shades logo

shades fills you in

AlphaFold is a project from DeepMind, Google’s AI wing. It has a daunting task: Predicting how proteins fold. Why’s that so hard? Well, there are billions of possible ways for a protein to fold, so it’s really hard to predict what a protein will look like (and therefore what it’ll do)
AlphaFold
DeepMind
AlphaFold is a project from DeepMind, Google’s AI wing. It has a daunting task: Predicting how proteins fold. Why’s that so hard? Well, there are billions of possible ways for a protein to fold, so it’s really hard to predict what a protein will look like (and therefore what it’ll do)
YouTube
A refresher from bio class: A protein is like a beaded necklace, with each bead representing an amino acid. Depending on the combination of amino acids (the “beads”), the protein (the “necklace”) will fold in on itself in a unique way. The final 3D shape determines what the protein will do
DeepMind
Mapping out proteins is a 50 year old problem. In the 70s, scientists thought maybe the number of protein shapes was finite, also known as solvable. Since then, the best and brightest have tried to crack this nut. None were successful until DeepMind had a breakthrough: their AI could predict the shape of a protein with incredible accuracy
Jakemp
In 2018 AlphaFold nailed the C.A.S.P. -- the “World Championship” competition of predicting how a protein will fold. It stands for Critical Assessment of Structure Prediction
DeepMind
2 Years Later, AlphaFold 2 Came Out, and It Was Even Better Than the First Version. Here’s a Timeline of the Breakthroughs!
BloPig
After 2020 C.A.S.P., DeepMind published an in-depth paper on the process. They also posted the source code for anyone to use. Blopig says the disclosure was a little disappointing -- no real surprises. The breakthrough was really about applying superb engineering and computing to known ideas
Foxchase
Foxchase agrees, and says the breakthrough will speed up development of new cancer meds
EMBOPress
But there are real limits. Biochemists remind us that AI is often only as good as its database, and there just isn’t much information on all the unique protein shapes
EMBL
EMBL agrees, and points to another AlphaFold 2 limit: It categorizes proteins while they’re static, but most proteins are dynamic and change based on their surroundings
MIT Technology Review
But Technology Review says the predictions are useful even when they aren’t super accurate -- approximations are often good enough
Analytics India Mag
Analytics India says the next step is inverse protein folding -- aka, analyzing a structure to see which protein will interact with it. Quantum computing will help a lot with this!
up next...
Soylent and the like
Soylent and the like
Transhumanism
Transhumanism
Immortality
Immortality
Nanotech
Nanotech
Buy a foldable?
Buy a foldable?
Snapchat AI, rogue?
Snapchat AI, rogue?
Android, better??
Android, better??
Pause AI?
Pause AI?
Free Lambda
Free Lambda
The qubit breakthrough
The qubit breakthrough
Quantum entanglement
Quantum entanglement
GMO mosquitoes
GMO mosquitoes
K18 Hair Mask
K18 Hair Mask
Mini brains
Mini brains
The war on superbugs
The war on superbugs
shades logo