multiple comparison across three or more conditions?

[browningChen]

Hi rita
Thanks for the powerful tools,
I have some question about the comparison across three or more conditions.
For example, here I have different age from 29 to 90, or I have three condition like control, disease and therapy.

[ritamtbsilva]

Hi @browningChen,

Thanks for the kind words!

Regarding comparisons across three or more conditions (e.g. age spanning 29-90, or categorical groups such as control/disease/therapy), the appropriate approach depends on both the module being used and the specific biological or statistical question. Could you clarify what you would like to test or visualise?

In any case, here are some insights that may help:

Scores module
For score visualisation using PlotScores, it is currently not possible to display more than two variables simultaneously. At the moment, one variable can be mapped to the x-axis and one additional variable to colour.

For more complex scenarios (e.g. three categorical variables or continuous + categorical combinations), I recommend using CalculateScores. This function returns a data frame containing per-sample scores together with the associated metadata. This output is directly compatible with ggplot2, allowing you to construct custom visualisations and perform more complex comparisons that you may be interested in.

Enrichment module
Here the strategy depends on how the design matrix is defined for differential expression. markeR supports “simple” comparisons without explicitly specifying a design matrix; in that case, you can use the variables argument to define the comparison of interest. Otherwise, a user-defined model matrix is required via the modelmat argument.

Note on continuous variables (e.g. age):
Age is treated as a continuous (numeric) variable and can be directly incorporated into markeR analyses. Detailed guidance and examples are available in markeR's documentation:

• Enrichment module: https://diseasetranscriptomicslab.github.io/markeR/articles/Article_BenchmarkingMode.html#note-on-continuous-variables
• Scores module (see the end of the same section): https://diseasetranscriptomicslab.github.io/markeR/articles/Article_BenchmarkingMode.html#logmedian-method

If this isn’t enough to address your question, feel free to share a bit more detail about your specific use case and I’ll be happy to help :)

[browningChen]

Hi @browningChen,

Thanks for the kind words!

Regarding comparisons across three or more conditions (e.g. age spanning 29-90, or categorical groups such as control/disease/therapy), the appropriate approach depends on both the module being used and the specific biological or statistical question. Could you clarify what you would like to test or visualise?

In any case, here are some insights that may help:

Scores module For score visualisation using PlotScores, it is currently not possible to display more than two variables simultaneously. At the moment, one variable can be mapped to the x-axis and one additional variable to colour.

For more complex scenarios (e.g. three categorical variables or continuous + categorical combinations), I recommend using CalculateScores. This function returns a data frame containing per-sample scores together with the associated metadata. This output is directly compatible with ggplot2, allowing you to construct custom visualisations and perform more complex comparisons that you may be interested in.

Enrichment module Here the strategy depends on how the design matrix is defined for differential expression. markeR supports “simple” comparisons without explicitly specifying a design matrix; in that case, you can use the variables argument to define the comparison of interest. Otherwise, a user-defined model matrix is required via the modelmat argument.

Note on continuous variables (e.g. age): Age is treated as a continuous (numeric) variable and can be directly incorporated into markeR analyses. Detailed guidance and examples are available in markeR's documentation:

• Enrichment module: https://diseasetranscriptomicslab.github.io/markeR/articles/Article_BenchmarkingMode.html#note-on-continuous-variables
• Scores module (see the end of the same section): https://diseasetranscriptomicslab.github.io/markeR/articles/Article_BenchmarkingMode.html#logmedian-method

If this isn’t enough to address your question, feel free to share a bit more detail about your specific use case and I’ll be happy to help :)

Thanks for reply, which help me a lot

DESeq can product multi-comparison for treat and untreat disease group to health control at the same time

coldata$condition <- relevel(coldata$condition, ref = "Control")
dds <- DESeqDataSetFromMatrix(countData = count,
                              colData = coldata,
                              design= ~condition)
dds <- DESeq(dds, betaPrior = FALSE)
resultsNames(dds)
#[1] "Intercept"                        "condition_treat_vs_WTControl"   "condition_untreat_vs_WTControl"       
results(dds, lfcThreshold = 0,alpha = 0.1,pAdjustMethod = "BH",name="condition_treat_vs_WTControl")

Some time we wanna show the effect of treatment comparing to untreated group as below, we will list then together and using clusterProfiler::compareCluster to enrich together

and heatmap to show the expression similarity of the DEG under those disease-related pathway

[ritamtbsilva]

Hi @browningChen,

Assuming your condition variable is a single factor containing three levels (WTControl, treat, and untreat), and that you want WTControl to be the baseline (reference level), the equivalent setup in markeR would be:

1. Define design matrix

metadata$condition <- factor(metadata$condition, levels=c("WTControl","treat","untreat") 

model_matrix <- model.matrix(~ condition, data = metadata)

colnames(model_matrix)
# "(Intercept)"
# "conditiontreat"
# "conditionuntreat"

In this parameterisation:

• (Intercept) = mean expression in WTControl
• conditiontreat = treat − WTControl
• conditionuntreat = untreat − WTControl

These are what we call contrasts.

2. Run differential expression

DEGs <- calculateDE(
  data     = counts,
  modelmat = model_matrix
)

Access DE tables per contrasts:

DEGs$conditiontreat
DEGs$conditionuntreat

For a direct comparison between treated and untreated:

DEGs <- calculateDE(
  data     = counts,
  modelmat = model_matrix,
  contrasts = c(
    "conditiontreat",
    "conditionuntreat",
    "conditiontreat - conditionuntreat"
  )
)

3. Run GSEA

GSEAresults <- runGSEA(
  DEGList            = DEGs,
  gene_sets          = genesets_example,
  stat               = NULL,
  ContrastCorrection = FALSE
)

This will perform enrichment separately for each coefficient/contrast, returning a named list with GSEA results for each contrast, for all gene sets provided.

4. Visualise results

This will produce what we call lollipop plots (you can find some more examples here) and that would allow to visualise the GSEA results for the contrasts / multi comparisons of interest, in relation to a set of provided gene sets.

plotNESlollipop(
  GSEA_results   = GSEAresults,
  saturation_value = NULL,
  nonsignif_color  = "#F4F4F4",
  signif_color     = "red",
  sig_threshold    = 0.05,
  grid             = FALSE,
  nrow             = NULL,
  ncol             = NULL,
  widthlabels      = 13,
  title            = NULL,
  titlesize        = 12
)

Each panel corresponds to one coefficient/contrast (e.g., conditiontreat or conditionuntreat).

If your condition variable is structured differently (for example, separate disease and treatment variables rather than a single three-level factor), the design matrix would need to be adjusted accordingly.

If anything is still unclear, feel free to share your metadata and the specific comparisons you’re trying to make so we can work through it together :)

[browningChen]

Thank you so much rita @ritamtbsilva !
really appreciated, I will try it : )