MIT Scientists Invent New Way to Confuse Themselves With Even Tinier Pieces of Protein
In what can only be described as science’s latest attempt to out-nerd itself, researchers at MIT have developed FragFold, an AI-powered system with one lofty goal: making proteins even smaller and somehow still complicated.
Using sophisticated machine learning, FragFold predicts how tiny protein fragments attach to targets, essentially turning them into microscopic puzzle pieces researchers will argue about for the next decade. Scientists claim this could have groundbreaking implications for biology and medicine, or at the very least, will provide them endless material for research grants.
“This is very exciting,” said Andrew Savinov, a postdoc at the Li Lab. “We’ve discovered that even incomplete proteins can still do important things. It’s kind of like realizing that a missing piece from your IKEA furniture still allows the table to wobble in a functional way.”
MIT’s team boldly decided that instead of studying full proteins—because apparently, that’s too mainstream—they would focus on minuscule fragments that latch onto bigger proteins like the world’s tiniest parasites. Among other important findings, FragFold was able to identify proteins that interact fleetingly, contributing to cellular processes in ways even leading biologists will pretend to fully understand.
The research is entirely dependent on artificial intelligence, which was likely delighted to take on a project less terrifying than predicting stock markets and replacing human workers. Scientists say FragFold builds on the success of AlphaFold, an existing program that predicts protein structures, but with a key improvement: FragFold ignores context and focuses strictly on disrupting proteins in ways that might be useful.
One significant case study in the research featured the FtsZ protein, which plays a role in cell division. Scientists found that due to an “intrinsically disordered region,” FtsZ was difficult to study—a description that could also apply to the average college student five shots into a night out. However, FragFold managed to unravel new binding interactions, which scientists assure us is progress, even if it currently solves none of humanity’s actual problems.
The team also examined how these fragments could inhibit proteins, potentially leading to targeted treatments for diseases, but definitely leading to more science journal articles with titles indecipherable to the general public.
Skeptics have pointed out that FragFold relies on deep-learning AI, meaning that while it makes highly accurate predictions, no one really knows how the hell it works. “It’s like training a dog to juggle,” admitted one anonymous researcher. “It’s impressive, but if you ask me to explain how, I’ll just start sweating.”
The grand question still remains: Can scientists use this data to design proteins on purpose instead of haphazardly stumbling upon useful ones? The jury is out, but excitement runs high, particularly in the department where researchers now compete to see who can design the smallest, most impractical protein fragment possible.
“This could reshape medicine as we know it,” said Savinov, before immediately backtracking. “Or at least, it’ll keep us occupied until the next big funding cycle.”
