3 ms·

The issue is rather complicated. Yes, you want to have a diverse sample, no questions here. However, there might be issues with the statistical design. Mice are

by suslik 4y ago

The issue is rather complicated. Yes, you want to have a diverse sample, no questions here. However, there might be issues with the statistical design. Mice are territorial animals, and male mice get especially aggressive in a presence of a female. Put a couple of male mice in a cage, they'd be fine; add a girl in there, and there will be casualties. That creates extreme stress in both male and female mice, and stress is an important confounding factor in neuroscience and beyond. Grouping mice by sex in different cages is obviously incorrect - cage effect is a major factor in animal research. Not sure how these problems are solved now - maybe there are statisticians in the audience who can shed some light on that.

digging 4y ago
> Put a couple of male mice in a cage, they'd be fine; > Grouping mice by sex in different cages is obviously incorrect - cage effect is a major factor in animal research. What does this mean?
- suslik 4y ago
  To simplify, this means: if you have a following setup Cage 1: male1-control, male2-control, male3-treatment, male4-treatment Cage 2: female1-control, female2-control, female3-treatment, female4-treatment and you see females responding to treatment better than males, what does that mean? It can be any of the following: 0: females respond differently to the treatment due to their biochemistry (say, some Y-chromosome gene is interfering with the treatment) 1: females respond differently to the treatment due to social effects (a male would respond to treatment but not in an environmnet where he has to compete with other males) 2: there is a cage effect (say, it is slighly warmer, or more light, or fresh breeze, or something else mice care about) that causes the animals in one cage to respond differently. Only with the above setup, you cannot know which answer is correct. Statisticians have a bunch of tricks to tackle some of the issues above. For example, you can crossover (swap cages after some time) or increase the number of cages and randomise across cage conditions (say, you have 20 cages, 10 all-male and 10 all-female, spread around the facility to cover the variability of the cage conditions). Still, you won't be able to distinguish between #0 and #1. This is not really my area though, so I want a knowledgeble reader to forgive my ignorance. I am a simple bioinformatician who was never too deeply involved into this type of pre-clinical biostatistics. I'm sure someone on HN has an answer how to properly design experiments like that. My message was that it's a bit more complicated than just 'we need female mice in research' - and I saw way too many irresposible biologists sacrificing 100s of mice in experiments where a correct inference couldn't be made.

by suslik 4y ago

digging 4y ago

> Put a couple of male mice in a cage, they'd be fine; > Grouping mice by sex in different cages is obviously incorrect - cage effect is a major factor in animal research. What does this mean?

suslik 4y ago
To simplify, this means: if you have a following setup Cage 1: male1-control, male2-control, male3-treatment, male4-treatment Cage 2: female1-control, female2-control, female3-treatment, female4-treatment and you see females responding to treatment better than males, what does that mean? It can be any of the following: 0: females respond differently to the treatment due to their biochemistry (say, some Y-chromosome gene is interfering with the treatment) 1: females respond differently to the treatment due to social effects (a male would respond to treatment but not in an environmnet where he has to compete with other males) 2: there is a cage effect (say, it is slighly warmer, or more light, or fresh breeze, or something else mice care about) that causes the animals in one cage to respond differently. Only with the above setup, you cannot know which answer is correct. Statisticians have a bunch of tricks to tackle some of the issues above. For example, you can crossover (swap cages after some time) or increase the number of cages and randomise across cage conditions (say, you have 20 cages, 10 all-male and 10 all-female, spread around the facility to cover the variability of the cage conditions). Still, you won't be able to distinguish between #0 and #1. This is not really my area though, so I want a knowledgeble reader to forgive my ignorance. I am a simple bioinformatician who was never too deeply involved into this type of pre-clinical biostatistics. I'm sure someone on HN has an answer how to properly design experiments like that. My message was that it's a bit more complicated than just 'we need female mice in research' - and I saw way too many irresposible biologists sacrificing 100s of mice in experiments where a correct inference couldn't be made.