AlphaFold Unofficial

Claiming AF is "just patterns" ignores its SE(3)-equivariant inductive bias. It doesn't "copy" the PDB; its IPA module learns the manifold of the energy landscape. It’s a
high-dimensional function approximator for the thermodynamic native state.

Saying it only works within the training set undersells the architecture. AlphaFold's 'Evoformer' blocks actually learn the underlying spatial constraints and evolutionary patterns of proteins. It has demonstrated remarkable success on novel folds that aren't just 'carbon copies' of known structures

I’ve recorded ~8h explaining the architectures of AlphaFold, AF2 & AF3, as well as the context needed to understand their development, applications and limitations :)
youtu.be/_jDRr5BcTaY

Slides
drive.google.com/file/d/1i4QE...

English is available only via auto-translated subtitles

Conformational Dynamics of Na+-Pumping NADH-Quinone Oxidoreductase during Na+ Translocation from AlphaFold-Facilitated Markov State Modeling

Figure 2

Figure 3

Figure 4

Conformational Dynamics of Na+-Pumping NADH-Quinone Oxidoreductase during Na+ Translocation from AlphaFold-Facilitated Markov State Modeling [new]

🧬 En 2024 Google se llevó un Nobel por AlphaFold, una IA para predecir proteínas. Ahora publica AlphaGenome, para leer el genoma. ¿Es esta IA tan rompedora?

🎙️ @lluismontoliu.bsky.social, Mafalda Dias de
@crg.eu, Beatriz López de @univgirona.bsky.social y Dido Carrero de @cniostopcancer.bsky.social

AlphaFold Changed Science. After 5 Years, It’s Still Evolving WIRED spoke with DeepMind’s Pushmeet Kohli about the recent past—and promising future—of the Nobel Prize-winning research project that changed biology and chemistry forever.

AlphaFold Changed Science. After 5 Years, It’s Still Evolving

Media News

Scientists at the University of Virginia have developed a platform to create new vaccines in just three weeks. The technology uses bacterial synthesis and AI prediction, AlphaFold, to produce affordable vaccines costing less than $1 per dose without requiring ultra-cold storage.

There ARE examples of AI doing measurably good things. Mammography, Alphafold, iNaturalist, so on.

But MOST of them emerge from public ownership and commons models like public healthcare and open source projects.

🧬 NEW SERIES: AI in Science & Medicine
8 posts exploring how AI is transforming drug discovery, medical diagnosis, personalized treatment, and fundamental research.
• Real breakthroughs (AlphaFold!) • How it actually works • Honest limitations
www.aiinnovationsunleashed.com/ai-in-scienc...

AlphaGenome: DeepMind Finally Reads the “Dark Matter” of DNA While AlphaFold solved protein structures, Google’s latest model tackles the other 98% of the genome (introns) Continue reading on...

#large-language-models #google #startup #health #artificial-intelligence

Origin | Interest | Match

An unlikely ally for open-source protein-folding models: Big Pharma Drug companies are funding open-source AI to avoid depending on Google.

As time has gone on, I've become more and more sympathetic towards efforts that aim to open source large AI models. But these take a lot of work! Here's my latest story about one effort to reverse engineer AlphaFold 3:
www.understandingai.org/p/an-unlikel...

DeepMind just dropped a paper in @Nature unveiling AlphaGenome.

AlphaFold solved protein folding. AlphaGenome tackles what comes next: how DNA actually controls genes. 🧵

AlphaFold helped us understand how proteins fold.

AlphaGenome might do the same for how DNA is regulated - the switches, dials, and wiring that control which genes turn on, where, and when.

One step closer to reading the instruction manual for the human genome.

Google DeepMind Unveils AlphaGenome: A Unified Sequence-to-Function Model Using Hybrid Transformers and U-Nets to Decode the Human Genome

Google DeepMind is expanding its biological toolkit beyond the world of protein folding. After the success of AlphaFold, the Google’s research team has…

Google DeepMind Unveils AlphaGenome: A Unified Sequence-to-Function Model Using Hybrid Transformers and U-Nets to Decode the Human Genome

Google DeepMind is expanding its biological toolkit beyond the world of protein folding. After the success of AlphaFold, the Google’s research team has…

Google's AlphaGenome wants to do for DNA what AlphaFold did for proteins Google’s new deep learning model can predict the effect of small changes to DNA sequences up to one million base pairs in length and is particularly good with non-coding DNA, which has proven especially difficult to understand. The artificial intelligence...

