Components in TraversalBlock

[rrfreimuth]

Can someone point me to a use case that demonstrates the need for Allele and CPB to be components in TraversalBlock, please? It makes sense in principle but I cannot think of when I might use those types in the same construct as Adjacency.

[larrybabb]

My understanding is that @Mrinal-Thomas-Epic is working on identifying an example or two and posting them here when he can.

[Mrinal-Thomas-Epic]

Allowing Allele and CPB to be components in a traversal block allows data producers to specify which copy of a duplicated region smaller mutations occur on. For example, among the non-identical gene duplications listed in the PharmVar CYP2D6 structural variants, you could get a different function or activity score depending on which alleles were together on a certain copy of the gene.

@d-cameron will likely be able to give better examples of how experiments or variant calling algorithms could produce outputs that we would represent with a combination of Adjacencies, Alleles and/or CPB as components in TraversalBlocks.

[Mrinal-Thomas-Epic]

Although looking at it now, even though its useful to have that type of phasing for Alleles and CPBs, I don't know if you would ever need to associate an orientation with Alleles or CPBs. It may make more sense to move Allele and CPB up one-level to DerivativeMolecule and only keep Adjacency within the TraversalBlock.

The initial use case for TraversalBlock was just so that we could normalize the representation of adjacencies, while still being able to represent specific orientations (and combinations of orientations) in the context of a derivative molecule. For example, these two adjacencies are equivalent in isolation, but may have different meanings in the context of a derivative molecule:

{
    "adj_1": {
        "type": "Adjacency",
        "adjoinedSequences": [
            {
                "type": "SequenceLocation",
                "sequenceReference": "NC_000001.10", // IRI reference
                "end": 123
            },
            {
                "type": "SequenceLocation",
                "sequenceReference": "NC_000002.11",
                "start": 456
            }
        ],
        "linker": "CCGT"
    },
    "adj_2": {
        "type": "Adjacency",
        "adjoinedSequences": [
            {
                "type": "SequenceLocation",
                "sequenceReference": "NC_000002.11",
                "start": 456
            },
            {
                "type": "SequenceLocation",
                "sequenceReference": "NC_000001.10",
                "end": 123
            }
        ],
        "linker": "ACGG"
    }
}

AFAIK, this concern doesn't apply to Alleles, CPB, or Terminus.

[Mrinal-Thomas-Epic]

I just added this toy example to my MR, which may be helpful in thinking about the need for explicit orientation.

[larrybabb]

@ahwagner I need some assistance from you following the intention here.

If we determine that the ordered components of the DerivativeMolecule do not need to always be encapsulated in a TraversalBlock because we can infer that their orientation is forward (which I think is what we are saying above). Then I am confused why we wouldn't only use TraversalBlock for reverse oriented sequence spans at which point we could simplly call it ReversalBlock? Otherwise you can have 2 ways to specify the forward sequence components.

Also, I'm not sure why we would NOT move Adjacency to an allowed component of DerivativeMolecule along with the other types.

Any clarification on where we are heading with this would be helpful.

[Mrinal-Thomas-Epic]

Can you give an example of what you mean by 2 ways to specify forward sequence components?

I'm not opposed to the combined change of switching TraversalBlock to ReversalBlock and allowing Adjacency directly on DerivativeMolecule. The benefit is the proposed model captures the same semantics, but is slightly more concise. However, it could also be slightly more work (but not much) for implementations to handle the logic of either interacting with an Adjacency or ReversalBlock on a DerivativeMolecule (rather than being guaranteed to always interact with a TraversalBlock containing an Adjacency).

[larrybabb]

Can you give an example of what you mean by 2 ways to specify forward sequence components?

@Mrinal-Thomas-Epic wouldn't it be possible to put the same elements in a TraversalBlock as you can in a DerivativeMolecule (except for Adjacency)? If so then couldn't you express the same Allele, CPB and Terminus combination in a DM and wouldn't that be the same as having a DM with a single TraversalBlock that had the same combo but with a forward orientation? (Maybe I'm missing a key component here)

[Mrinal-Thomas-Epic]

In the old/current model:
I interpreted the definition as only allowing a single molecular variation per TraversalBlock. If you had a combination of Allele, CPB and Terminus each one would get their own TraversalBlock.

In the model proposed in #554:
A combination of Allele, CPB and Terminus can only be specified directly on the DerivativeMolecule (with orientation inferred). Instances of those three classes can no longer be components of TraversalBlock.