Dear RMA experts, I recently had a necessity to look into details of the affy package and was confused on the estimation of alpha parameter for normexp convolution model in RMA. Namely, alpha is estimated as 1/mode(data > overall_mode). Now, 1. Why mode is used to estimate mean of exponential distribution? 2. for continuous unimodal distribution, wouldn't mode(data > overall_mode) be the same as overall_mode (disregarding edge effects) ? What am I missing? Thank you in advance, Zufar Here is affy R code that estimates RMA model parameters --- > bg.parameters function (pm, n.pts = 214) { max.density <- function(x, n.pts) { aux <- density(x, kernel = "epanechnikov", n = n.pts, na.rm = TRUE) aux$x[order(-aux$y)[1]] } pmbg <- max.density(pm, n.pts) bg.data <- pm[pm < pmbg] pmbg <- max.density(bg.data, n.pts) bg.data <- pm[pm < pmbg] bg.data <- bg.data - pmbg bgsd <- sqrt(sum(bg.data2)/(length(bg.data) - 1)) * sqrt(2) sig.data <- pm[pm > pmbg] sig.data <- sig.data - pmbg expmean <- max.density(sig.data, n.pts) alpha <- 1/expmean mubg <- pmbg list(alpha = alpha, mu = mubg, sigma = bgsd) } <environment: namespace:affy=""> [[alternative HTML version deleted]]