SQANTI-Reads and General Workflow question

[mpizzagalli777]

Hi all, I have a general question regarding a workflow I have been working on. In order to study alternative splicing and novel transcripts in a series of patient derived cells, I want to generate a high confidence transcriptome. To this end, I have been using SQANTI as my last step. Currently my workflow (per cell line) is as follows: Pychopper on individual LR samples, followed by FLAIR align and correct. After that, I perform FLAIR collapse on the concatenated bed files and pass the gtf output into SQANTI, where I include some SR data as orthogonal support. However, I recently saw SQANTI-reads and was wondering if this approach is incorrect?

Finally, I would like to eventually create a universal transcriptome for these samples. I have four cell lines and triplicates from LR data for each sample. How would I go about generating this combined transcriptome? It seems that using the concatenation and FLAIR collapse approach would allow me to do this but would prevent me from using SQANTI-reads as that requires SQANTI-QC to be run on each individual sample. I would like to create this universal transcriptome in order to ensure that any novel transcripts that appear across cell lines will have a common name between cell lines.

Thanks in advance for any help :)

[carolinamonzo]

Hi @mpizzagalli777,

Your workflow seems great! SQANTI-reads is meant to be a quality control of the actual reads, or to compare long-reads sequencing experiments between samples in order to identify outliers or strange patterns in individual samples when compared to others. It also includes some metrics on the direct quality control of the reads (you can read more about it in it's preprint: sqanti-reads. Since it is a quality control tool, it does not have any impact in your workflow.

If you are interested in comparing the transcriptomes of the samples, you could run the samples through SQANTI-reads, but if the multi-sample comparison is not of your interest, you can go ahead with the workflow you mentioned.

Best wishes,
Carolina.

[andysaiz]

When running SQANTI3 on each individual sample for input into SQANTIreads, which GTF do you use?

I was thinking on running SQANTI-reads and I was a little confused on whether to use the filtered/updated GTF (output from running SQANTI3 with all samples together/aggregate GTF) or the original GTF.

I already ran SQANTI3 (QC, filter, rescue) on the aggregate GTF that goes with all samples. Got an updated GTF of the transcripts retained after filtering/rescue. Do I now use this filtered GTF to run SQANTI3 QC on each individual sample? Or do I use the original GTF from IsoQuant that represents isoforms from all samples?

[carolinamonzo]

Hi Andy,

I'm not super sure I understand your question.

1. If you want to compare raw samples to get a feeling of the differences in sequencing quality between samples of the same or different runs, you can take the original individual GTFs, run each of them through SQANTI3 and then pass them to SQANTI-reads (this is the fast mode you'll find in the wiki).

2. If you want to use SQANTI-reads to do a comparison between samples that you have run through IsoQuant (reconstructed isoforms), you need to start from one GTF per sample, run each sample independently through SQANTI3-QC, (Filter and resque are optional but highly recommended to clean your reconstructed transcriptomes), and then use the resqued GTFs of each individual sample to run them through SQANTI-reads.

SQANTI-reads in reality sort of works like a sqanti-multisample, and is a simple module to directly compare gtfs between samples. It does not start from an aggregated GTF, but from GTFs per sample and a metadata file indicating the particularities of each of them.

I hope this solved your question.

Best,
Carolina.

[andysaiz]

Yes, thank you so much! This is exactly what I was looking for! want to see sample-level QC (SQANTI-reads), but I also want to see how transcripts vary across biological groups (WT, KO1, KO2), so I devised a strategy to prioritize both! What do you think?

[carolinamonzo]

Hi Andy,
It looks great! I think with your workflow you will cover all aspects that you are interested in :)
One tip, in SQANTI-reads, when you make your metadata, you can include a column that will be used to group the samples according to that condition, so for example, every row in the metadata would be a different sample, and then in the grouping column you would call them WT, KO1, KO2, so you will have faceted plots (in the same scale) and get both information on the quality of the samples and the groups.
All best wishes,
Carolina.

[mpizzagalli777]

Thanks so much for you response and help Carolia!