Supercomputer analysis of Covid-19 leads to new theory
elemental.medium.com
elemental.medium.com
One thing that I think strongly suggests that this hypothesis is wrong is that there is no strong relation between ACE-inhibitors and Covid mortality. Indeed, most of the studies that I've seen suggest that ACE-inhibitors have a somewhat protective effect whereas ARBs actually seem to have a minor detrimental effect [1]. So for the article to claim that covid behaves pharmacologically like ACE-inhibitors seems wrong at face value.
Perhaps you can clear up something for me. We've all heard how obesity and hypertension are risk factors for covid morbidity. But I could never get a clarification regarding treated vs. untreated hypertension.
AFAIK, many obese people take ACE inhibitors to treat hypertension. If we divide obese people into three groups: (a) untreated, (b) treated with ACE inhibitors, and (c) treated with other medications, how do their covid moribity rates compare?
How does NPI/mask effectiveness impact the study moving regions?
By the time you design a study, recruit a pool and wait for some of them to get covid, not enough get it for the study to have enough statistical power.
This article does a great job outlining the current state of the knowledge on the subject of Covid/hypertension as well as some clinical trials that should be posting results early next year [1].
I haven't seen any studies that try and tease apart all of the complex relationships amongst various comorbidities, but I think we have seen pretty conclusively that obesity is a very significant risk factor.
[1]https://www.acc.org/latest-in-cardiology/articles/2020/07/06...
(I know this because I have the PubPeer extension installed which puts a big red warning by it.)
> ACE2 counters the activity of the related angiotensin-converting enzyme (ACE) by reducing the amount of angiotensin-II and increasing Ang(1-7)
The cell/tissue tropism of SARS-CoV-2 must logically have some effect on the function of the renin–angiotensin system (RAS). What is frustrating is that these questions remain unanswered.
[1] https://en.wikipedia.org/wiki/Angiotensin-converting_enzyme_...
[2] https://en.wikipedia.org/wiki/Pulmonary_alveolus#Type_II_cel...
"Put simply, medics found that severely ill flu patients nursed outdoors recovered better than those treated indoors. A combination of fresh air and sunlight seems to have prevented deaths among patients; and infections among medical staff.[1] There is scientific support for this. Research shows that outdoor air is a natural disinfectant. Fresh air can kill the flu virus and other harmful germs. Equally, sunlight is germicidal and there is now evidence it can kill the flu virus."
[1]: https://medium.com/@ra.hobday/coronavirus-and-the-sun-a-less...
Vitamin D may have modulatory effects on the biochemical pathways linked in the original article. Most pharmaceutical products we study aren't going to directly kill bacteria or viruses, usually they have a particular physical effect on a specific molecule that then leads, through various cellular pathways, to the death or suppression of the pathogen. I highly doubt Vitamin D directly kills anything, it could modulate symptoms or the pathophysiology of COVID-19 the disease, vs. SARS-COV-2 the virus itself.
If "it" means UV light, yes there are UV sterilizers aplenty today. If "it" means outdoors/fresh air, there are a lot of components to that and it's not like "fresh air" enters "in vivo" per se.
Yes, I'm aware of UV sterilizer, but does it work when the microorganisms are inside you? You don't want to get an "internal tan"
I read that vitamin D and observed benefits for those who have vitamin D is a correlation. Meaning that taking vitamin D supplements might not be as helpful as getting sunlight (a natural way to get vitamin D)
With regards to sunlight exposure, sunbathing in the alps was a go-to tuberculosis treatment prior to having antibiotics.
Very very interesting theories as it also helped during the 1918 flu.
"Here, we perform a new analysis on gene expression data from cells in bronchoalveolar lavage fluid (BALF) from COVID-19 patients that were used to sequence the virus. Comparison with BALF from controls identifies a critical imbalance in RAS represented by decreased expression of ACE in combination with increases in ACE2, renin, angiotensin, key RAS receptors, kinogen and many kallikrein enzymes that activate it, and both bradykinin receptors. This very atypical pattern of the RAS is predicted to elevate bradykinin levels in multiple tissues and systems that will likely cause increases in vascular dilation, vascular permeability and hypotension."
FTFY. They didn't measure patients. They modeled genomic interactions to make some predictions about biochemical effects on patients. Then they noted that some of those predicted effects correlate with symptoms of Covid patients. They went further and shotgunned a list of treatments which are known to affect the same biochemical processes. The farther along the path of inference, the weaker the conclusions get, but it sounds like a promising arrow for research to me.
> why they couldn't just measure bradykinine levels directly? Is that too hard?
Don't need nearly as much permission or human resources to run computer simulations on offline data as you do to take measurements of patients in the hospital.
