Hacker News new | past | comments | ask | show | jobs | submit login
Predicting sex from retinal fundus photographs using automated deep learning (nature.com)
37 points by dayve on Dec 5, 2022 | hide | past | favorite | 20 comments



Wow, what a roller-coaster that headline was for me.

I read "predicting sex from rectal fungus...", so I was thinking this was somehow going to relate gut bacteria micro-biomes to peoples' sexual activities.

Ah no; it's determining a person's sex by looking at their retina.

I kinda want to read that other imaginary article now, though...


Let the AI do the job for you:

Introduction:

The human gastrointestinal tract is a complex and dynamic ecosystem, containing a diverse array of microorganisms that play a crucial role in maintaining health and well-being. Recent research has suggested that the composition of the gut microbiome may be associated with a variety of health outcomes, including sexual behavior and preferences. In this study, we sought to investigate whether it is possible to predict an individual's sexual activities from the presence and abundance of rectal fungal species.

Methods:

We recruited a sample of 100 participants (50 male and 50 female) from a sexually active population, and collected rectal swab samples from each individual. The samples were processed and analyzed using high-throughput DNA sequencing techniques, which allowed us to identify and quantify the abundance of fungal species in each sample.

We then used statistical analysis to assess the relationship between the presence and abundance of rectal fungal species and the sexual behaviors and preferences reported by each participant. In particular, we focused on three specific behaviors: frequency of sexual activity, number of sexual partners, and preference for same- or opposite-sex partners.

Results:

Our analysis revealed a significant association between the presence and abundance of certain rectal fungal species and sexual behavior. In particular, individuals who reported a higher frequency of sexual activity were more likely to harbor greater abundances of the fungal species Candida albicans and Saccharomyces cerevisiae. Similarly, individuals who reported a higher number of sexual partners were more likely to have higher abundances of Candida glabrata and Cryptococcus neoformans. Finally, individuals who reported a preference for same-sex partners were more likely to have higher abundances of Aspergillus niger and Trichophyton mentagrophytes.

Discussion:

Our findings suggest that the presence and abundance of certain rectal fungal species may be predictive of an individual's sexual behavior and preferences. Further research is needed to confirm these findings and to investigate the potential mechanisms underlying these associations. In particular, it will be important to understand whether the presence of these fungal species directly influences sexual behavior, or whether it is simply a marker for other factors that affect sexual behavior.

Conclusion:

In conclusion, our study provides preliminary evidence that it may be possible to predict an individual's sexual behavior and preferences from the presence and abundance of rectal fungal species. Further research is needed to confirm and expand upon these findings, and to explore the potential mechanisms underlying these associations.


Holy shit.

If you posted it somewhere out of context, and with a made-up DOI, I'd assume you copy-pasted from a real, legitimate paper. It looks short, but that's obviously because the PDF contains images and tables that couldn't be pasted into a HN comment textbox. Right? For my casual interests, I probably wouldn't bother looking up the original PDF, or even the names of the fungal species in the text.

Can't wait for a model that can generate full PDF papers, complete with plausibly-looking images.


I just asked ChatGPT to make it for me. Presumably I could have it make some LaTeX markup to make the paper look pretty, and maybe some markup for the graphs, and turn that into a PDF.


That's what scares me. I didn't know you could generate plausible-sounding papers (with flaws being easily explained away by limitations of the medium of text-based news/discussion boards). And by a generally-accessible tool for that matter - correct me if I'm wrong, but ChatGPT is pretty much the only case where one can use a state-of-the-art model without getting on some long waitlist for blessed individuals, or paying up front.


Hmm, so the less-healthy-sounding fungi (black mold and ringworm) are associated with homosexuality. I wonder if this is a subtle case of "machine learning is like money laundering for bias", or I'm reading too much into it (:


For a second I thought this was going to be "Prediction that two people in the photo or video frame are going to engage in intercourse" with deep learning. That would be an interesting article though.


This is what I also thought -- that an image taken of someone's retina could predict whether they would later engage in sex with the person they were looking at.

Someone smarter than me, build the thing, put it on a phone app, and make the next big dating app!

No more swiping, just look at the picture for a subconscious second, and let the phone camera + our algorithm swipe for you!


That would be fantastic. Someone should totally build a model to do that!


> However, deep learning has shown that these algorithms demonstrate capability in tasks which were not previously thought possible. Harnessing this power, new insights into relationships between retinal structure and systemic pathophysiology could expand existing knowledge of disease mechanisms... Here, the physiologic cause and effect relationships are not readily apparent to domain experts21.

Interesting. I'm curious to learn how a black box model help expand existing knowledge of disease mechanisms? It seems to me that doing so would require the ability to determine what the predictive indicators the model is using are.


Finding novel relationships gives us news places to look to try to understand how things are. I didn't know you you could tell gender from retina. If indeed we can, that leaves room to then all why is that biologically happening. We don't be the black box AI to explain. If it read a signal there, it's up to use user other methods to figure why that signal exists and what the implications are to our current understanding


> I didn't know you you could tell gender from retina.

This is determining sex from retinal scans - not gender. Thats what makes this interesting, and potentially inflammatory, because it highlights even further subtle physiological differences between the two majority human sexes.


How do you know? I didn't see anything in the experimental design that would distinguish between detecting sex vs gender. Sex and gender are fairly highly correlated and it doesn't look like they did any analysis to rule out that they weren't actually detecting gender and not sex. Of course that would be a much more surprising result.

That said, i don't think subtle physiological differences are really that inflamatory. For example, we already accept that men are more likely to be colour blind and a myrid of other minor differences. Usually these things are only controversial when someone posits a major physiological difference and then mostly because much of the time its someone with an axe to grind significantly overstating the evidence.


> I'm curious to learn how a black box model help expand existing knowledge of disease mechanisms?

Even knowing that there is a predictive indicator is valuable, as it can guide future efforts. Even with no other clues, I suspect that researchers will discover the indicators much faster than they otherwise would merely because it is proven to be possible. Knowing something is possible has a powerful psychological impact.


Is this just like nearly every other AI paper like this where the black box actually learned how to distinguish metadata instead? How were the files named, were the female and male pictures collected using two different instruments that would have different noise distributions etc. 9 times out of 10 this kind of paper just turns out to be "We didn't properly normalize our data and reality slipped through", not anything actually meaningful.


The paper says it both predicts sex, but also says gender. I am guessing what it is actually able to predict is which sex hormone is dominant. For example: If a man with prostate cancer is taking estrogen to suppress T (and hopefully prevent the prostate cancer from growing/spreading) for even six months the colors he will perceive will shift as estrogen is dominant and the cells in his eyes will respond to that change. The paper briefly mentions this: "Others have demonstrated variation of ocular blood flow and have suggested the effect of sex hormones, but thus far, consensus is lacking".


Or it's dependent on increased gene expression from dual X-chromosomes or genes from an Y chromosome. Or sex-specific methylation patterns or some other factor, that are irreversible once the organism has grown beyond the fetus-stage.

There is way more to sex than just steroid hormones.


Adding to the misread titles, I thought it would predict if you get laid based on your photos and got a bit sad haha.


[flagged]


In this gender-fluid times I foresee something like that would cause quite a bit of controversy.

I was surprised to read they used "self-reported sex" for training, not sure how you could predict a biological trait from biological features using a self-reported truth as training label. It works because most of the time genetic sex and self-reported coincide but they list at least one testing error where the model predicted genetic sex and it was different from self-reported one.


An ocular scan just so kids can put different pants on? Christ on a stick this sounds awful.

You just described a nightmare scenario




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: