Seqanswers Leaderboard Ad

Collapse

Announcement

Collapse
No announcement yet.
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • EdgeR LogCPM Bug?

    When I output the logCPM values for my genes, they are identical across contrasts. However, when I output the logFold values these seem to be correctly calculated for each contrast. Is this a bug or am I not understanding what the logCPM is that is returned by glmLRT?

    Code:
    library(edgeR)
    
    # Read in the quantification data
    data <- read.table('File.txt', header = TRUE, row.names = 1)
    
    keep <- rowSums(cpm(data) > 2) >= min(summary(3)) 
    data <- data[keep,]
    
    y <- DGEList(counts = data)
    y <- calcNormFactors(y)
    
    rownames(design) <- colnames(y)
    
    y <- estimateGLMCommonDisp(y, design, verbose = TRUE)
    y <- estimateGLMTrendedDisp(y, design)
    y <- estimateGLMTagwiseDisp(y, design)
    
    # Fit the linear model to the design matrix
    fit <- glmFit(y, design)
    
    # Make Contrast Matrix
    my.contrasts <- makeContrasts(....... # Different contrasts
    	levels = design)
    
    
    # Loop to calculate the contrasts
    for (i in 1:dim(my.contrasts)[2]) {
    # Create the log ratios
    	lrt <- glmLRT(y, fit, contrast = my.contrasts[,i])
    	tempLR <- lrt$table[2]
    	if(exists("LogCPM")){  LogCPM<  <- cbind(LogCPM, tempLR)   }
    	else{ LogCPM  <- tempLR }
    }
    
    # rows of LogCPM are identical?!?

  • #2
    This is not a bug, rather you are not understanding what logCPM represents.

    The help page for glmLRT says that logCPM is "the average log2-counts-per-million". The average is taken over all libraries in your dataset y, and hence is always the same regardless of the contrast being tested.

    logPCM is intended as a measure of the overall expression level of the transcript, and is displayed by functions such as plotBCV().

    BTW, it is not a simple average. The logCPM is computed using the edgeR function mglmOneGroup(), taking into account the estimated dispersions and the library sizes.

    Gordon

    Comment


    • #3
      Thanks! I was interpreting the documentation as meaning that the average was calculated for specific factor level combinations / contrasts. Wouldn't plotting the actual "A" for a specific contrast be more informative? Presumably that can be extracted by multiplying the matrix with the particular coefficients. What's the rationale for plotting the experiment-wide "A" rather than the condition-specific one? My question also applies to dispersion? What about condition-specific dispersion? Any help you could provide would be greatly appreciated!

      D

      Comment


      • #4
        Originally posted by danielfortin86 View Post
        Thanks! I was interpreting the documentation as meaning that the average was calculated for specific factor level combinations / contrasts.
        But there's nothing in the documentation that says so.
        Wouldn't plotting the actual "A" for a specific contrast be more informative? Presumably that can be extracted by multiplying the matrix with the particular coefficients.
        Since this is your suggestion, the onus is on you to suggest a reason why this would be more informative. For my part, I don't know even how you would define "the actual A" to be for a general contrast. For example, suppose you have contrast=c(-0.4,0.5,0,-0.1). How you would you define A and what would it tell you?
        What's the rationale for plotting the experiment-wide "A" rather than the condition-specific one?
        I already told you in my previous reply.
        My question also applies to dispersion? What about condition-specific dispersion? Any help you could provide would be greatly appreciated!
        I wonder whether you might be underestimating the statistical penalties and uncertainties introduced by fitting overly complicated models and estimating huge numbers of parameters, especially dispersion parameters. It is very unlikely that fitting condition-specific dispersions would be worthwhile, let alone a contrast-specific dispersions. But see

        Comment


        • #5
          Originally posted by Gordon Smyth View Post
          But there's nothing in the documentation that says so.

          Since this is your suggestion, the onus is on you to suggest a reason why this would be more informative. For my part, I don't know even how you would define "the actual A" to be for a general contrast. For example, suppose you have contrast=c(-0.4,0.5,0,-0.1). How you would you define A and what would it tell you?
          It *might* be informative to show a summary of original expression values and variability per group - just so you can see if edgeR's being silly and might need some tweaking ... (like when expression values are zero, or when there's outliers, or when you forgot to use tagwise dispersion, or when you used it... and would like to know what value edgeR used.... etc).

          Comment


          • #6
            Originally posted by earonesty View Post
            It *might* be informative to show a summary of original expression values and variability per group - just so you can see if edgeR's being silly and might need some tweaking ... (like when expression values are zero, or when there's outliers, or when you forgot to use tagwise dispersion, or when you used it... and would like to know what value edgeR used.... etc).
            Certainly may be useful to look individual expression values for leading genes, and all the edgeR cases studies show how to do this. The original poster was asking for something different however, for the definition of logCPM to be changed.

            Comment


            • #7
              I meant as extra columns for edgeR, not as a "leading gene thing". Stuff like a normalized mean and variability for each group. I've taken to moderating edgeR's results by removing results where the within-group variability is high, for example. (I think people are interpreting the logCPM value in that context... because they are expecting outputs that summarize group-wise summary information and don't see it.)

              Comment


              • #8
                when you perform edgeR analysis comparing two groups (mock vs treat), you get a list of genes with probabilities of differential expression with certain logFC. Assuming for this logFC edgeR is calculating a mean expression value for samples within each group,
                (1) how can I see what value is used (by edgeR) for each gene of each group?
                (2) how is this mean expression value calculated? (is it an average cpm values of samples of each group?)

                Comment

                Latest Articles

                Collapse

                • seqadmin
                  Best Practices for Single-Cell Sequencing Analysis
                  by seqadmin



                  While isolating and preparing single cells for sequencing was historically the bottleneck, recent technological advancements have shifted the challenge to data analysis. This highlights the rapidly evolving nature of single-cell sequencing. The inherent complexity of single-cell analysis has intensified with the surge in data volume and the incorporation of diverse and more complex datasets. This article explores the challenges in analysis, examines common pitfalls, offers...
                  06-06-2024, 07:15 AM
                • seqadmin
                  Latest Developments in Precision Medicine
                  by seqadmin



                  Technological advances have led to drastic improvements in the field of precision medicine, enabling more personalized approaches to treatment. This article explores four leading groups that are overcoming many of the challenges of genomic profiling and precision medicine through their innovative platforms and technologies.

                  Somatic Genomics
                  “We have such a tremendous amount of genetic diversity that exists within each of us, and not just between us as individuals,”...
                  05-24-2024, 01:16 PM

                ad_right_rmr

                Collapse

                News

                Collapse

                Topics Statistics Last Post
                Started by seqadmin, Yesterday, 08:58 AM
                0 responses
                9 views
                0 likes
                Last Post seqadmin  
                Started by seqadmin, 06-12-2024, 02:20 PM
                0 responses
                15 views
                0 likes
                Last Post seqadmin  
                Started by seqadmin, 06-07-2024, 06:58 AM
                0 responses
                182 views
                0 likes
                Last Post seqadmin  
                Started by seqadmin, 06-06-2024, 08:18 AM
                0 responses
                231 views
                0 likes
                Last Post seqadmin  
                Working...
                X