It is. To measure gene expression, you isolate total mRNA and sequence it. This tells you the expression of all genes simultaneously. The protocol is fairly standard, cheap, and quick. That doesn't tell you anything about bradykinine, though, because there is no mRNA that codes for it.
In contrast, no such protocol exists for proteins. Sequencing a single protein is comparatively difficult, and no high throughput device exists that sequences lots of proteins, let alone quantifies their abundance. The traditional lab methods like PAGE gels are slow and labor intensive.
Could I ask you for any hints about where/how to learn more? And/or if there's a way to follow your work online?
If this info has been out since 5 months I'd expect several trials already that target the bradykinin storm.
https://elifesciences.org/articles/59177
Excerpts from the abstract:
Bradykinin is a potent part of the vasopressor system that induces
hypotension and vasodilation and is degraded by ACE and enhanced by
the angiotensin1-9 produced by ACE2.
This very atypical pattern of the RAS is predicted to elevate
bradykinin levels in multiple tissues and systems that will likely
cause increases in vascular dilation, vascular permeability and
hypotension.
Lay people use the two interchangeably.
In at least two of the thesaurus I have at hand hypothesis is synonymous is theory.
A theory is a explanatory framework for a body of knowledge. Unlike a hypothesis, it's not inherently a question or a guess. That doesn't mean it's "true." It also can be unconfirmed (as you say, string theory) or even demonstrably false (phlogiston theory, Ptolemaic theory) and still be a theory.
Obviously there's considerable overlap between the two concepts and as you say they are sometimes used almost interchangeably. Colloquially, "I'm testing my hypothesis that orally ingesting booze provides protection against infection, which if confirmed will be a key part of a theory of booze immunology" gets collapsed into "I'm testing my theory of booze immunology." Big deal. It really only matters because people let themselves get bent out of shape about the whole "evolution is just a theory" thing.
And since I'm ranting already, evolution isn't "just a theory" because evolution itself isn't a "theory," evolution is the natural phenomenon that is being theorized about.
I could find only the article below on IBM's news section (they created this super computer). Speaking about the results of the 2 day analysis, Jeremy Smith, Governor’s Chair at the University of Tennessee, director of the UT/ORNL Center for Molecular Biophysics, and principal researcher in the study: “Our results don’t mean that we have found a cure or treatment for COVID-19. We are very hopeful, though, that our computational findings will both inform future studies and provide a framework that experimentalists will use to further investigate these compounds. Only then will we know whether any of them exhibit the characteristics needed to mitigate this virus.”
https://newsroom.ibm.com/US-Dept-of-Energy-Brings-the-Worlds...
Really hoping this is better by the next few months or my Seattle winter will be even more pill popping to maintain basic human functionality....
For example: https://www.milkwood.net/2014/03/31/want-extra-vitamin-d-pla...
This works for both shiitake and button (portobello) mushrooms.
This trick came from Paul Stamets, one of the world’s foremost mycologist. He is based out in Casscadia. You may have to track down exactly what “2 days” means... I wouldn’t be surprised if he went looking for a way to maintain his Vitamin D health.
You only need 10g of this four times a week, according to this article. While it is summer time in Seattle, you can prepare a bunch and then dry them, so they last until next spring.
It does not simulate complex biochemical interactions in different parts of the body.
From the description, they did something that requires a lot more horsepower.
There certainly wasn't any "heavy biochemical calculations"; this work is entirely comparative genomics, so just operating on DNA strings.
I see this fluff article https://www.ornl.gov/blog/genomics-code-exceeds-exaops-summi... and there may be more detail here: https://www.hpcuserforum.com/presentations/april2019/Joubert... which shows near-linear performance they ascribe to "Made possible by aggressive communication overlap and low-congestion Mellanox Infiniband fat tree network with adaptive routing"
So there may actually be an HPC/supercomputer story in there, but I'm having trouble figuring out what they did in this most recent work.
> RNA-Seq analysis was performed using the latest version of the human transcriptome
I found this article discussing read mapping for RNA sequence analysis: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4833417/
> In particular, RNA sequencing (RNA-seq) technology,1 which provides a comprehensive profile of a transcriptome, is increasingly replacing conventional expression microarrays.2 Primary data processing in RNA-seq (as well as in other massive sequencing experiments, including genome resequencing) involves mapping reads onto a reference genome. This step constitutes a computationally expensive process in which, in addition, sensitivity is a serious concern
Reads mapping is a massively embarassingly parallel computation, again not something you would need or want a supercomputer for. You mainly need disk IO to/from the source reads and the mapping table you produce.
