Question: Correct orientation for 5p data

[wmacnair]

Hi @rob-p, very happy to hear that you have some time for simpleaf &co again :D

We have a small question regarding the right orientation for 5' data.

We are pretty sure that the right call looks like:

simpleaf quant --expected-ori rc ...

However, this seems to conflict with the specifications given in chemistries.json:

  "10xv2-5p": {
    "geometry": "__builtin",
    "expected_ori": "fw",
    "plist_name": "dda0309f511ded5d801081a55c66b9a44cab4edbf0e07a9223f539e248d8e090",
    "version": "0.1.0",
    "meta": {
      "cr_filename": "737k-august-2016.txt"
    },
    "remote_url": "https://umd.box.com/shared/static/jbs2wszgbj7k4ic2hass9ts6nhqkwq1p"
  },
...
  "10xv3-5p": {
    "geometry": "__builtin",
    "expected_ori": "fw",
    "plist_name": "0acedbe7ef38360c3731dbda78043b7bacbd75a0bb4e7e749e4fefa70e76a537",
    "version": "0.1.0",
    "meta": {
      "cr_filename": "3M-5pgex-jan-2023.txt.gz"
    },
    "remote_url": "https://umd.box.com/shared/static/cbpv1c4zi6ty81nvcgy3pyta2oj7vea1.txt"
  },

(we got this file via simpleaf chemistry refresh).

Is this maybe an oversight in the chemistries.json document? Or are we missing something?

Thanks!
Will
(also Maruša @marusakod)

[rob-p]

Hi @wmacnair,

I'll cc @DongzeHE on this as well. It's been a little while, but I recall having some substantial discussion about this when we first designed it. And the reasoning has to do with how orientations are encoded upstream in piscem depending on if there is >1 biologically mapable reads. Essentially, it is more common with the 5' tech to actually sequence biological material in part of R1 in addition to the R2 biological sequence. In this case, we have the following conundrum (which future versions of alevin-fry and simpleaf will alleviate with improved information in the RAD file): RAD fragments carry only a single orientation, and so if you have a paired fragment, you are left with the question of which fragment you should record orientation of. In the moment, we decided that it made sense to record the (implied) orientation of the first fragment that might contain biological sequence. So if we have a read pair with r1 and r2 mapping, and with r1 fw and r2 rev, we record f1. If we have a read pair where r1 might map, but only r2 maps (in the rc orientation), we still record fw (the expected orientation of the first read that might contain biological sequence).

This is, of course, wildly confusing, as evidenced by the fact that I am still struggling to recall the logic now. It stems from the fact that the RAD format initially was designed to carry only a single orientation per fragment. We've now generalized it considerably and lifted these limitations, but have not updated the whole stack. So, I'd like to ask @DongzeHE to chime in, as, presumably, he's processed some of this data and can confirm. Likewise, this definitely seems like something that should be documented somewhere, so we should figure out where to document it, and how to do so in as un-confusing a manner as possible.

Best,
Rob