A Surprising Comparison of Male vs. Female Microbiomes

I must admit, I was curious. So I went over to the desk of our brilliant Lead Data Scientist, Dr. Siavosh Rezvan-Behbahani, to find out.

Could you look at all of uBiome’s gut samples, I asked, and see what the difference in microbiome composition is between men and women? And the corollary, is it possible to predict from microbiomial data whether the person giving the sample was male or female?

With human DNA, of course you can determine gender based on the chromosome signature XX vs. XY. But does the microbiome have a gender signature too?

Siavosh dove in. He spent many hours analyzing, plotting numbers, running different machine learning classifier algorithms. He looked at healthy male and female samples in part of our dataset, all the way down to genus level.

And here’s what he found, which blew my mind.

male female ubiome

It turns out that in our dataset, there is no statistically significant difference between male microbiomes and female microbiomes. And, given a random sample, we would not be able to determine if it came from a man or a woman.

This result is fascinating to me, because it suggests that maybe men and women aren’t that different in some ways. We all have two eyes, and belly buttons, and similar proportions of bacteria swimming around inside our intestines.

(Of course there’s the standard disclaimer that this is just what we observe in our gut dataset, and may not be representative of the entire human population. It’s also possible that there is a difference but it’s much more subtle than we expect. In any case, this result is encouraging me to think up other questions to ask!)

If you’re curious about something too, please tell me in the comments what kinds of data discoveries you’d like to hear about next.

38 thoughts on “A Surprising Comparison of Male vs. Female Microbiomes

  1. Tami

    Can you show us the absence of signal isn’t due to insufficient statistical power? I trust that you have large sample size– but I am interested in seeing some attempt at a power estimate.

  2. Ana

    It would be rather shocking if at the *PHYLUM* level there were statistically significant differences between men and women. It’s just such a high level of taxonomic classification, incorporating all sorts of bacteria that could do very different tasks. If they were statistically different at the phylum this would suggest that men and women have some enormously different bacterial flora – I would be more likely to believe that result is an error than a valid result. To give you some perspective, humans would be classified as Chordata at the phylum level – which includes species from mammals, fish, reptiles, birds and even some weird little guy called a sea squirt.

    The trade off is of course that it can be difficult to precisely identify bacteria genera or species that would allow us to make a more specific statement about bacterial diversity between genders or even individuals.

    1. Alexandra Carmichael

      Thanks for the question, Ana! The chart shows phylum level data, but we did analyze all the way down to the genus level, and didn’t find a statistically significant difference.

  3. Alexandra Carmichael

    Another reader wrote in with these great suggestions, so I’ll post them here to keep everything in one thread:

    What about a weight-specific analysis? Is there a microbiome of a “thin” person or “overweight” person, and what implications might this have for taking certain probiotics to maintain a healthy weight or supplement a weight loss program?

    Also, can you predict if someone is a smoker or heavy drinker, based on their microbiome?

  4. Scott

    I’d be much more surprised if they differed much at such a low level. I mean, Venus and Mars, but we’re still the same species, right?

  5. nwmotogeek

    I would like to see a comparison between people who drink beer vs those who do not. There has been a long held myth that beer is good for you.

  6. Enrique

    Numbers do not mean anything without their standard deviation or their estimated margin of error. Publish the male and female proportions with their standard deviation in the dataset and then we will draw conclusions.

    1. Alexandra Carmichael

      Thanks Enrique! We tested for statistical significance to a 95% confidence level, and will try to report more details about future analyses.

  7. Thanks for this insight! Just a quick question; was this analysis done on the fecal microbiomes, or on a more general, averaged total of microbiomes from multiple sites (e.g. also including skin, oral, etc)? Just wondering which particular type of microbiome you are reporting about. Thanks!

  8. Pingback: February 17, 2015 | Microbiome Digest – Bik's Picks

  9. Gnadigefrau

    Would it be possible to compare the microbiomes of people who are unable to recover from Lyme disease (chronic Lyme disease/post-Lyme disease syndrome) with people who recover on the standard regimen? Lots of research being done at Johns Hopkins right now, and I believe that many doctors suspect a genetic mutation is behind the 10-20% who never recover (or continually relapse). But why not look at the microbiome as well. It would be hard to tease out the effects of long term antibiotics, however. I have wondered if fecal transplants from recovered Lyme patients would help those who struggle.

