Although you seem to only disagree with two “root cause” errors haha—were there other errors included in “most of this”?
I’m also generally skeptical of the throw-stuff-at-the-wall-and-see-what-sticks approach to pharmaceutical discovery, but that’s for separate reasons. In particular, if the stuff doesn’t stick, then we don’t gain anything; “phenotypic” search requires that we get lucky and hit the target, and doesn’t give us any way to “move closer” to a solution based on what we see. That’s fine if our search space is small enough (in particular if it’s low-dimensional), but biological systems are not that simple.
(I’m also skeptical of the effectiveness of many existing pharmaceuticals found through this sort of search. It reeks of p-hacking potential, and I have little doubt that pharma companies are very good at p-hacking. Although I would also guess that it’s gotten less problematic in more recent years, as I hear clinical trial standards have tightened up somewhat. That said, there do seem to be some kinds of interventions which are “easy” to stumble upon, in the sense that there’s a ton of things which work—like antiinflammatories or antibiotics.)
I really like your homeostasis argument
Thanks! :)
But I don’t see why the situation can’t be closer to a fully-connected graph of many causes, requiring intervention on many nodes. (To be clear what I meant by “many causes,” I’d consider 4+ to be many.) That’s closer to the model I have in my head. What am I missing here?
If there’s a bunch of causes hooked up to a bunch of effects, then there shouldn’t be any reason for all the causes to push the effects in the same directions. It’s entirely plausible that two different causes both have some impact on Alzheimers, cancer, osteoporosis, atherosclerosis, arthritis, cataracts, and sarcopenia. But it’s pretty suspicious if two different causes both increase the risk of every single one of those things—if we had a complete graph with random weights on all the connections, then some factors should push some diseases up, while others should push down. Instead, we see a variety of mutations/interventions (like progerias, calorie restriction, etc) which all push most of these things in the same direction, which pretty strongly suggests that they’re operating through the same pathway.
Statistically, we can think of this in terms of principal components analysis. A fully-connected graph should have lots of strong principal components (after adjusting for age), rather than one component dominating (i.e. effects strongly correlated even after adjusting for age).
Also, do you have links for your last 3 bullet points?
Stem cells are a pain, because there’s a bunch of different types and the differences are not small. But I’ll give some info for HSCs specifically, since they’re relatively well-studied. Here’s a decent overview, and here’s a turnover measurement; turnover is on the order of once a year. Takeaway: HSCs turn over ~ once per year, so their count should be in equilibrium on a timescale of ~1 year. If the count is decreasing on a timescale much longer than that, then something upstream must either be decreasing their reproduction rate or increasing their removal rate.
I don’t have a good link on hand for telomeres, but wikipedia does mention the key puzzle piece: actual telomere shortening rates are around 50-100 bp per cell cycle (in vitro), whereas end-replication should only shorten by 20 bp per cycle; the difference is supposedly explained by shortening due to oxidative stress (though I have not personally checked those numbers). Also of note (though I don’t have a link for this): the experiments which show increase in (average) lifespan usually involve continuous telomerase expression; a one-time pulse won’t reset the aging clock. That generally means we’re treating an intermediate factor, not a root cause.
Here’s a particularly excellent paper on senescent cell turnover. Timescale is ~ weeks, so if the count is changing on a much longer timescale than weeks, then either the production rate is increasing or the removal rate is decreasing. Also, much like telomerase, senolytics experiments use continuous administration, not a one-time pulse.
I’m also skeptical of the effectiveness of many existing pharmaceuticals found through this sort of search. It reeks of p-hacking potential, and I have little doubt that pharma companies are very good at p-hacking.
I think p-hacking is effectively fought with the standard of preregistration and requiring two studies. What’s however not effectively fought is goodharting.
If you have a substance that changes a bunch of things in the body and you find out that one of those things is positive in a very specific case under specific trial conditions you can get the drug approved.
If you have a substance that changes a bunch of things in the body and you find out that one of those things is positive in a very specific case under specific trial conditions you can get the drug approved.
