Why can’t we have full-length transcripts with droplet-based technologies ?

Question

[sagane.joye]

In Smart-seq3 technology, full-length cDNA molecules are pre-amplified by PCR and then cut by Tn5 tagmentation in near-random locations, resulting in many molecules with the same 5' UMI end but different 3' ends. Using paired-end sequencing, it results in 3′ end sequences that span different parts of the initial cDNA molecule (linked to a specific molecule based on the 5′ UMI sequence) and enable in silico reconstruction of the initial RNA molecules.

So my question is why can’t we, or at least no one did, use the same method for droplet-based technologies to enable full-length transcripts reconstruction?

Answer 1

[i.sudbery]

In smart seq-3 you get two sorts of reads. You get end tagging reads, which have UMIs, and "internal" reads, that don't have a UMI. The reads that cover the full length of the transcript are not marked with UMIs as best I can tell from the smart-seq3 paper.

Therefore, we obtained strand-specific 5′ UMI-containing reads and internal reads from either strand spanning the full transcript without UMIs in the same sequencing reaction

In theory, you could probably recover these "internal" reads in a droplet protocol I think.

The next step in Smart-seq3 is make tranditional sequencing libraries, which adds a library barcode to the reads in each well. Here is where you couldn't do that with droplet - the library barcodes are added at the same time as the UMIs on bead-linked oligodT primers. Thus only bead linked fragments get cell barcodes. There would be no way to get a cell barcode on the the "internal" reads that were now detached from the beads while guarenteeing that each cell would get a unique barcdoe.

[sagane.joye]

Thank you very much for your answer ! As I understood the smart-seq3 paper they are able to reconstruct the molecule only using UMIs reads (section Molecule reconstruction and transcript isoform assignment in methods), as their 3'end are spanning different regions of the molecule. Then they just uses "internal" reads to filter compatible isoforms of the reconstructed molecules.

But yes I get the point of not be able to tag internal reads in droplets ! But are they really necessary if we have enough molecules to cover all 3'end regions paired with 5' UMIs end?

Answer 2

[jared.andrews07]

I mean, you kind of can, with some homebrew, see here or here for ISO-seq examples using the 10X system.

The output is not really usable for any sort of expression analysis, but coupled with a typical 10X library, you hypothetically get a typical 10X dataset with much more accurate mapping.