AlanRuttenberg: Is this supposed to model something like "Cells were stimulated by Carbachol after which an increase in the quantity of phosphorylated PKD1 was observered" ?
Gary: Basically, though 'stimulated' is not well defined - more like 'Carbachol was added to cells, which led to an increase in PKD1 phosphorylation (through an unknown mechanism). Also, gf109203x can inhibit the activation of phosphorylation by Carbachol'
Alan: So the pathway is essentially a reification of whatever happened in between. Seems like a heavy hammer for DX, particularly if there isn't anything else known about this pathway. The issue of representing the unknown in between should probably be addressed with more care. For example there are kinds of interaction that are prone to be called an interaction type but have proven to be indirect over time, for example degradation is often the indirect version of proteolysis. A full proposal for indirect control would make room for handling this case as well.
For the short term it might be cleaner to use a different property, for instance we could have PHYSICAL-CONTROLLER to mean things like enzyme and kinase controllers which exert their influence through a direct physical interation and CONTROLLER, a superproperty of PHYSICAL-CONTROLLER which can be indirect, or indeterminate. Or if there is resistence to changing the current usage of CONTROLLER, then make a superproperty CONTROLLER-POSSIBLY-INDIRECT and use that for indirect control. This would be unambiguous and leave the complication of empty pathways for the future.
Gary: Interesting idea, however, there is usually a need to name the process, such as "degradation" in your example and it is useful to be able to attach GO biological process or INOH process terms, if known. Not sure if you can call proteolysis direct in the cell (maybe in a test tube, if there was a single protease present). Mostly, it is a process e.g. proteasome degradation
Alan: When I say proteolysis, I am referring to the physical interactions which lead to a peptide being cleaved. In this case the proteasome would be "enzyme" in a control reaction. If you want to talk about the larger process of proteasomal degradation, this would be a pathway, in which there were other steps, interactions such as ubiquitin ligation. The mistake would be to put, in the same way as a physical interaction, a finding that says that e.g. ubiquitin controls degradation. This statement is a short hand for a more involved process. If we allow such kinds of "interactions" then we need, at a minimum, to call them something different, and eventually make provision for relating them to the more explicit pathway which they are short hand for. This requirement for distinguishing direct versus indirect interactions is a long standing requirement, and is important, for example, for creating the inputs to the Activity Center Algorithm
I have a similar criticism of the gene regulation proposal, which, as far as i can tell, doesn't adequately distinguish between the sort of experiment where one treat a system in some way and later observes up-regulation of protein or mrna, which for all we know could have resulted from a cascade of events - a pathway - from more direct regulation like binding of transcription factors, or splicing.
Gary: pathway databases need shorthand abstractions to represent common cases - this and gene regulation are examples - they are interactions. The direct/indirect issue is the real issue, however, pathway databases stay away from this issue because it is subjective (depends on the level of detail you choose to represent). The user should decide what they want to call direct vs. indirect for their use case, just like activity centers has a number of rules for this. Other people may have different rules.