Is RSEM required for RNA-seq data analysis using STAR and edgeR?

Question

[Apprentice]

I have fastq files (n=40) obtained by paired-end unstranded RNA-seq. I would like to analyze these files using STAR for mapping and perform the differential expression analysis between the two groups by edgeR. I plan to apply TPM for normalized count. I am wondering if RSEM is currently necessary since STAR can generate raw count data per gene. I would be grateful if you could enlighten me.

[Ming Tommy Tang]

DESeq2 and EdgeR uses raw counts as input.

Answer 1

[Gordon Smyth]

A complete example analysis is given here (using Rsubread instead of STAR): https://bioconductor.org/packages/release/workflows/vignettes/RnaSeqGeneEdgeRQL/inst/doc/edgeRQL.html

[swbarnes2]

RSEM would also be a smarter algorithm than subread.

[Apprentice]

Thank you for your response. I can use the counts calculated by STAR, please give me a rationale why using the Rsubreade algorithm is better than using the counts generated by STAR.

Answer 2

[swbarnes2]

EdgeR wants raw counts, not TPM. You can use the counts STAR calculates, but RSEM's algorithm is better.

[Apprentice]

Thank you for your response. You are right that I can use the counts calculated by STAR, please give me a rationale why using the RSEM algorithm is better than using the counts generated by STAR. Is it reported in papers or other sources that the RSEM generated counts are more accurate than the STAR counts?

[tbayer]

You could just read the RSEM paper to find out how it processes the counts to make them better! It involves dealing with multi-mapping reads.

[Apprentice]

Thank you for your reply. I'll read the RSEM paper.