Visualising insertion distributions with IGoR

[willbradshaw]

Hi there,

I'm trying to access the insertion/deletion/gene-choice/etc probabilities from an inferred IGoR model for visualisation using pygor. I imported the pygor.models.genmodel.GenModel object and ran

g = GenModel(<path_to_parms_file>,<path_to_marginals_file>)

for the final_parms.txt and final_marginals.txt files in the igor/evaluate directory. Sticking with VD-insertions for now as an example, I could find two entries that looked promising:

• g.marginals[0]["vd_ins"] gives me a vector of what look like probabilities, presumably of different numbers of insertions.
• g.get_event("vd_ins", True).realizations gives me a list of Event_realization objects of the same length as the marginal array from the previous point. Each realisation R has an associated integer accessible by R.value (or R.index - the two are always the same).

To get from here to a plot I need some more information about how data is organised in the GenModel object, which I'd be grateful if you could shed some light on:

• Are the probabilities in the marginals array in order (i.e. element 0 is the probability of 0 insertions, element 1 the probability of 1 insertion, and so on)? If not, what is the order based on?
• Are the realization.value integers the number of insertions associated with each realization? What does the order of realizations in the event tell me?
• What's the relationship between the array of probabilities and the list of realizations?

Some of these questions are probably redundant. :-)

Thanks! Will

[qmarcou]

Hi Will, Indeed your three questions can be answered by a single answer:

• The in g.marginals[0]["even_nickname"] is a vector or higher order array of probabilities. Additional dimensions will be present in case the realization of the considered event is conditionally dependent on other events. g.marginals[1]["even_nickname"] will give you the order of the dimensions.
• As for realizations themselves there are three different properties you are referring to, it will be easier to understand for the case of gene choice as for the peculiar choice of insertions the 3 can be redundant:
  • R.name : the realization name, e.g for gene choices the name of the gene
  • R.value : the actual value associated with the realization name, e.g for gene choice the actual gene sequence
  • R.index: the realization index on the marginal array. So if you want the probability of the realization R just access ``g.marginals[0]["even_nickname"][R.index] (note this only works with the 1 dimensional example, otherwise you will have to specify indices for the others dimensions too)

These properties are the same than the ones described for the C++ interface: https://github.com/qmarcou/IGoR#inference-and-evaluation-output

The fact that for insertions the 3 properties are equal is true only if the model has been constructed in a special way defining a range of authorized number of insertions. Bottom line: stick with using R.index even for insertions

Hope this helps