> RNA-Seq analysis was performed using the latest version of the human transcriptome (GRCh38_latest_rna.fna, 160,062 transcripts to which we appended the SARS-CoV-2 reference genome, MN908947). Mapping parameters were set with a mismatch cost of two, insertion and deletion cost of three, and both length and similarity fraction were set to 0.985. TPMs were generated for all 160,063 transcripts for the nine COVID-19 samples and the 40 controls (Supplementary file 2). The resulting transcript mappings for genes of interest were manually inspected to account for any expression artifacts, such as reads mapping solely to repetitive elements such as the Alu transposable element or all reads mapping to a UTR or pseudogene therein. Transcripts whose counts came solely from (or were dominated by) reads at repetitive elements were removed from the analysis. For the controls cases we ran an outlier analysis using the prcomp function in the R package factoextra. Input data were TPM for transcripts that averaged greater than one across all samples (30,102, Supplementary file 2).
For these guys, it's likely their best and only option. They probably weren't given the money to build a cluster optimized for their needs or to maintain a cloud instance. Why? Its oakridge, their main priority is HPC physics. It's hard to argue when you have access to such a HPC center. And HPC sites need all the customers they can get, lest their clusters get shoved into the cloud. It's a real fear. They'd end up with hidden costs, data lockin, and poor interconnects. To help pay for those peak massive simulations, traditional HPC need to fill up that last 10-15% and bioinformatics needs most of what they offer. Perhaps all that hard won knowledge will rub off on the burgeoning field too. :)
A vision of bio-oriented HPC is IU's clusters. Though they have a shiny new cray shasta with ampere and slingshot, several of their other clusters are 10 gigs with high mem nodes. All connected to the same storage too. The hospital is the main customer and dictates their designs.
Ditto on the paper though. It's what I disliked about Bioinformatics. All the glory to the researchers designing the experiments and they can't even bother to mention what software they used.
As for the cost structure for research computing, the argument that the costs are externalized isn't a good one- that overhead that pays for the facility, and the networking, comes out of your grant money, and using grad student time to admin your cluster often just causes your grad students to leave for FAAMG.
That has not been my experience. There are lots of scientific workflows that only need 10s of TB at most, yet can still consume lots of cycles.
> As for the cost structure for research computing, the argument that the costs are externalized isn't a good one- that overhead that pays for the facility, and the networking, comes out of your grant money, and using grad student time to admin your cluster often just causes your grad students to leave for FAAMG.
At the universities I've worked at, equipment (large purchases) is except from overhead, or results in a lower overhead charge. (Researchers balk at paying a ~50% overhead rate on a $1million instrument). Using grad student time to admin your cluster is dumb, but I'm more talking about users who need single-digit numbers of computers. If you need real HPC, you're in the world of queues, national and regional supercomputers, etc. etc.
The article is fluff.
98% of all bioinformatics is done on "supercomputers" or 'high performance computing environments" saying the researchers used supercomputers to analyze the expression data is like saying someone used a shovel to dig a hole.
That said, many problems that previously would have required a supercomputer, can now be solved on phones.
The shovel had racing stripes. Got it.
The job doesn't change based on what hardware it is run on.
Hard to question that.
It supposedly acts on the RAS and down regulates ace2
[1]: https://journals.physiology.org/doi/full/10.1152/ajpregu.000...
Why do you believe that?
Theories are attempts to explain the mechanism of something based on the observed data. Given data on patient symptoms and known drugs (ACE inhibitors in this case) and their effects, a computer could easily produce a theory that the disease acted like ACE inhibitors. It'd still take a human to write the program to generate these theories, but a computer could do it.
But what if the data represents ideas?
https://en.wikipedia.org/wiki/Automated_theorem_proving
> Unless the computer is conscious, it can not generate the theory.
Why would you say that consciousness is necessary for theory generation? It isn't for arithmetic, equation solving, natural language processing or image identification, etc.
>But what if the data represents ideas?
Then the computer would still be generating and analyzing data, not processing ideas.
>> Unless the computer is conscious, it can not generate the theory. >Why would you say that consciousness is necessary for theory generation? It isn't for arithmetic, equation solving, natural language processing or image identification, etc.
I think that the conscious analyst/observer is an intrinsic part of theory discovery, in the same way that a computer can not understand Chinese[1].
If the conscious observer is not necessary for a theory to exist, why is the computer necessary either? Certainly the phenomenon and data exist without it?
Arithmetic was deliberately mentioned, you might as well say "Of course a calculator app on your phone isn't _really_ doing arithmetic, the conscious analyst/observer is an intrinsic part of discovering the correct answer, in the same way that a computer can not understand compound interest".
There is a sense in which you are correct, but it is a very uninteresting one. Practical applications of computation follow from ignoring this semantic debate.
