[FRIAM] whackadoodles go mainstream!
Marcus Daniels
marcus at snoutfarm.com
Tue Apr 21 22:14:04 EDT 2020
Eric writes:
< The project of looking for minima at the level of protein-protein interactions is the sort of thing that molecular-dynamics modelers can do, and that I think is part of their ordinary work already. >
In principle there could be different energy minima depending on host genetics. So another way is to forget about structure and just look at their functional consequences. That is, a viral gene VG1 makes a protein VP1 and some human gene HG1 makes another protein HP1. Sure it is actually VP1 and HP1 that interact, but the physics of that interaction could require an exascale Folding at Home calculation to sort out. Or one can just say that VP1 and HP1 have some functional interactions F that can be studied without really understanding the physical form of VP1 and HP1. VP1 and HP1 are merely intermediate variables for getting at F, one can simplify to Fg(VG1,HG1) instead of F(VP1,HP1) A problem with this plan is that not just VG1 and HG1 will do, one needs Fg evaluated on thousands of arguments (VG1/HG2, VG2/HG1, .. VGN/HGN) and with replicates to ensure Fg is not really reporting the action of noise or unseen variables. Molecular dynamics calculations could give a deeper or more explanatory sense of what those classes of viable combinations might be and why. But if the goal is to tune a vaccine to cover certain HG values, you may not really care why the biophysics works.
Marcus
From: Friam <friam-bounces at redfish.com> on behalf of David Eric Smith <desmith at santafe.edu>
Reply-To: The Friday Morning Applied Complexity Coffee Group <friam at redfish.com>
Date: Tuesday, April 21, 2020 at 6:23 PM
To: The Friday Morning Applied Complexity Coffee Group <friam at redfish.com>
Subject: Re: [FRIAM] whackadoodles go mainstream!
Both of these seem to be ideal and also interesting projects. Some parts harder than others. I can’t speak with expertise, but can provide small details about some parts where I think I understand them.
On Apr 22, 2020, at 1:34 AM, Prof David West <profwest at fastmail.fm<mailto:profwest at fastmail.fm>> wrote:
Two naive questions.
1) Diseases mutate/evolve immunity or resistance to antibiotics, or so I believe i have been told. And I think this discussion notes that Covid evolves in response to pressures originating in the host. And I have read articles talking about genetic difference among hosts as partially responsible for the variance in severity of the disease. Is there a way to build some kind of "co-evolutionary" model — virus adapting to host / host adapting to virus — that would predict a stable minima?
The project of looking for minima at the level of protein-protein interactions is the sort of thing that molecular-dynamics modelers can do, and that I think is part of their ordinary work already. It was implicit in the Nature letter to the editor, in the discussion about interpreting the six mutations. The challenges that set the de facto limits on that work are 1) there is some computational cost in the models themselves, but moreso, any given relaxation model normally uses a structural template and can be biased by what protein structures you know about, and; 2) when you try to get past templates, the dimensionality of the computational search becomes very high, and not something that we can do with even a modestly-costly molecular model needing to be applied to each test case. The problem is that you have to search for the gross features of the fold, and not just optimize the little local relaxations. The more actual sequence variation you can get (which is to Marcus’s point about more comprehensive databases), the better, although what you really need is for sequences to be accompanied by structures if there is significant structural change, and structures are the slowest and most costly of all. To the extent that you have data support of that kind, you can hope to be less severely biased, because nature will have done some of the searching for you. But prospective modeling of counterfactuals remains a problem of search, and so is hard to do with good coverage or confidence.
or is this whatis meant by "herd immunity?”
I think I can say with some safety that “herd immunity” is a much simpler and more generic term that doesn’t have to do with any modeling or design. It is more of a summary-statistic in character. In a way that is completely phenomenological, epidemiologists can try to characterize the transmissibility of a disease in this lumped parameter R0, which aggregates how many downstream new infections result from any extant infection through its full course. The initial R0 that you would get for a novel virus like this one is simplified by the assumption (probably approximately correct) that nobody has pre-existing immunity, though the empirical estimator still has a context of people’s degree of contact, kinds of contact (passing in the park versus sitting across a dinner table and talking versus being at a bar and shouting while somewhat drunk), and so the data that go into such an estimate will vary from urban to rural, private to public transport, etc. I think (and here I don’t know details) that the people who do this try to get an estimate that averages somehow over all those circumstances when they broadcast a number. In better models, one would try to get a distribution of values conditioned on variables you think you can estimate.
In SIR or SEIR models, the actual number of new infections created by one extant infection, even if the social structure is held fixed, will change as the number who are immune by recovery increases. I don’t know whether the convention is to absorb that into an “effective R0”, or to continue to use R0 for the naive population, and then say that the actual transmissions have growth rate R0*(fraction susceptible), with the time dependence only in the second factor. (I believe the convention is the latter.). Assuming the second convention, “herd immunity” just refers to the condition in which the fraction susceptible has decreased to the point where R0*(fraction susceptible) < 1, so the long-term trajectory of the disease is extinction if there aren’t refugia for it.
