Hi,

After running my linear model using microarray data, I am using plotSA to check if there is a generally monotonic relation between the average log expression and the residual standard deviation. The plot detects a few outlier probes and highlights them in red. I was wondering if you wanted to know which are those probes exactly and not guess based on the mean and SD, I was wondering how can I do that?

```

Check the plotSA source function on how outliers are detected : https://rdrr.io/bioc/limma/src/R/plots-fit.R

Particularly the "Check for outlier variances" part, you can run the function step by step and you can identify which points are assigned as outliers :

if(length(fit$df.prior)>1L) {
        df2 <- max(fit$df.prior)
        s2 <- fit$sigma^2 / fit$s2.prior
        pdn <- pf(s2, df1=fit$df.residual, df2=df2)
        pup <- pf(s2, df1=fit$df.residual, df2=df2, lower.tail=FALSE)
        FDR <- p.adjust(2*pmin(pdn,pup),method="BH")
        colv[FDR <= 0.5] <- col[2]
    }

Identifying and trouble-shooting the genes with outlier variances is fine. Removing them just because they are outliers is not fine. limma has already treated them correctly and you do not need to micromanage the limma analysis. Taking it upon yourself to remove the outlier probes manually may damage or bias the analysis.

The hyper-variable genes are those with small values for df.prior. Assuming that your fit object includes gene annotation:

Genes <- data.frame(fit$genes, df.prior=fit$df.prior)
o <- order(Genes$df.prior)
Genes <- Genes[o,]
head(Genes)