I think the fact that a computer can execute a program to compound interest isn’t a particularly novel one or interesting idea to me.
Going back to the original article, I think it was an unnecessary and incorrect anthropromorpiziation to write a computer discovered a theory of disease. Why isn’t my lazy laptop curing diseases?
I think there is a lot of interesting ideas in this semantic area. Can a computer compose all possible melodies and release them into the public domain[1]. If I write a script that formulates and posts every combination of "x variable cures cancer", did the computer or I discover a theory? If no, what are the minimum requirements?
https://www.google.com/amp/s/www.vice.com/amp/en_us/article/...
Yes, you're correct. However, identifying the correlations is a necessary precondition to making a theory about them. And the automated analysis can help with that step.
1. When a system is very simple, we explain its actions in terms of its properties and external forces acting on it.
2. When it's a medium complexity system, we tend to explain its actions in terms of its design, putting ourselves in the designer's shoes.
3. When it's a complex enough goal-seeking system, we begin to empathise with the system itself, thinking why "it chose" a course of action.
I remember how impressed I was with this classification and how much sense it made in terms of how we're able to understand the world and predict what will happen next.
From this angle, much of modern computer software is clearly in category 3, and it just makes things easier for us to think of it as having a mind of its own.
Quoting from wikipedia "For the simplest vehicles, the motion of the vehicle is directly controlled by some sensors (for example photo cells). Yet the resulting behaviour may appear complex or even intelligent."
Or do we simply fall into the trap the GP described, where once something is complex enough that we don't understand what is going on from a purely mechanical perspective we consider it close enough to human to empathize with it?
A much simpler question can be answered: is it possible to be cruel to an animal? The accepted answer revolves around whether or not you can influence the behavior of the animal (or machine). Crabs naturally hide under rocks. If you shock them when they hide under the rocks they eventually stop hiding under the rocks. Meanwhile a bacteria, by itself, will never learn to associate a stimulus with a condition.
That may capture the capability for thought. There are higher levels past it, like learning to model the behavior of other organisms, and those models eventually turning into a theory of self.
One of the best nuclear astrophysicists I know thinks about stars as entities that want to stay alive -- "I'm running out of hydrogen, what can I burn next?" That approach yields the right phenomenology almost all the time.
https://plato.stanford.edu/entries/intentionality/
"However, in the absence of detailed knowledge of the physical laws that govern the behavior of a physical system, the intentional idiom is a useful stance for predicting a system’s behavior."
When you try to describe a complex or subtle thing concisely, you might find it hard. Even if the system is neither animal nor human, you too might notice yourself reaching for "character with motivations" to describe it.
> Covid-19 is like a burglar who slips in your unlocked second-floor window and starts to ransack your house.
https://www.cs.utexas.edu/users/EWD/transcriptions/EWD10xx/E...
"It is probably more illuminating to go a little bit further back, to the Middle Ages. One of its characteristics was that "reasoning by analogy" was rampant; another characteristic was almost total intellectual stagnation, and we now see why the two go together. A reason for mentioning this is to point out that, by developing a keen ear for unwarranted analogies, one can detect a lot of medieval thinking today."
https://www.cs.utexas.edu/users/EWD/transcriptions/EWD08xx/E...
And the ever-popular
https://www.cs.utexas.edu/users/EWD/transcriptions/EWD12xx/E...
"When we returned from the interview, some more legal professionals had arrived and there was a lively discussion going on. For me the exposure was a cultural shock, instructive, but also rather disorienting. Of course I knew that lawyers are not scientists, yet the atmosphere of a trade school took me by surprise. Of course I knew that lawyers mainly deal with national law, yet I was unprepared for the prevailing parochialism. (Now I come to think of it, the system of common law, based —as it is— on custom and precedent, could very well strengthen this phenomenon.) but the most disorienting thing was that I found myself suddenly submerged in a verbal tradition that was totally foreign to me! They were on the average very verbose —some even repetitive—, they had a tendency to "reason" by analogy and more than once I felt that speakers cared more about the potential influence of their words than about what they actually said. (Are these common professional deformations of the trial lawyer?) I spoke for ten minutes, that is, I tried to do so: after several hours of exposure I no longer knew how to address this crowd."
You might phrase a title that way if you believe that we trust computers more than we trust scientists. Is that true? I don't think so. But if it is, how horrifying.
We evolved in kinship groups. The most important phenomena to understand were your fellow humans, followed by animals.
The human brain has a highly-optimised "Character with motivations taking actions" parser.
"A person used a supercomputer to analyse Covid-19" conveys no more knowledge than "A supercomputer analyzed Covid-19".