Content
@
https://warpcast.com/~/channel/synbio
0 reply
0 recast
0 reaction
Ayan
@gmo
you’ve heard of ChatGPT, but what about protein LLMs? For those new to proteins, they are the workhorse of all of your cells. They are made of amino acid chains that fold into precise, tiny molecular structures. Their form is super critical- for example, when they fold incorrectly, you can get diseases like Alzheimer's. Researchers at Meta (now Evolutionary Scale) took the same “attention” mechanism used to weigh which parts of text passages are most important for understanding context in LLMs and leveraged it for determining 3D protein structures directly from 2D data 60x faster than other methods. This is also really important because experimentally determined protein structures cover less than 0.03% of all known protein sequences and this is before we even understand their function. Thanks to the magic of LLMs being applied to protein sequences, we’re seeing huge breakthroughs in uncovering structure and function 🙏🏽
5 replies
2 recasts
21 reactions
Ayan
@gmo
Learn more here: https://arstechnica.com/science/2023/03/large-language-models-also-work-for-protein-structures/ And in this paper: https://www.science.org/doi/10.1126/science.ade2574
2 replies
0 recast
6 reactions
Drew Volpe
@drew
It’s fascinating how well NLP tech that was developed for human languages can work with bio data.
1 reply
0 recast
1 reaction
Ayan
@gmo
Super fascinating, there is a whole exchange and rediscovery happening where bio can leverage models that weren’t intended for it
0 reply
0 recast
0 reaction
