Hacker News Reader

Comment by throwaway4666 | Hacker News Reader

Disclaimer: haven't seen the video (I never click any video link, sorry), but I did read the text links provided in the thread. That said: one of the big reason to use mice as animal models is that we've obtained a bunch of pure-bred, homozygous lines for a bunch of different genes, and that process is expensive and above all lengthy. Mice have both (a) a relatively short generation time and (b) been experimented on for a very long time. This makes them particularly suited for experiments involving knocked-out genes and suchlike because you have near-perfect controls at your disposal.

Now just because we have a bunch of pure-bred homozygous mice doesn't mean mice behave the same way as humans with respect to e.g. metabolic pathways, immune system etc. It's just that we often don't have better models.

View on HN

The problem is that the breeding protocols inadvertently created a selection bias towards long telomeres. Mice with extra long telomeres have exceptional tissue repair capacity while also being susceptible to tumor growth.

The supply of lab mice can be corrected to have a more natural telomere length. All past studies that used mouse models to study the impact of drugs are skewed to under report toxicity and over report carcinogenicity.

That fixes only this one specific problem though, the breeding protocol has probably introduced a lot of biases.

I imagine that “the fix” involves a new strain of more natural mice and a better breeding protocol moving forward. Part of Weinstein’s issue is that this fix has not been transparent. I hope someone can provide greater detail.

> Mice with extra long telomeres have exceptional tissue repair capacity while also being susceptible to tumor growth....

> All past studies that used mouse models to study the impact of drugs are skewed to under report toxicity and over report carcinogenicity.

Could the data from long telomere mice be de-skewed to compensate for new-understood effects the long telomeres?

I don't think it is linear. Fast tissue recovery can make side effects unobservable

I don't think it is just about mice though- it is also about "pure bred" anything. Imagine there was a single human that we could do all kinds of knockout experiments with, but we would always be starting with that base human (and on top of that it would be inbred and always live in the same kind of environment). Some of the conclusions we would draw would certainly not extend to all humans, probably not to most even. If the entire field became centered around this human, we would be missing huge amounts of information in space we are not exploring at all.

Controls are great, and it makes perfect sense to have studies that produce these extremely well controlled results. But controlling does have its own tradeoffs in practice, as results may very well not be robust/not generalize to things we care about. My problem is that the entire system operates under essentially the same paradigm, so we have very little work trying to address these downsides.

The solution to population variability is not to use variable models. That just trashes your result set.

There are enough confounding, uncontrollable variables in biology already.

Seeing how identical individuals from literally the same cell line in an exactly identical and controlled environment sometimes give different results I don't think you can ever achieve 'perfect controls' anyway. The solution is to delve deep into what things actually do (in the case of people I know, that involves a lot of live single-cell imaging and suchlike) but that's expensive.

I didn't propose any specific solution, all I said is that the field doesn't do enough to account for population variability. It is hardly acknowledged! You will find PhD students at top programs who say "huh hadn't thought about that being a problem before". Because the training and grants and whatnot largely revolve around this same experimental setup. And IMO it is not a good approach to have everyone working on these problems think in largely the same way.

Btw, in these cases where they find a result seems to only apply to a particular mouse strain, that feels like a pretty trash result to me too. What are we supposed to do with that knowledge? Currently what we often do is build on it, cite it completely out of context, and happily walk away with a publication.

Reply on news.ycombinator.com