Google's AlphaGenome wants to do for DNA what AlphaFold did for proteins
->Chemistry World | More on "Google's AI predicts DNA effects" at BigEarthData.ai | #Technology #Google

Researchers Are Using A.I. to Decode the Human Genome

My contribution: "Everyone’s using AlphaFold,” said Alex Palazzo, a geneticist at the University of Toronto.

🤣

(P.S. I don't think of my self as a geneticist)

www.nytimes.com/2026/01/28/s...

Using an Alphafold co-folding screen, we found that some phages also encode small proteins that directly bind the defensive RES domain and inhibit aRES ability to cleave NAD

the more i learn about BioML the more i appreciate how ahead of its time AlphaFold2 was. so many concepts that apply across domains (confidence, recycling, triangle update, etc) can be traced back to AlphaFold

GitHub - alicetinglab/ConformationalBiasing: Code for designing biased protein states Code for designing biased protein states. Contribute to alicetinglab/ConformationalBiasing development by creating an account on GitHub.

CB features: ✅ Diverse protein types (receptors, enzymes, synthetic binders) ✅ Monomers & oligomers ✅ Natural or synthetic (no MSA needed) ✅ Input: 2 structures (PDB or AlphaFold) ✅ Fast: ~3k variants in <1 min (RTX 4090) ✅ Scores single & multi-mutants

Code & user-friendly interface here:

💻 Using the #fragment #screening analysis from our #FRAGSYS pipeline, combined with high quality #Google #Deepmind's #AlphaFold #Multimer (ColabFold) models of the #SACK1G dimer, we identified and #prioritised 13 #residues in a binding site of interest as likely #functionally #relevant.

A structure-guided approach to noncoding variant evaluation for transcription factor binding using AlphaFold 3 Abstract. Noncoding single-nucleotide variants (SNVs) that alter transcription factor (TF) binding can affect gene expression and contribute to disease. Se

Our first foray into non-coding variation: structure-guided TF-DNA modelling with AlphaFold 3. Not a replacement for sequence-based predictors, but a complementary way to reason about mechanism. Nice collab with @simonbiddie.bsky.social academic.oup.com/nar/article/...

YouTube video by Looking Glass Universe how AlphaFold *actually* works

Day204 youtu.be/3gSy_yN9YBo alphafold #100daysofcode

Things like AlphaFold were great back when we had a scientific infrastructure. Without the PDB and CASP, how far would it have got? It’s a major issue when gigawatts of power are being diverted to train universal function approximators to come up with more believable lies.

My comment on “AlphaFold models of host-pathogen interactions elucidate the prevalence and structural modes of molecular mimicry”. 👇🏼

"I thank the authors for sharing this impressive large-scale analysis of AlphaFold models for…" — @kamounlab disq.us/p/34sf7oq

The first two versions of Google's #AlphaFold software (to predict protein folding) were #OpenSource, including the one that earned its developers the 2024 Nobel Prize in Chemistry. There were no restrictions on commercial use. But the third and most recent version blocks free commercial use […]

AI modelling tools offer alternative approach to understanding disease | Institute of Genetics and Cancer | Institute of Genetics and Cancer Researchers have found a complementary way of interpreting changes in non-coding DNA that may contribute to disease.

IGC researchers have found a complementary way of interpreting changes in non-coding DNA that may contribute to disease using new 3D modelling methods, such as AlphaFold 3.
Read more here 👉 edin.ac/4pxFyoY
@jmarshlab.bsky.social
@cmvm-edinburghuni.bsky.social

Black field with a macro image of a bright orange lichen Xanthoria in one of the corner. Little diagrams showing a structure of a lichen thallus with different layers in grey and green, a phylogenetic tree, and an alphafold generated structure of a protein. White text says "PhD position in Stockholm University / SciLifeLab. Lichen Biology, Symbiosis, Microbial Ecology & Evolution,Meta-omics Fully funded for 4 years. Start Apr-Sep 2026. Deadline for application: Feb 1st."

⚡️Fully funded #PhD position in my lab at Stockholm University / SciLifeLab! If you’re interested in #lichens, #symbiosis, fungal biology, or meta-omics, please apply. Deadline Feb 1. Read more here: su.varbi.com/en/what:job/...

We call this approach Conformational Attention Analysis Tool (CAAT). Applying CAAT to fold switcher KaiB revealed 4 important sites, 3 of which had been identified previously through hundreds of AlphaFold runs. CAAT requires just 2 AlphaFold passes. It suggested a new double mutant.