Comparing Sample replicates with PtR (Trinity). Are they normalized?

Question

[nidhi.vijayan13]

So I have 9 replicates for one sample. When I applied PtR --compare_replicates and --log2 --CPM --prin_comp 3 on them, 6 clustered together and 3 clustered differently. I got similar differentiation using --sample_cor_matrix. However, these samples are not normalized right? The 3 samples that cluster apart have twice the number of sequencing reads compared to the others. If they are not normalized, how would I normalize and then run PtR?

[andres.firrincieli]

your count data were simply 'normalized' by sequencing depth (CPM, count per million) and transfromed in log2 scale log2(CPM+1). That's it.

This is the chunck of code in PtR that transform your count data in CPM:

if ($CPM) {

        $Rscript .= "cs = colSums(data)\n";
        $Rscript .= "data = t( t(data)/cs) * 1e6;\n";
        $output_prefix .= ".CPM";

    }

And then convert the CPM values in log2 scale:

 if ($LOG2) {
        $Rscript .= "data = log2(data+1)\n";
        $output_prefix .= ".log2";
        $Rscript .= "write.table(data, file=\"$output_prefix.dat\", quote=F, sep=\"\t\")\n" if $write_intermediate_data_tables_flag;
    }

In your case, see how samples clusters, I would use a different normalization approach (cross-sample normalization), just like those implemented in edgeR (TMM-normalization) and DESeq2 (Variance stabilizing transformation (vst) or rlog)