import: experimental designs with more than one covariate

[lgeistlinger]

@lgeistlinger , thank-you for the update! (esp. import).

# I used apeglm for the res:
res <- lfcShrink(dds, coef="FM_1_vs_0", type="apeglm")
se <- import(dds, res, from = "DESeq2")
Error in .importFromDESeq2(obj, res) : 
  all(rnames %in% colnames(res)) is not TRUE
colnames(res)
[1] "baseMean"       "log2FoldChange" "lfcSE"         
[4] "pvalue"         "padj"  

# repeated without "apeglm"
res <- results(dds)
se <- import(dds, res, from = "DESeq2")
Error in .importFromDESeq2(obj, res) : 
Supported experimental designs include binary group comparisons 
with an optional blocking variable for paired samples / sample batches

design(dds)
~ age + batch + condition

I have 4 questions for you:

1. Is LFC shrinkage needed / recommended with EnrichmentBrowser?

2. I have to use age (scaled, continuous) and batch (binary) as covariates to analyze for my condition of interest (binary)? How can I use this information with EnrichmentBrowser to avoid any false positives?

3. To circumvent covariate problem, I was thinking of using ranked gene list with pi scores res$pi <- res$log2FoldChange*(-log(res$pvalue)), can I do this for downstream topology-based methods in EnrichmentBrowser, if yes, how?

4. Could you suggest a better way to rank / score genes? I am not sure how should the tie broken in case of non-unique scores.

Originally posted by @vivek-verma202 in #23 (comment)