AlphaFold2 Explained | Google’s DeepMind Solves Protein Folding

In 2020 DeepMind’s AlphaFold 2 solved a 50 year old Biology challenge.

INTERESTED IN AI? – Check out our list of the best AI newsletters here

This is one of the biggest advancements in AI and structural biology in decades.

In this video I will explain simply what this means and what advancements we can expect next.

First to understand is that there are Amino Acids, which are the basic building blocks of life.

Think of them like Lego Blocks.

With these Lego Blocks we can build proteins.

Proteins are both the building blocks of life, but also the machines that make life work.

It’s what allows us to see light, move, breathe and metabolize the things we eat.

Each protein has an intricate shape that defines what it does and how it works.

We know of over 200 millions of proteins and that number is going up as we discover more and more.

But humans have been really bad at understanding the exact 3d shape of all of these.

So if we are looking at a Lego castle, we know which lego pieces are used to build it and we know what it looks like, we just can’t build an exact replica of it.

Our ability to understand protein folding would allow us to understand the actual machinery that makes life function.

The sequence of amino acids is folded into unique and specific 3D structures which determines the protein’s function.

It’s important to understand that a string of amino acids can fold in many different ways.

That’s an understatement.

To give you an idea, the number of atoms in the universe is


Yet the number of ways a protein can fold is


So basically, it would take an enormously long time to go through all the possible configurations, Yet proteins can fold in seconds or less.

This is called “Levinthal’s Paradox”

So out of 200 plus million of proteins that we are aware of, we were able to determine the 3d structure of about 170,000 of them.

We did that using x-ray crystallography.

Which by the way was what Walter White from Breaking Bad was really good at.

His breakthroughs in x-ray crystallography is what allowed that multi billion dollar company “Grey Matter” to form.

That’s completely unrelated to anything here, I just thought I would include that.

Anyways, AlphaFold is able to take our existing data set of the proteins whose structure we know and then predict the 3d structure of other proteins that we do NOT know.

This is an important thing to understand.

Without AI, it takes us hundreds of thousands of dollars and years to map out these 3d structures.

It’s slow, expensive and probably limited to only some of the proteins.

With AI, we are able to have it predict new protein structures with ever increasing accuracy.

As it predicts more protein structures, our datasets grow and allows us to improve the AIs ability.

Why is this important?

How will we use our newfound understanding in the real world?

Well, the ability to understand protein function would allow us to know the unknown functions of proteins in our DNA.

As well as quickly design new proteins that alter the function of other proteins.

This would allow us to treat many diseases.

Create biomaterials in use in building and agriculture.

Tissue and organ regeneration as well as supplements for health and anti-aging.

In the long term it will allow us to create biological simulations and potentially lead to being able to engineer biological life.