Google DeepMind and IsoMorphic Labs have just surprised the industry by releasing AlphaFold 3 and announcing how good it truly is.

It says AlphaFold 3 predicts the structure and interaction of all of life's molecules.

You can see here it states introducing AlphaFold 3 a new AI model developed by Google DeepMind and IsoMorphic Labs by accurately predicting the structure of proteins, DNA, RNA, ligands and more and how they interact.

We hope it will transform our understanding of the biological world and drug discovery.

You have to think about it like this, in every cell of every living thing, plant, animals, even us, there are literally billions of these microscopic machines.

These machines are made up of proteins, DNA and other funky molecules.

The thing is, is that none of these pieces work alone.

They're all kind of interacting, combining in millions of ways.

Only by seeing that interaction can we actually understand how life works.

This is where AlphaFold 3 comes in.

They literally just published a picture in Nature Magazine, which we'll get into later, in which they talk about this amazing new AI model.

This model can literally predict the structure of life's very molecules and even how they interact with crazy good accuracy.

We're actually talking a significant leap on what we could do before.

The crazy thing is, is that this is probably about to show us how all those crazy, crazy machines fit together, why they behave the way they do.

This is the key to understanding everything from diseases to how plants grow and how to fix problems or build new awesome stuff.

The details of this are pretty, pretty crazy.

It's pretty incredible that they've managed to build upon the success of AlphaFold 2, which was the one that changed the game in understanding proteins.

Researchers everywhere actually do use it.

Think malaria vaccines, cancer research, the entire whole deal.

AlphaFold's already been such a game changer.

It's even won major science awards.

AlphaFold 3, it actually goes way beyond just proteins to all sorts of biomolecules.

Think eco-friendly materials, stronger crops, supercharged medicine.

This is pretty much next level science with the power to change the world.

How does this actually work?

Given a list of input molecules, AlphaFold 3 generates their joint 3D structure, revealing how they all fit together.

It models large biomolecules such as proteins, DNA and RNA, as well as small molecules, also known as ligands, a category encompassing many different drugs.

They can model chemical modifications to these molecules, which control the healthy functioning of cells that when disrupted can lead to the disease.

We also see that AlphaFold 3's capabilities come from its next generation architecture and the training that now covers all of life's molecules.

At the core of the model is an improved version of the EVO former module, which is essentially a module that kind of learns the grammar of protein folding by studying evolutionary examples.

Uses that knowledge to predict the 3D structure of new amino acid sequences, much like how we can kind of predict the meaning of a new sentence after learning the grammar of a language.

It's a deep learning architecture that underpinned AlphaFold 2's incredible performance.

After processing the input, AlphaFold 3 assembles its predictions using a diffusion network akin to those found in AI image generators.

The diffusion process starts with a cloud of atoms and over many steps converges on its final most accurate molecular structure.

Here's where we have one of the predictions.

In this, well in this case, what we do have here is we have the ground truth shown in grey and then we have the real one shown in the actual colours.

In this example, what we do have here is we have us looking at the spike of a protein of a common cold virus.

The spike protein is a part of the virus that helps it infect our cells and the AI model actually accurately predicted how this spike protein interacts with antibodies, which are the immune system's defence proteins that attach the virus and neutralise it and of course simple sugars.

In this prediction that you're currently seeing on screen, the spike protein is in blue, the antibodies which try to stop the virus are shown in turquoise.

The simple sugars are also shown in yellow and these predictions essentially closely match what scientists have observed in real life experiments which are shown in grey.

By using animations of this interaction, scientists can then see exactly how the virus actually interacts with antibodies and sugars.

This information is crucial because understanding how these immune system processes helps us Figure out how to fight different viruses including COVID-19 leading to the potential of better treatments and AlphaFold 3 saves so much time by providing accurate predictions that would otherwise require lengthy and expensive laboratory equipments.

