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.
> 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.
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.
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.
> 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.