Raw counts vs expected counts

Question

[mhyunjunkang]

Hi All,

I'm new to RNAseq, so I'm confused with a couple of terms.

As I understand, raw counts are generated by sequence aligning. Is it right?

And then, expected counts are generated by running RSEM with raw counts, right?

This is to correct sequence reads mapped to more than one spot or alternatively spliced variants. Is this right?

Then, expected counts are not normalized by length of genes. Did I understand correct?

I guess, this is pretty simple things, but I'm still a little confused when someone asks me..-.-v

Please correct, if I have misunderstanding.

Thanks in advance. MH

Answer 1

[Devon Ryan]

Raw counts are what you get when you align to the genome and count how many reads (generally discounting multimappers) align to each gene.

What you refer to as "expected counts" are better called "estimated counts". This involves using expectation-maximization (or similar) to try to apportion multimapping reads in the most likely manner. This can also be used to get isoform-level metrics. Aside from RSEM, salmon and kallisto can also do this.

Estimated counts are typically presented in TPMs and/or estimated counts. TPMs are length normalized, while estimated counts are not.

[mhyunjunkang]

Thank you for your reply and precise explanation. I totally agree with you that "estimated counts" is more proper word, after EM. I think I need to study about salmon and kallisto, which I have no knowledge about. Thanks again. It helped me a lot. MH

[Devon Ryan]

Happy to help! Salmon/kallisto (and RapMap/sailfish/etc.) are about "the best thing since sliced bread" in the RNAseq world :)

[kishorssf91]

Hi Ryan, Could you please tell what should be cutoff value if i want to filter the transcripts based on "expected_count" value?