AlphaFold reveals how sperm and egg hook up in intimate detail | Hacker News Reader

AlphaFold reveals how sperm and egg hook up in intimate detail

nature.com

23 pointsBrajeshwar10 comments

This is such a terrible bit of "science journalism".

First, they don't bother to link to the paper (presumably because it was published in Cell?). So if you're wondering, here it is:

https://www.cell.com/cell/fulltext/S0092-8674(24)01093-6

Second, the writer doesn't bother to mention any of the actual science done here: the AlphaFold step, however sexy, was just a hypothesis-generating step. The reason the paper is in Cell is because they did a bunch of other, more important work to verify that the hypothesis was correct.

Third, I'm kind of offended that they make it sound like this was some unknown theory until AlphaFold revealed the magic of the universe. If you go do a publication search, you'll see that people have been challenging the 2-protein theory for a while. So it's cool that they found another factor, but again, AlphaFold didn't magically unlock this area of research.

It's really cool that a computer model was used to generate a hypothesis that was subsequently validated with (IMO) hurculean amounts of lab work; it's not cool that the lab work was minimized to a factor that was omitted from the summary, because that lab work is the only thing that got the paper in Cell (and in fact, what "reveals" how sperm-and-egg "hook up" yada yada yada.) Computer predictions are a gigantic haystack, but validated computer predictions are diamonds in that haystack. As always, you have no idea how many incorrect computer-generated theories the authors discarded before finding the one that got published.

In short: beware hype generated by "journalists" writing in the toilet-reading section of Nature and Science.

The "just" a hypothesis-generating step is THE HYPOTHESIS.

Generating THE HYPOTHESIS is THE non-scientific step in the SCIENTIFIC METHOD. To not undersell this: where the actual fuck does a hypothesis come from? Particularly one that works out?

Looking at your profile (now), you have a PhD in computation biology and you have 29k+ karma, but: what? I don't understand your objection, man.

A good hypothesis is hard to come by.

Can you help me out?

(computers generating hypotheses beat non-computer aided humans generating hypotheses should adequately engage a large class of arguments that your comment seems to suggest that you'd like to make)

Sure. You're wrong: a good hypothesis is trivial to come by. It's hard to prove one. You're confusing the latter step for the former, in the sense that you're defining "good" as "likely to be proven". And there's the rub.

Literally nobody in science needs a computer model to generate a question for them. Everybody wants the computer to do the part where it tells them the hypothesis that is most worth investing their effort in figuring out, but it cannot actually do that.

Or at least, nothing about this paper suggests otherwise. You have no idea how many computer-generated ideas they churned through before this one. If I ask AlphaFold to fold/dock me a protein, it will happily do so, and return me a structure. Is it right enough that I should run off and spend the next N years of my life validating it? Mu.

You also don't know how many experimental results that didn't coincide with their hypothesis they discarded (this is quite common in molecular biology- "run that gel until you get the bands in the spot you want!").

> to bypass the difficulties of working with membrane proteins in the laboratory, the team used AlphaFold to predict interactions between proteins

> AlphaFold predicted that three sperm proteins work together to form a complex. Two of these proteins were previously known to be important for fertility.

https://www.nature.com/articles/nature18448

Here’s the paper linked in the article.

Does anyone have experience using AlphaFold? I don’t have a Nature subscription to see how they specifically accomplished this, but I glanced at the AlphaFold GitHub, and I can’t quite wrap my head around getting it to predict something like this. I’m sure it’s not as simple as prompting ChatGPT with a verbal question.

That's not actually the paper discussed in the article. They didn't even bother to link the paper they're discussing.

The URL you've linked to is a summary of two other papers from Cell, which solved crystal structures of 2 of the 3 proteins involved.

Edit - here is the paper in question: https://www.cell.com/cell/fulltext/S0092-8674(24)01093-6

The link to the paper is at the bottom of this article. Or more correctly they put the text of a link address at the bottom of the article (it's not actually a link); the DOI address itself resolves to the Cell article.

They use AlphaFold Multimer [0]. Demo here [1].

[0]: https://cosmic-cryoem.org/tools/alphafoldmultimer/

[1]: https://www.youtube.com/watch?v=iGhdrPo6XQM

One step closer to mating with "V'Ger"

"Unable to evolve past pure machine logic, V'Ger returned home to Earth to find its Creator - human beings. Commander Decker volunteered to merge with V'Ger through Ilia, which caused the entity to complete its evolution into a new lifeform. V'Ger vanished and Earth was saved"

Reply on news.ycombinator.com