I may be doing something very wrong, because there is very little correlation.

Here is how I made the graph:

des <- res_subset_df
bg <- results_genes_subset

des <- des[des$baseMean > 0,]
des <- tibble::rownames_to_column(des, var = 'id')

des <- des[rank(des$pvalue)<5000,]
bg <- bg[rank(bg$pval)<5000,]

des <- des[match(bg$id, des$id, nomatch = 0),]
bg <- bg[match(des$id, bg$id,  nomatch = 0),]

des <- des[order(des$id),]
bg <- bg[order(bg$id),]

plot(rank(des$pvalue), rank(bg$pval), pch = '.',  xlab = "deseq2 rank p",
     ylab = "ballgown rank p",
     main = expression("deseq vs ballgown"))

To generate the DESeq results I'm doing this:

dds <- DESeqDataSetFromMatrix(countData = countData,
            colData = colData, design = ~diagnosis)

#subset to remove unaffected symptomatic

ddssubset <- dds[, colData(dds)$diagnosis == "affected" | colData(dds)$symptoms == "control"]

ddssubset <- DESeq(ddssubset)

res_subset <- results(ddssubset, contrast = c("diagnosis", "affected", "control"))
res_subset <- res_subset[order(res_subset$padj),]
res_subset_df <- data.frame(res_subset)

And for ballgown results, this:

bg = ballgown(dataDir = "ballgown", samplePattern = "P", pData=datainfo, 
              meas = "all")

bg_filt <- exprfilter(bg, cutoff = 10, meas = "cov")

#subset to remove unaffected symptomatic

statssubset = subset(bg_filt, "diagnosis != 'symptomatic'", 
                     genomesubset = FALSE)

results_genes_subset = stattest(statssubset, feature="gene",
                         covariate = "diagnosis", getFC = TRUE, 
                         meas = "FPKM")

Well I’m biased, but there is an statistical advantage to working on the counts with offsets, instead of the normalized expression (FPKM).