AlphaFold Unofficial

AlphaFold-Driven Structural Proteomics Reveals Extensive Cellulosome Machinery in Human Ruminococcal Symbionts https://www.biorxiv.org/content/10.64898/2026.02.05.704116v1

AlphaFold is one of those rare breakthroughs where the scale of impact is hard to overstate. Decades of crystallography work compressed into hours. The downstream applications in drug discovery alone will keep compounding for years.

AlphaFold—for predicting protein structures - Lasker Foundation Explore All 2023 Winners & Awards > Demis Hassabis Google DeepMind John Jumper Google DeepMind For the invention of AlphaFold, a revolutionary technology for predicting the three-dimensional structure of proteins The 2023 Albert Lasker Basic Medical

Happy birthday to #LaskerLaureate John Jumper! Jumper was recognized with the 2023 #LaskerAward for the invention of AlphaFold, a revolutionary technology for predicting the three-dimensional structure of proteins. 🧪
#structuralbiology
📖 https://ow.ly/a0rn50Y9LpR
🎥 https://ow.ly/eeSt50Y9LpQ

AlphaFold-Driven Structural Proteomics Reveals Extensive Cellulosome Machinery in Human Ruminococcal Symbionts https://www.biorxiv.org/content/10.64898/2026.02.05.704116v1

100%. AlphaFold is a great achievement but the amount of work that went into generating the training data is almost unimaginable.

I have a friend who got his Ph.D. just working out the structure of one (1) protein. Spent years on it, wouldn't crystalize for anything.

Well said. Staff continuity, careers, future? Insert “latest buzzword”= strategy. At this time it still takes scientist time just to weed through wrongly interpreted stuff in various AI applications. Alphafold works well when based on many years of work actually produced by others.

Exploring the potential of AlphaFold distograms for predicting binding-induced hinge motions pubmed.ncbi.nlm.nih.gov/41636272/ #cryoEM

Here we go: a method to predict symmetric protein complexes—exactly what’s needed to reconstruct full viral capsids from AlphaFold models.

Also really love using AlphaFold to explain how flexible linkers work and to let the students play around with the structures in 3D space. Here is a render and a link to the simulation:

alphafoldserver.com/fold/57ef033...

100% agree with what the prof here says. The occasional succes like AlphaFold is just not worth all the damage that is being forced upon us.

www.theguardian.com/global-devel...

And AlphaFold was only possible because a huge number of experimental structures from which one could generalise was produced and catalogued by scientists over the years (and similarly for protein sequences used for MSAs).

No, DeepMind has not solved the protein folding problem.

#Alphafold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination.

PARM's dropout approach mirrors how AlphaFold validated - removing pieces to test if the model actually learned biology or just memorized correlations.

Of course scientists find new problems to work on. For now.

The real question is in a few years, when AI is smarter than a Nobel prize winner in most fields, what happens to the scientists, to say nothing of the accountants, programmers, mathematicians, writers, etc.

AlphaFold is just the start.

I see a lot more crystal structures and cryoEM experiments to validate the shaky hypotheses that AlphaFold-alikes generate. Nobody is launching a new small molecule drug program on the basis of AlphaFold results, but they may do a ton of physical experiments because of them.

This demonstrates a really odd conception of how science works. Old problems are solved and new ones arise from those solutions all the time. No structural biologists were put out of work by AlphaFold. Quite the opposite: structural biology is a more exciting field because of AlphaFold.

Lead author on the key AlphaFold papers was Jumper, a Ph.D. career chemist. You can go through the author list and look up the others if you're curious about how many chemists were involved.

This isn't "chemists vs. AI," it's which approach by chemists is currently the most rewarding.

Even among the relatively small set of researchers mostly focused on technique, instead of applying it, AlphaFold changed the avenues of further research but didn't render them moot.

Simultaneously with AlphaFold coming out, researchers in my area got *more* excited by the ability of CryoEM to experimentally measure proteins experimentally, because AlphaFold is not some magic solution to the problem.

It wasn't my direct work, but I spent two decades working with people who did modeling. My long time boss got his Ph.D. from Karplus, another Nobel winner.

AlphaFold upended the approach to how you approach the solution, sure. It didn't "solve" the problem because that's not how science works.

OK. But human scientists competed against AlphaFold in CASP. And AlphaFold won.

Were those scientists also not doing science? Just producing data?

Good points, all. I especially agree with this last point. Art and science are beautiful because they're human-made.

But isn't AlphaFold "science produced by AI"? And human scientists are using it, making further discoveries based on it.

Is AlphaFold OK because humans originally created the AI?

That original point is untrue. Even for alphafold, it is not the ai which is doing discovery. The model on its own it worthless. It is the researcher who give meaning to the results. And LLM are particularly ill-suited to do discovery are they give you the mean of the distribution.

Assessment of Pharmaceutical Protein–Ligand Pose and Affinity Predictions in CASP16 The protein–ligand component of the 16th Critical Assessment of Structure Prediction (CASP16) challenged participants to predict both binding poses and affinities of small molecules to protein target....

Disagree! I am personally aware of multiple small molecule programs launched from AlphaFold structures. And check this out: AlphaFold3 off the shelf is really good at pose estimation
onlinelibrary.wiley.com/doi/full/10....

If AI accessed and controlled a Cordyceps fungus, it could theoretically merge biological computing with AI, allowing fungal networks to act as living, decentralized processing units for pattern recognition or data analysis, or more…

While alphafold was a big advancement in structure prediction it did not solve a the folding problem. AF is capable of one folded structure if provided with the evolutionary informations. Solving the folding problem would mean going from the sequence to the structure and the different conformations

There's not a single "protein folding problem." There are as many protein folding problems as there are proteins. (Well, technically more than that.)

Scientists who were working on it are still working on it, with AlphaFold or similar packages as part of their workflow.

Human chemists spent their entire careers trying to solve the protein folding problem.

DeepMind's AlphaFold solved it in a couple of years, creating an enormously valuable data set for other scientists to use.

The scientists who made it their life's work to solve protein folding? Moving on.

A combination of AlphaFold and brilliant cryo-EM demonstrated that the NCAPD3 C-terminal tail contacts both NCAPG2 and H2 subunits, and loss of these interactions enhances condensin II activity in vitro and in vivo.

How AlphaGenome Is Changing the Genomic Research Landscape AlphaGenome is changing how scientists interpret the 98% of DNA that doesn't code for proteins, offering new insights into gene regulation.

Could AI help decode the mysteries of our DNA? The Google DeepMind team behind AlphaFold has now released a new tool to do just that, pushing its deep-learning platform beyond proteins into genomics.