Interesting and probably unwanted behaviour for strain_interaction_to_cross_immunity

[kokbent]

The utils function to convert strain interaction to cross immunity can produce some counterintuitive cross immunity matrix.

import numpy as np
from utils import strain_interaction_to_cross_immunity

interact = np.array([[0.6, 1.0], [0.3, 1.0]])

strain_interaction_to_cross_immunity(2, interact)
# Array([[0. , 0.6, 1. , 0.6],
#        [0. , 0.3, 1. , 1. ]], dtype=float32)

So in this case, if we say that the "interaction" for:

1. challenging strain 1 to a strain 1 infected is 0.6, and
2. challenging strain 1 to strain 2 infected is 1.0

A person who is infected by strain 2 only has cross immunity of 1.0 towards challenging strain 1, but if they are infected by both strain 1 & 2, their cross immunity drops to 0.6 towards challenging strain 1.

[SamuelBrand1]

Two things:
1.

interact = np.array([[0.6, 1.0], [0.3, 1.0]])

seems a bit weird? Is that saying getting infected with strain 1 gives more protection against strain 2?

https://github.com/cdcent/cfa-scenarios-model/blob/12a76646b61fdcdd4b75b038e80cbe4062a3b042/mechanistic_model/utils.py#L906

But I don't think you have "most recent strain" as a concept in the history?

[SamuelBrand1]

Maybe easiest sensible solution is to use the maximum immunity over pre-infected strains.

[arik-shurygin]

Your strain interactions matrix is a bit confusing here though.
The strain interactions matrix you provide looks like this:
[0.6, 1.0]
[0.3, 1.0].

In every case thus far we have always had a 1.0 diagonal across the strain interactions matrix. what this 0.6 says here is that strain 0 vs strain 0 is not a perfect match which does not exactly make sense to me.

[arik-shurygin]

As Ben pointed out in verbal discussion, the 1.0 diagonal is usually enforced, however during strain definition collapse (between epochs when we combine strains into the ancestral one) this can occur. This should be the primary focus of this GH issue, either always enforcing a 1.0 diagonal and patching the bug with strain combination, or allowing the user to pass nonsense if they so wish.

[arik-shurygin]

Had a discussion with Tom, he is not convinced that this behavior is entirely unfounded. If you combine strain definitions, they are still different strains in reality, so there may not be perfect protection between them (diagonal = 1). For example, if we combine BA2/4/5 all together, a challenge from this combined strain is not necessarily met with perfect protection. An individual may have been infected with BA2, is now being met with BA5, but these are now marked as the same strain... Tabling this issue for a future discussion but noting the points for now.

[arik-shurygin]

labeling wontfix for now since this is something that should be on our radar but is not actively going to be worked on