It does this because determining the 3D structure of proteins using experimental methods like x-ray crystallography or cryo-electron microscopy can take months or even years and AlphaFold 3 can predict these structures in literally hours or days and with predicted structures available, scientists can then focus on the most promising drug targets or biological questions without spending time exploring dead ends.

This allows researchers to test hypotheses so there's hypothesis generation where it can generate new hypotheses about how biological molecules function or interact and researchers can then test these hypotheses directly reducing the need for broad exploratory studies.

There was also another example here in which we can see Tim 3.

This is a protein that researchers are studying because it might be useful/a useful target for cancer treatments and scientists found a way to create small molecules that could stick to this protein and potentially block the harmful effects.

Before this study there weren't any clear images showing how these small drug-like molecules would actually fit into the Tim 3 structure and without this information designing effective drugs was pretty challenging.

Essentially AlphaFold 3 comes in.

Scientists use AlphaFold 3 to predict what Tim 3 would look like when these small molecules bind to it and they only provided the AI with the protein sequence which is essentially the recipe and a simple description of the drug-like molecules.

That's where we get to the outcome.

AlphaFold 3 actually predicted how these molecules would fit together perfectly into the pocket of the Tim 3 protein aligning almost exactly with the structures that scientists discovered through experiments.

It also showed that without these drug molecules present the pocket actually looked different proving that AlphaFold 3 can actually recognize changes in protein shapes when other molecules are around and this essentially means that AlphaFold 3 can predict how drug molecules will interact with proteins more accurately than ever before and this is going to lead to faster and more efficient drug discovery helping scientists develop better treatments for cancer and other diseases.

Of course it makes sense to talk about isoMorphic labs.

One of the things that they talk about leading the drug discovery at isoMorphic labs is that AlphaFold 3 is 50% more accurate than the best traditional methods on the Pose busters benchmark without needing the input of any structural information making AlphaFold 3 the first AI system to surpass physics-based tools for biomolecular structure prediction.

IsoMorphic labs is using AlphaFold 3 to accelerate and improve the success of drug design by helping understand how to approach new disease targets and developing novel ways to pursue existing ones that were previously out of reach.

They also talk about how we can now create and test hypotheses at the atomic level and produce highly accurate structure predictions within seconds standing in stark contrast to the months or even years required to experimentally determine answers to similar questions.

They also talk about how they're using this.

They state that already we are using AlphaFold 3 day to day.

Our scientists have seen that designing small molecules against AlphaFold 3's structural predictions helps create designs that bind effectively to a target protein.

The improved structural accuracy of protein-to-protein interfaces with AlphaFold 3 opens up the possibility of designing new treatment modalities, such as antibodies or other therapeutic proteins.

A richer understanding of a novel target can be achieved by looking at the structure of targets in their full biological context, in complex with other protein binding partners, DNA, RNA, and ligand cofactors.

Of course, there was a paper that was released with the AlphaFold 3 release.

However, this is currently an Accelerated Article Preview (an AAP), which is a version of a scientific article that's published online before it appears in a traditional print journal.

It basically just means that it's designed to get important research findings out to scientists and the public as quickly as possible.

They kind of skip some of the editing and formatting that can actually slow down a traditional journal publication.

That means there is a watermark on each page to remind you that these aren't the final polished versions.

Here's where we actually talk about the main thing, which is, of course, the AlphaFold server.

Google actually launched this tool that helps you do this stuff, so scientists can use it for free.

There's no fancy subscription needed.

Basically, with a few clicks, any biologist can use AlphaFold 3 to make models of proteins, DNA, RNA, and other important molecules.

This is pretty huge because the AlphaFold server lets scientists quickly come up with new ideas to test out in the lab.

This cuts down the time wasted on guesswork.

Just picture your science projects going way faster, and remember, this used to take a huge amount of time.

All of that is being saved.

This is the quick trailer I'll leave you with, and if you need any of the links, they'll be either in a top comment or a link in the description.