Weird correlation between RIP-seq replicates

Question

[Hiba]

Hello,

I have a RIP-seq experiment with 3 replicates per condition. When following the classic RNA-seq analysis pipeline and after quantification with Salmon I've done a transformation and compared the correlation between replicates. This is the code :

sample_info_12DD <- DataFrame(
                    condition= c('12DD','12DD','12DD',
                                'input','input','input'),
                    row.names = c("12DD_rep1",  "12DD_rep2", "12DD_rep3", 
                                 "input_rep1", "input_rep2", "input_rep3"))

dds_12DD <- DESeqDataSetFromTximport(txi_12DD,colData = sample_info_12DD,design = ~ condition)



#here we remove genes with 0 reads in all samples
keep_genes <- rowSums(counts(dds_12DD)) > 0 #TRUE if the gene has more that 0 reads, if not : FALSE
dds_12DD <- dds_12DD[ keep_genes, ]
#cpm filtering
keep <- rowSums(cpm(counts(dds_12DD)) > 1) >= 3
dds_12DD<- dds_12DD[keep , ]

dds_12DD <- DESeq(dds_12DD)

#rlog transformation
rld <- rlog(dds_12DD)
#vst transformation
vst<- vst(dds_12DD)

df <- bind_rows(as.data.frame(log2(counts(dds_12DD, normalized=TRUE)+1)) %>%
    mutate(transformation = "log2(x + 1)"), as.data.frame(assay(vst)) %>% mutate(transformation = "vst"), as.data.frame(assay(rld)) %>% mutate(transformation = "rlog"))


lvls <- c("log2(x + 1)", "vst", "rlog")
df$transformation <- factor(df$transformation, levels=lvls)

The results :

PS : when tryin the 3rd replicate and also other conditions, it has the same correlation profil

My question is : can I pursue Differential gene analysis knowing that I have a weird correlation between replicates ?

[DGTool]

As i.sudbery said below, to me nothing looks too odd either. Would you be able to specify roughly what you mean by weird? That might help narrow down the answer to something more specific (and also help other people if they encounter similar looking plots and have the same question)

Answer 1

[i.sudbery]

I don't think there is anything particularly out of the ordinary here. The log(x+1) and the VST look exactly as I'd expect them. The rlog is a bit werid, but thats probably just a functino of the the rlog transformation.