  10. gseattle

    Well drat, I was hoping for a big difference, maybe influenced by higher prevalence of testosterone and estrogen, eventually discovering that sharing collections of microbes by occasionally kissing–plus another sharing mechanism that some of you might know–is extremely healthy.

    Hello sweetheart, I happen to have the best mix of microbial additions for your optimal health during the next several hours would have been my new pickup line.

    1. Alexandra Carmichael

      Hahaha well, I’m sorry to dampen your microbial dating spark! Guess you’ll have to use your wit instead. 🙂

  11. Michelle Berry

    I was wondering which models you ran. Did certain classifiers work better than others, or were they all pretty terrible? What sort of error rates did your models get? I experimented with the AGP dataset using random forests and SVM to try to model whether the person who had provided the sample had recently taken antibiotics. Given the more resolute trends I had seen in published studies of mouse models, I was surprised how poorly these algorithms performed when i ran them on the AGP dataset.

  12. Is a significant portion of your data set couples by any chance? Seems like that could skew the results.

    Also, I’m interested in how coffee affects gut microbes.

  13. Autoimmune

    would love to see the difference btwn those with and without autoimmune diseases! There is a lot of scientific data out there suggesting a correlation between certain gut bacteria and autoimmune disease.

    1. Biophile

      I vote for this one too.
      But it looks to me like the percentage of Verrucomicrobia in females is twice that of males; is it really insignificant?

  14. Alexandra Carmichael

    One more request came in by email! “I have been wondering if there is a difference in the gut biome between women on hormone replacement therapy, and women of the same age who are not on hormone replacement therapy.”

  15. eddie

    How about comparing RNA16 Bacteria , and RNA18 of fungi…… and coming up with a SCAN of fungi.. it would be good to compare ones fungi with there bacteria. Many block them , many with low good bacteria have high fungi and disease Please make an affordable RNA18 scan personally I would love for you to compare my past americangut results from people over there to my current UBIOME.. I am completely different bacteria wise– and now have NO gluten intolerance , dairy or starch problems… due to changing my FUNGI foot print as well bacteria —- they both work with each other

  16. Wildflowyr

    I’m wondering– is there any connection between one’s microbiome and the gene expression that influences methylation? Also, is there a connection between gut bacteria and the ability of the body to produce and use vitamin D from sunlight (as opposed to supplements)?


  17. Sima

    Very interesting, but ….How many people was the analysis based on? How many males? How many females? All adults? Or children also? And if children also, what percentage?

  18. Laura Henze Russell

    Could you look at the microbiomes of those with mercury toxicity?

    Of those with methylation genomic variants that make it harder to excrete mercury?

    Of those with and without dental amalgam (50% mercury) fillings?

    Of those with higher vs. lower levels of high-mercury fish consumption?

  19. OonaBelle

    What about a comparison of the microbiomes of folks with different IQ’s? Old versus young? Does the microbiome “age” Also those of different socioecomoic status? This is a touchy subject for some folks, for obvious reasons, but still, it would be fascinating to explore!

  20. Rob Boulton

    I submit that it might – or might *have* been, once – possible to determine the region, or at any rate the extent to which the contributors of samples have maintained traditional diets or have succumbed to the SAD diet structure of junk “foods” and carbonated drinks. This theory is based mostly on an excerpt from Michael Pollan’s book, “In Defense of Food”, which details the case of ten Australian aborigines living on a Western style diet in a small town and the range of diseases they had. A nutrition researcher persuaded them to go back to the bush style of living – eating the diet their ancestors had eaten for centuries. The story is available at https://farmerwu.wordpress.com/quotations/michael-pollan-from-in-defense-of-food/.

    I think it’s almost certain that the microbiome from samples at the time they were living in town and suffering from a range of diseases would be markedly different from the microbiome obtained had they given samples at the end of the seven weeks of the experiment.

Comments are closed.