/Tech4h ago

Linguists' Sequence Modeling Research Powers AlphaFold2 Protein Breakthrough

2100565

Original post

Computational linguistics sustained the research (seq modeling, encoder/decoder, transformer,...) that AlphaFold2 adapted to protein structure prediction.

"The lineage is counterintuitive: solving a grand challenge in structural biology required having had linguists in AI." 3/

Rachel Thomas@math_rachel

Evolutionary concept of exaption: when traits evolve for one function & later crucial for another - bones for jaw mechanics ➡️ repurposed into middle ear - magnetrons: radar ➡️ microwave cooking - GPUs: game graphics ➡️ deep learning - U-nets: segmentation ➡️ generative AI 2/

8:35 PM · Jun 27, 2026 · 517 Views

Sentiment

Sentiment building, check back later.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS215LIKES2

Rachel Thomas@math_rachel

I recommend reading the full paper 6/

4h2152

REPLIES1

Rachel Thomas@math_rachel

Re: backprop & DL: "The deeper lesson is that the field could not have specified in advance what it needed, and therefore could not have benchmarked toward it.

"Someone had to be working at the boundaries, in contact with multiple traditions, for that bridge to be built" 4/

Rachel Thomas@math_rachel

Computational linguistics sustained the research (seq modeling, encoder/decoder, transformer,...) that AlphaFold2 adapted to protein structure prediction.

"The lineage is counterintuitive: solving a grand challenge in structural biology required having had linguists in AI." 3/

4h4800

Rachel Thomas@math_rachel

"Benchmark-driven selection has been extraordinarily effective at harvesting accumulated exploration capital...

"Benchmark-driven selection has harvested more than it has planted" 5/

4h1481