Comment by ProjectArcturis | Hacker News Reader

Comment by ProjectArcturis | Hacker News Reader

ProjectArcturis

Neuro PhD here. Here's the experiment they actually did: create a transgenic mouse that overexpesses beta-amyloid. They're all messed up, not just their memory, but we think beta-amyloid causes Alzheimer's, so this is our mouse model of Alzheimer's. Then further screw with the mouse by blocking its CSF drainage. Anti-amyloid drugs help the standard transgenic mice, but don't help the blocked ones. Then unblock the blocked mice, and Voila! you've fixed them. A cure for Alzheimer's is right around the corner, right?

Except, this doesn't necessarily have anything to do with Alzheimer's in humans. You're assuming it's caused by beta-amyloid, but there have been a dozen giant Phase 3 trials testing amyloid-removal drugs and they've all failed. Fundamentally, all you've done is mess with mice in two different ways, then when you undo the ways you've screwed with them, they're more or less back to normal.

This all tells you virtually nothing about how Alzheimer's actually happens in humans, much less how to cure it.

> This all tells you virtually nothing about...

This seems like a somewhat common trend in medical research...or something along the lines of findings being meaningless due to "causal vs correlated" questions. After reading these types of research issues, I always think that there's gotta be a better way to go about it by utilizing wisdom of crowds across industries.

For example, there's a chance electrical battery engineers needed to figure out a way to drain electrons in a special way to increase the life of the batteries. And say for example that same underlying logic is the solution to curing alzheimer's...there's no current way (that I'm aware of) to compare underlying processes that are parts of complex systems. The internal parts can be complex systems themselves, but if we can distill complex systems down to their core components, separate them into separate "repositories" that include the detailed processes and/or components necessary to make that given repository "work", we could essentially create "templates" of core knowledge....almost like a github for knowledge.

My main point is the individual pieces required to solve these types of complex issues may already be solved...would be cool to create a platform to manage those pieces.

How far off do you think we are from simulating an entire human on a protein level in accelerated time?

Imagine a world where a drug concept could be simulated on an entire human model in parallel cradle to grave 1000x faster than normal. What level of compute is necessary to pull that off?

Decades at least, we don't even have close to a complete list of the proteins in our bodies which would be needed to start thinking avout simulating them.

We're still struggling to accurately simulate a C. elegans. We're a little way off simulating 50+ kg of conscious organism.

Not to mention the ethical implications. If we're happy to run "cradle to the grave" tests on a human simulated to that level of accuracy, why not just run them on actual humans bred for research? We can do that right now, far cheaper.

...of course, maybe that just means that we're the disease model...

Ethics aside, if we could simulate a whole brain, then we could simulate billions of brains and try all sorts of things, something we wouldn't be able to do with physical humans. At least not without being labelled as mass murderers and turning into a bird.

This made me laugh. If you put ethics aside, couldn't you just do all those things on physical humans?

I have a hard time believing we'd find it ethical (at least in the long term) to experiment on a simulated brain. If it were accurate enough, might it be considered a person?

Why wouldn't an accurately simulated brain be considered a person?

> Why wouldn't an accurately simulated brain be considered a person?

Ah, and here we come to the crux of the matter! What makes a person a person? Is it merely electrochemical activity in a certain type of tissue?

It's important to recognize that this question comes to the border that marks the end of the territory of biology and the beginning of the territory of philosophy. It's possible to charge right through this border (as Stephen Hawking did in his book The Grand Design) but there is a peril in doing so.

https://www.theguardian.com/commentisfree/2013/may/27/physic...

> The attempt to fit consciousness into the material world, > usually by identifying it with activity in the brain, has failed dismally, > if only because there is no way of accounting for the fact that certain nerve impulses > are supposed to be conscious (of themselves or of the world) > while the overwhelming majority (physically essentially the same) are not.

Is he charging through the border, or gently rebuking the concept of a border?

OK this is still bugging me a day later... I'm missing the reference with the bird bit. What's that from?

If you are willing to bend some ethics, brain organoids and other similar organ-on-chip solutions are quite promising. Though it is important to note that scaled down brains may not necessarily express the same behavior vis-a-vis a full sized human cortex.

Well, if they did, you could never power them down...

I did say if you are willing to bend some ethics.

500+ years? Last I looked into what some major players in computational neuroscience were doing simulating highly accurate neurons they were doing on the order of maybe thousands of neurons. It’s been a few years so maybe they can simulate 100k today. The human brain has 100billion neurons. And that’s just part of a system. I don’t know how many other cells you’d need to simulate to model just the space between your ears. And I don’t think we’re even talking at the protein level yet of simulations. Proteins are notoriously difficult to simulate. If bitcoin had managed to use protein folding as it’s proof of work I’m not sure we’d even be able to fill the current demand for research let alone simulating an entire human. I don’t think I can fathom the amount of compute we’re talking about to simulate a human at the drug interaction level to any real degree of accuracy.

That seems overly pessimistic to me, in the last hundred years we've gone from basically no understanding on a molecular level of the human body to a decent understanding of much of the chemistry of the body, it seems like the next hundred years could conceivably advance to the level where we have a good understanding of even the smallest interactions.

we can’t simulate a single protein long enough to determine how it folds, even with a supercomputer. we’re so many orders of magnitude away from such a thing it’s basically unthinkable.

Very, very far. There are 10^27 atoms in a human body. Even if you go down to individual proteins, that saves you maybe 4-5 orders of magnitude. That's about a zettabyte of data just to describe initial conditions.

So is the issue that beta-amyloid is the smoke, rather than the fire? Or rather that mice aren't humans?

I'm not a doctor. I haven't worked in a wet lab. All this is from second hand conversations and anecdotes.

Mice aren't human.

There's lots of ways to cure cancer in mice. it's practically a joke. Mice can tell you a bit about danger - if your treatment kills the mouse, think pretty hard about that treatment on a human. It might kill the human too, but maybe not. Might be ok for the mouse but not the human, like maybe thalidomide.

The impression I get is, if it's ok for mice it _might_ be ok for humans, so it helps avoid a large swath of super dangerous stuff. But doesn't tell you a whole lot about what happens in humans. I'm sure there are things that kill mice that are effective treatments for humans, but both of those edges are outlier-ish. If it kills the mouse it's probably not worth continuing without a good understanding of why it might work in humans.

It's that transgenic beta-amyloid mice are not humans with Alzheimer's.

BUT!

As another neuro PhD: this is still an important finding for our understanding of protein (dis)aggregation in the brain. The opposite finding was totally possible. The protein disaggregation could have been independent of CSF flow. The typical HN hot-take is "not so fast, this translational research actually underdelivers," but I prefer, "this fundamental neurosci research has the added bonus of being preformed on a clinically relevant model."

I would disagree that transgenic A-beta mice are a clinically relevant model. I mean, I understand why we started there -- at first glance Alzheimer's is amyloidosis in the brain. But after all this time, it's pretty clear that removing amyloid doesn't cure AD, and consequently that overexpressing amyloid in a mouse is not a good animal model of AD.

It could be that amyloid is the smoke. It could also be residue from the body's fire extinguishers. That is, something breaks, and amyloid accumulation is a consequence of whatever the body does to try to fix it. Maybe it makes it worse. The point is, we don't know, and 30 years of presuming that we do has led us to a giant dead end.

Reply on news.ycombinator.com