Of course, since whatever one means by R0 can change a lot if there are behavior changes (not sharing trains, wearing masks, separation, etc.), by decreasing the R0 one can lower the fraction recovered needed to get R0*(fraction susceptible) < 1. Unfortunately, for any interventions short of vaccination, you still need fractions of the population to be recovered that are much larger than we want to subject either people or the medical system to, on a timeframe as short as a year. So pretty severe behavior modification seems to be needed throughout the pre-vaccine period to keep a disease endemic, but not epidemic.
2) Supposedly (two articles, one in JAMA), 80% of COVID transmissions occur in homes, including nursing homes, and 34% occur in public transit, and near zero transmissions outside. (I know the numbers do not equal 100, but I think that is an artifact of being summarized.) Again supposedly, the only outbreaks from all the panned spring break activity were attributed to the public transport (airplanes) transmission, not to beach/party activity. Would it not be possible determine some kind of "person-density metric" — humans per cubic volume — that could be used as a basis for reopening stores, restaurants, etc. etc.
This seems like an eminently buildable model that it would be great to have. The key will be when we can get data-based estimates for it that we know how to contextualize properly in the model and that we think are reliable in the measurement methods. I am very glad to see your cites that this is striating to become available. I would like to see this be a significant area of work quite broadly through the next year.
I have also wondered about more complicated strategies, like adding immunoglobulin donations to the social model and the medical treatment regime. If _all_ of the cases were only miserable colds, we probably wouldn’t worry too much about allowing it to sweep the population for a couple of miserable years. If there were a way, with serological testing, to know who has antibodies, and we could do a regular program of plasma donations on massive scale, ramping up with each new recovered person, and if those immune donations together with antivirals or whatever drugs we have, could be used to catch and mitigate some larger fraction of the bad cases, maybe a socially conscientious but not very aggressive strategy like Sweden seems to be pursuing could be made rational. (What they are doing right now does not look very good to me, in the sense that the have about 5x as many people dead per thousand as Norway, and 2.5x as many as Denmark, for no reason I can understand).
The dynamics of that model in principle seem appealing. Initially you can only allow a small fraction of illness, because you don’t have much serum to catch its bad cases. But you can allow a little more infection than you could with no serum, and because you get more recoveries, you can get more serum, which allows you to admit a larger fraction of infections etc. So the difference should be a change in the time constant of the exponential growth toward herd immunity, in proportion to what part of the bad-case tail you can mitigate. Getting anything like this past the stage of emergency management, and into some kind of a program, is probably vastly complicated, though. You would have to get good medical prediction for what you can and can’t do (safety and efficacy) with antibody transfer, which could take as long as the vaccine development. If anyplace were to start it systematically, I would guess it would be Germany, given how they are setting up at this time.
All best,
Eric
davew
On Tue, Apr 21, 2020, at 9:59 AM, Marcus Daniels wrote:
Glen writes:
< I can see that the diversity seems to peak as the infections ramp up in the US and EU. And there's only a handful of spots where the diversity shoots up, only to level off later. >
Well, the color-coded phylogeny is looking at a tiny sampling of the infected world. I wouldn't draw any conclusion from that other than that viral drift is possible. It hasn't even been tabulated systematically against any phenotype besides where the sample was collected. Looking at the GenBank data, there are even variants within the same patient.
Hopefully there will be engagement on compilation of anonymized clinical data outside of academia (esp. from hospitals and the major diagnostic labs). Then we can stop talking about what we "expect" and look at what actually happens!
This morning I was told of this: https://covid19researchdatabase.org/
Marcus
________________________________
From: Friam <friam-bounces at redfish.com<mailto:friam-bounces at redfish.com>> on behalf of uǝlƃ ☣ <gepropella at gmail.com<mailto:gepropella at gmail.com>>
Sent: Tuesday, April 21, 2020 9:36 AM
To: FriAM <friam at redfish.com<mailto:friam at redfish.com>>
Subject: Re: [FRIAM] whackadoodles go mainstream!
Well, from the animation Marcus posted, even I can see that the diversity seems to peak as the infections ramp up in the US and EU. And there's only a handful of spots where the diversity shoots up, only to level off later.
The Gorman article was very helpful. I suppose to really grok the inter-species vs. intra-species patterns enough to build an expectation, you'd have to work at a bat lab. By selecting h3n2 in the left panel, it seems the flu maintains a higher diversity across the board. So my 1st attempt at an expectation would be that a) jumping species, b) jumping species from a tolerant host like a bat, and c) the novelty of the strain should lead to a kind of pressure to get a better foothold in the hosts. Then as the host and parasite relax into one another, any obvious pressure would fade and it would look more random.