But it’s pretty suspicious if two different causes both increase the risk of every single one of those things—if we had a complete graph with random weights on all the connections, then some factors should push some diseases up, while others should push down. Instead, we see a variety of mutations/interventions (like progerias, calorie restriction, etc) which all push most of these things in the same direction, which pretty strongly suggests that they’re operating through the same pathway.
It’s all about the damage to repair/replacement balance. Birds have higher repair rates, as do NMRs, as do bowhead whales. CR decreases synthesis of proteins that could be damaged while simultaneously (and strangely) upregulating repair. It might be instrumental to look at all the TFs activated by CR.
I’m also generally skeptical of the throw-stuff-at-the-wall-and-see-what-sticks approach to pharmaceutical discovery, but that’s for separate reasons. In particular, if the stuff doesn’t stick, then we don’t gain anything; “phenotypic” search requires that we get lucky and hit the target, and doesn’t give us any way to “move closer” to a solution based on what we see. That’s fine if our search space is small enough (in particular if it’s low-dimensional), but biological systems are not that simple.
(I’m also skeptical of the effectiveness of many existing pharmaceuticals found through this sort of search. It reeks of p-hacking potential, and I have little doubt that pharma companies are very good at p-hacking. Although I would also guess that it’s gotten less problematic in more recent years, as I hear clinical trial standards have tightened up somewhat. That said, there do seem to be some kinds of interventions which are “easy” to stumble upon, in the sense that there’s a ton of things which work—like antiinflammatories or antibiotics.)
Thanks! :)
If there’s a bunch of causes hooked up to a bunch of effects, then there shouldn’t be any reason for all the causes to push the effects in the same directions. It’s entirely plausible that two different causes both have some impact on Alzheimers, cancer, osteoporosis, atherosclerosis, arthritis, cataracts, and sarcopenia. But it’s pretty suspicious if two different causes both increase the risk of every single one of those things—if we had a complete graph with random weights on all the connections, then some factors should push some diseases up, while others should push down. Instead, we see a variety of mutations/interventions (like progerias, calorie restriction, etc) which all push most of these things in the same direction, which pretty strongly suggests that they’re operating through the same pathway.
Statistically, we can think of this in terms of principal components analysis. A fully-connected graph should have lots of strong principal components (after adjusting for age), rather than one component dominating (i.e. effects strongly correlated even after adjusting for age).
Stem cells are a pain, because there’s a bunch of different types and the differences are not small. But I’ll give some info for HSCs specifically, since they’re relatively well-studied. Here’s a decent overview, and here’s a turnover measurement; turnover is on the order of once a year. Takeaway: HSCs turn over ~ once per year, so their count should be in equilibrium on a timescale of ~1 year. If the count is decreasing on a timescale much longer than that, then something upstream must either be decreasing their reproduction rate or increasing their removal rate.
I don’t have a good link on hand for telomeres, but wikipedia does mention the key puzzle piece: actual telomere shortening rates are around 50-100 bp per cell cycle (in vitro), whereas end-replication should only shorten by 20 bp per cycle; the difference is supposedly explained by shortening due to oxidative stress (though I have not personally checked those numbers). Also of note (though I don’t have a link for this): the experiments which show increase in (average) lifespan usually involve continuous telomerase expression; a one-time pulse won’t reset the aging clock. That generally means we’re treating an intermediate factor, not a root cause.
Here’s a particularly excellent paper on senescent cell turnover. Timescale is ~ weeks, so if the count is changing on a much longer timescale than weeks, then either the production rate is increasing or the removal rate is decreasing. Also, much like telomerase, senolytics experiments use continuous administration, not a one-time pulse.
I think p-hacking is effectively fought with the standard of preregistration and requiring two studies. What’s however not effectively fought is goodharting.
If you have a substance that changes a bunch of things in the body and you find out that one of those things is positive in a very specific case under specific trial conditions you can get the drug approved.
Yeah, that sounds right. Good point.
It’s all about the damage to repair/replacement balance. Birds have higher repair rates, as do NMRs, as do bowhead whales. CR decreases synthesis of proteins that could be damaged while simultaneously (and strangely) upregulating repair. It might be instrumental to look at all the TFs activated by CR.