Question on TU filter steps

[aswathyseb]

Hi,

I am new to TT-seq and I came across TT-Filter app. Thank you for the well written code.

I am going through it to understand the steps in the pipeline. Could you help me with the following few questions that I have trouble understanding?

1. After getting the coverages in the bins, do you do any normalization on the bins? I see some papers mention RPK.
2. I see that you have a size factor calculation in the code. How is it calculated and what is it used for?
3. Do you sum the coverages from all samples for the STAN algorithm or STAN is run individually on each sample file?

Thanks,

[shaorray]

Hi,

Thanks for testing this app, it was made for comparison between nascent RNA-seq samples and easy-of-use.

After getting the coverages in the bins, do you do any normalization on the bins? I see some papers mention RPK.

Yes, bin coverage of each sample is subject to size factor normalization (DESeq's median of geometric mean).

I see that you have a size factor calculation in the code. How is it calculated and what is it used for?

RPK represents sequencing depth, while DESeq's size factor is more robust for handling different shapes of read count distribution. With different sequencing methods and treatment conditions, size factor can do the job nearly perfect.

Do you sum the coverages from all samples for the STAN algorithm or STAN is run individually on each sample file?

Read coverage is calculated by Rsamtool from genomic bins. Then STAN is applied for calling 'transcribed' regions, which is the active hidden Markov state.

Best,
Rui

[aswathyseb]

Thank you for the answers. I just want to make sure I understand the last answer.

If there are 4 samples, after getting the read coverages using RSamtools, the 4 coverage files are provided as input to STAN, correct?

Or is it that you agrregate the coverages from all samples and provide a simple file as STAN input?

[shaorray]

If there are 4 samples, after getting the read coverages using RSamtools, the 4 coverage files are provided as input to STAN, correct?

Correct.

Or is it that you agrregate the coverages from all samples and provide a simple file as STAN input?

If you select multiple bam files in the "sample 1" box, all reads in these files will be summed up to a single coverage for STAN's transcribed region calling.

Then, if there are files uploaded in "sample 2/3/4...", size factors will correct the sequencing depth difference of these samples.

In other word, the larger bam file(s), the higher faction of rare transcription events of each sample. Then 'expr' column keeps normalized RPK info of each TU, after comparing reads distribution among uploaded samples.

[aswathyseb]

I am sorry about the repeated question, but I am a bit confused about the last sentence "Then 'expr' column keeps normalized RPK info of each TU, after comparing reads distribution among uploaded samples".

So are you calculating RPK? I thought you adjust using the DESEq2-based size factors.