The biological world in HD: Google AI predicts the structure and interactions of all the molecules of life

Inside each cell there are billions of molecular machines and understanding how they work is key to understanding and treating diseases. The latest version of AlphaFold, an artificial intelligence system Googleis capable of predicting the structure and interactions of ‘all’ the molecules of life.

Its description is published in the journal Nature and, according to those responsible, AlphaFold 3 brings “the biological world to high definition.” It allows scientists to see cellular systems in all their complexity, through their structures, interactions and modifications.

According to DeepMind, responsible for this artificial intelligence (AI) together with Isomorphic Labs, it is a “revolutionary model” that improves the previous ones and works with unprecedented precision.

Inside every plant, animal and human cell there are billions of molecular machines that are made up of proteins, DNA and other molecules, but none of them work on their own. Only by seeing how they interact with each other, through millions of types of combinations, can you begin to truly understand the processes of life.

The new model is based on the foundations of AlphaFold 2, which in 2020 and the following years represented a ‘fundamental advance’ in the prediction of protein structure (in 2022, predictions of the three-dimensional structure of almost all proteins were published -200 million- from its amino acid sequence).

Millions of researchers around the world have used that version to make discoveries in areas such as malaria vaccines, cancer treatments and enzyme design, says a statement from Google DeepMind.

Beyond proteins

Substantial improvements to the deep learning architecture and training system now make it possible to more accurately predict the structure of a wide range of biomolecular systems in a unified framework.

For protein interactions with other types of molecules, it achieves an improvement of at least 50% compared to existing prediction methods, and for some important interaction categories the prediction accuracy has doubled.

“AlphaFold 3 takes us beyond proteins to encompass a broad spectrum of biomolecules. This leap could lead to more transformative science, from the development of biorenewable materials and more resistant crops to the acceleration of drug design and genomic research,” the note adds.

From a list of molecules, AlphaFold 3 is able to generate their joint three-dimensional structure, showing how they fit together. It models large biomolecules such as proteins, DNA and RNA, as well as small molecules, also known as ligands.

Additionally, it can model chemical modifications of these molecules that control the healthy functioning of cells and that, when altered, can cause disease.

This new window into the molecules of life reveals how they are all connected and helps understand how those connections affect biological functions, such as drug action, hormone production, and the DNA repair process that preserves health.

A ‘google maps’ of molecules, open

Scientists can access most of its features for free through the newly launched AlphaFold server. With a few clicks, they can harness the power of AlphaFold 3 to model structures composed of proteins, DNA, RNA, and a selection of ligands, ions, and chemical modifications.

“AlphaFold 3 has the potential to be as innovative as AlphaFold when it was first released. With the server, it is no longer just about predicting structures, but about generously facilitating access and allowing researchers to ask bold questions and accelerate discoveries,” says Céline Bouchoux, from the Francis Crick Institute.

“Understanding the biomolecular world within us and how the complex networks of molecules interact in our cells is a crucial starting point for understanding and treating diseases through rational drug design,” says Isomorphic Labs.

In this sense, and to advance this understanding, this innovative AI model has been developed that provides a precise vision at the atomic level of the structure of biomolecular systems, concludes Isomorphic, which is already in contact with companies in the sector for its implementation.