(3.0, 3.0))
    g <- calculate_fitted_density(pi, mu, var, max_ks)
    h <- ntot * 1.5 / sum(g)
    vx <- seq(1, 100) * (max_ks / 100)
    ymax <- max(nc)
    barplot(nc, space=0.25, offset=0, width=0.04, xlim=c(0, max_ks), ylim=c(0, ymax), col="lightpink1", border="lightpink3")
    # Add x-axis.
    axis(1)
    for (i in 1:length(mu)) {
       lines(vx, g[,i] * h, lwd=2, col=component_colors[i])
    }
}

calculate_fitted_density <- function(pi, mu, var, max_ks) {
    comp <- length(pi)
    var <- var/mu^2
    mu <- log(mu)
    # Calculate lognormal density.
    vx <- seq(1, 100) * (max_ks / 100)
    fx <- matrix(0, 100, comp)
    for (i in 1:100) {
        for (j in 1:comp) {
            fx[i, j] <- pi[j] * dlnorm(vx[i], meanlog=mu[j], sdlog=(sqrt(var[j])))
            if (is.nan(fx[i,j])) {
                fx[i,j]<-0
            }
        }
     }
    return(fx)
}

# Read in the components data.
components_data <- read.delim(opt$components_input, header=TRUE)
number_comp <- as.integer(opt$number_comp)
if (number_comp == 0) {
    # Default to 1 component to allow functional testing.
    number_comp <- 1
}

# Set component colors.
component_colors <- set_component_colors(opt$specified_colors, number_comp)

# Set pi, mu, var.
items <- get_pi_mu_var(components_data, number_comp)
if (number_comp == 1) {
    pi <- items[1]
    mu <- items[2]
    var <- items[3]
} else if (number_comp == 2) {
    pi <- items[1:2]
    mu <- items[3:4]
    var <- items[5:6]
} else if (number_comp == 3) {
    pi <- items[1:3]
    mu <- items[4:6]
    var <- items[7:9]
} else if (number_comp == 4) {
    pi <- items[1:4]
    mu <- items[5:8]
    var <- items[9:12]
} else if (number_comp == 5) {
    pi <- items[1:5]
    mu <- items[6:10]
    var <- items[11:15]
} else if (number_comp == 6) {
    pi <- items[1:6]
    mu <- items[7:12]
    var <- items[13:18]
}

# Plot the output.
plot_ks(opt$kaks_input, component_colors, opt$output, pi, mu, var)