DGE with PEER normalized data

Question

[Prasad]

I am working with time series RNA-seq data(10 samples, each with 3 biological replicates). I performed sample clustering using hclust but the replicate wise clusters were not proper, so I was asked to do further normalization of FPKM values using PEER R package . Now I want to calculate differential expression using this normalized data but could not find any appropriate packages. This is how my normalized data looks like.

Transcript_ID   A   B   C
Contig10    6.50308322747151E-007   -3.90183130249966E-009  2.78059042457812E-009
Contig9999  7.53524395658012E-011   8.57470422488404E-006   -1.14360755105736E-005
Contig9995  0.000002363 0.0001350445    -4.99606248922646E-005

As per the suggestions I am trying out voom in limma not so sure about it so please suggest some packages which can take normalized data for getting DEG's.

Answer 1

[ATpoint]

To give this an answer: You should always correct for library size AND composition. FPKM (so naive per-million) scaling is typically not recommended. Use established procedures such as RLE from DESeq2 or TMM from edgeR. Then use these normalized counts, transform to log2 and check by MDS or PCA plots for signs of batch effects and explore with MA-plots if the normalization performed well. If necessary try and correct for batch effects prior to clustering.