Regardless of the design question, it seems like a virus that is *likely* to jump species in the first place would see less pressure to fit in immediately and would more quickly begin looking random.
Anyway, thanks for your tolerance.
On 4/20/20 5:15 PM, David Eric Smith wrote:
> Not my area of expertise, though for about the past year I have spent some time learning about proteins, so I am not as completely lost as I would have been before that.
>
> The number of mechanisms that jointly get termed “evolution” in a complicated, mosaic, multi-host virus like this makes inference a many-dimensional problem.
>
> Putting aside Marcus’s information on HLA variability, because I have not yet made the time to read it (though I hope to), just looking at the viral spike protein genes in the Nature letter to the editor, there seem to be at least two qualitatively different things going on. The six mutated positions in the receptor binding domain are interpreted by the letter’s authors as plausible convergent mutations. Since a lot of protein evolution seems to get locked in by structural or self-assembly constraints at many places, the number of labile positions on shorter timescales is some smaller number than the 20^211 that Marcus notes as an upper bound for brute force search (but again that is in the context of leukocytes, whereas these six mutations involve binding affinity to ACE2). But when we see two solutions that seem to be in completely different basins of attraction in humans, as SARS-CoV and SARS-CoV-2 appear to be, there looks to be a big valley of non-viability between
> them, with very low probability to have all mutations occur conjointly to cross it. I think people believe that between bats, cats et al., and now I guess pangolins (and I think I saw something about snakes from one letter that went around from an early researcher), there is an enormous reservoir of different strains, with a quite large diversity of ACE-type host proteins. So we would be looking for island-hopping routes that make the SARS-COV and SARS-CoV-2 solutions mutually intelligible. So the analysis of protein change mechanisms gets put into a larger context of ecological analysis of species contacts, which is probably as badly under-sampled as the viral diversity is.
>
> The foregoing is independent of this “furin cleavage site”, and some special proline in a turn that facilitates attachment of surface amino-sugar chains. This article suggests that such features are under selection from either or both of infectivity based on how the proteins are translated, and immunosurveillance. Both of those will involve translational and immune proteins that vary significantly from one host to another. (I assume this is where HLA variability becomes central to this story.). There was an article in NYT a week or two ago by James Gorman on why bats seem to be a reservoir for “so many viruses”
> https://www.nytimes.com/2020/01/28/science/bats-coronavirus-Wuhan.html
> (I don’t actually know whether they carry uncommonly more, per bat species, than other groups do per-species, since we often don’t know about things until they interest us enough to do a thorough survey, but if Gorman is writing with knowledge, that claim might be okay.). Certainly, there are an awfully large number of bats species, considering that they are just one little branch of insectivores. If there is something different about bat immune systems, which pairs in an important way with the genome evolutionary position of coronaviruses, that would seem to be the first place to look for immune selection, as context for later asking about what is different in people. The interesting thing is that these cleavage sites have not been found in other beta-lineage coronaviruses. (He doesn’t say where they have been found, though that shouldn’t be a hard dig to get to.). The suggestion is that the CoV-2 strain is a mosaic that is mostly beta-coronavirus with something else.
> The question is then, what else looks closest, where do the two occur together; what are the mechanisms for combining them. Now we are in an epidemiology/ecology question.
>
> Inferring a chain of origin with all these factors in play looks challenging to me.
>
> Eric
>
>
>> On Apr 21, 2020, at 8:35 AM, uǝlƃ ☣ <gepropella at gmail.com<mailto:gepropella at gmail.com> <mailto:gepropella at gmail.com>> wrote:
>>
>> What *would* you people who can read all this stuff *expect* to happen?
--
☣ uǝlƃ
.-. .- -. -.. --- -- -..-. -.. --- - ... -..-. .- -. -.. -..-. -.. .- ... .... . ...
FRIAM Applied Complexity Group listserv
Zoom Fridays 9:30a-12p Mtn GMT-6 bit.ly/virtualfriam<http://bit.ly/virtualfriam>
unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/
.-. .- -. -.. --- -- -..-. -.. --- - ... -..-. .- -. -.. -..-. -.. .- ... .... . ...
FRIAM Applied Complexity Group listserv
Zoom Fridays 9:30a-12p Mtn GMT-6 bit.ly/virtualfriam<http://bit.ly/virtualfriam>
unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/
.-. .- -. -.. --- -- -..-. -.. --- - ... -..-. .- -. -.. -..-. -.. .- ... .... . ...
FRIAM Applied Complexity Group listserv
Zoom Fridays 9:30a-12p Mtn GMT-6 bit.ly/virtualfriam<http://bit.ly/virtualfriam>
unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20200422/b792b612/attachment-0001.html>
More information about the Friam
mailing list