Researcher Alex Sigal has followed the mutations of the coronavirus in an HIV patient over months [1], and he believes that the pattern of mutation in the Omicron variant is similar to what they had observed in that patient. So he favors the HIV scenario for the development of Omicron.
And he speculates that, given this scenario, this would increase the chances of Omicron being milder than other variants, because, as he puts it, a virus that persists for months in an immunodeficient patient is probably not particularly deadly.
It doesn’t quite match my own intuition, since the major damage of a corona infection seems to come from your own immune response, rather than directly from the virus. But my expertise is quite limited, so I trust Sigal’s gut feeling on this question much more than my own. However, I think that “gut feeling” and “speculation” does describe the level of confidence.
[1] https://www.eurekalert.org/news-releases/609570 (edit: corrected link)
(All infos from German magazine SPIEGEL, which is unfortunately paywalled and in German.)
Fun fact: in the field of optimization there are heuristics which are modeled after evolutionary principles. These “evolutionary algorithms” also work with populations, offspring generation through mutation and mating, selective pressure, diversity preservation, and so on.
As a rule of thumb, these algorithms also work better when sexual reproduction is used. For example, a standard theoretical benchmark are monotone functions on bit-strings, where each gene takes only two values zero and one, and flipping a zero into a one gives higher fitness in all situations and environments. This seems like an easy situation to optimize, but asexual algorithm don’t find the optimum (where all genes are one) for exponential time (in the number n of genes) if the mutation rate is large. Algorithms which use mating have no trouble. This comes from Muller’s ratchet. [1]
More remarkably, even if the mutation rate is arbitrarily small, asexual algorithms don’t find the optimum for exponential time if they use populations, where in each round they produce a new offspring and prune the least fit individual. Again, algorithms with mating don’t have these problems. This also comes from a version of Muller’s ratchet, but on population level. Essentially, if there is a beneficial mutation then other mutations have time to accumulate until the beneficial mutation has taken over the whole population, and this takes long enough to accumulate very many bad mutations, even for extremely low mutation rates. [2]
[1] https://link.springer.com/chapter/10.1007/978-3-319-99259-4_1
[2] https://www.sciencedirect.com/science/article/abs/pii/S030439752100178X
(free preprints available on arxiv)