AlphaFold 3 is launched, comprehensively predicting the interactions and structures of proteins and all living molecules, with far greater accuracy than ever before

Editor | Radish Skin

Since the release of the powerful AlphaFold2 in 2021, scientists have been using protein structure prediction models to map various protein structures within cells, discover drugs, and map each A "cosmic map" of known protein interactions.

Just now, Google DeepMind released the AlphaFold3 model, which is capable of joint structure prediction of complexes including proteins, nucleic acids, small molecules, ions and modified residues. The accuracy of

AlphaFold3 is significantly improved compared to many dedicated tools in the past (protein-ligand interaction, protein-nucleic acid interaction, antibody-antigen prediction). This demonstrates that high-accuracy modeling across the biomolecule space is possible within a single unified deep learning framework.

Meanwhile, the team has newly launched AlphaFold Server, an easy-to-use research tool that provides free access to most of the features of AlphaFold3.

Frank Uhlmann, a biochemist at the Francis Crick Institute in London, came across AlphaFold3 early on and was impressed by its capabilities. "This is simply revolutionary." He said, "This will democratize structural biology research."

The study is titled "Accurate structure prediction of biomolecular interactions with AlphaFold 3" and was released on May 8, 2024 In "Nature".

1. Molecular machines in cells: Cells contain billions of molecular machines composed of proteins, nucleic acids and sugars. These machines work together so that the processes of life proceed smoothly.

2. The Revolution of AlphaFold3: Google DeepMind launches AlphaFold3, which predicts the structure and interactions of all living molecules with unprecedented accuracy.
3. Interactions with other molecule types: Compared with existing methods, AlphaFold3 improves the accuracy of predictions of protein interactions with other molecule types by at least 50%, and even doubles the accuracy in some categories.
4. Updates to AlphaFold3: DeepMind has made a major update to AlphaFold3 that reduces the reliance on target sequence-related protein information.
5. Usage of Diffusion Model: AlphaFold3 uses a diffusion model machine learning network, the same one used in the artificial intelligence image generator Midjourney. "This is a major change," said John Jumper, AlphaFold project leader.

   

   How AlphaFold3 reveals the molecules of life

Video link: https://mp.weixin.qq.com/s/IcKOIAqEM4B2fdOsE5cfVA

AlphaFold3 Features

Given Input list of molecules, AlphaFold3 Their joint 3D structure is generated, revealing how they fit together. It can simulate:

1. Large biomolecules (proteins, DNA, RNA)
2. Small molecules (ligands)
3. Chemical modifications

Technical advantages

AlphaFold3 uses improved architecture and training to cover all life molecules . Core technologies include:

1. Evoformer module
2. Diffusion network assembly

Accuracy

AlphaFold3’s prediction of molecular interactions exceeds the accuracy of all existing systems, with a unique ability to unify scientific insights.

Drug Discovery

AlphaFold3 improves drug design capabilities and can predict:

1. ligands
2. antibodies

It achieves unprecedented accuracy in predicting drug-like interactions, improving by 50% over traditional methods %, becoming artificial intelligence systems that transcend physics-based prediction tools.

Isomorphic Labs

Isomorphic Labs combines AlphaFold3 with in-house AI models for:

1. Internal projects
2. Drug design with pharmaceutical partners

It leverages AlphaFold3 to accelerate and improve the success rate of drug design, cope with New disease targets and existing targets.

Balance

DeepMind balances AlphaFold3’s accessibility, scientific impact and commercial drug discovery capabilities.

Jim Fan, senior research manager at NVIDIA and head of Embodied AI (GEAR Labs), tweeted and commented:

1. "AlphaFold3 is here, the latest version of the greatest breakthrough in artificial intelligence in biology. Novelty AlphaFold3 uses diffusion to "render" molecular structures. It starts with a blurry cloud of atoms and then gradually materializes the molecules through denoising. "
2. " In the timeline we live in, learning from Llama and Sora can inform and accelerate the life sciences. I find the generality absolutely incredible. The same transformer+diffusion backbone that generates beautiful pixels can too Think of proteins, as long as you convert the data to a floating point sequence accordingly. "
3. "We haven't reached the level of a single AGI model, but we have successfully built a universal AI recipe menu that can transfer training, data and neural networks across domains. Architecture. This shouldn't work, but thank God it does!"

Of course, there are also dissenting voices, due to limitations in modeling protein interactions with possible drugs, said Brian, a medicinal chemist at the University of California, San Francisco. "I don't see it having the same impact as AlphaFold2," said Shoichet, who has been using the AlphaFold structure to find drug candidates.

A free and easy-to-use research tool

Unlike RoseTTAFold and AlphaFold2, scientists will not be able to run their own version of AlphaFold3, and information such as the underlying code of AlphaFold3 will not be made public at this time. Instead, researchers will have access to DeepMind’s new “AlphaFold Server” to use AlphaFold3’s capabilities.

AlphaFold Server is a free platform that scientists around the world can use to conduct non-commercial research. With just a few clicks, biologists can harness the power of AlphaFold 3 to simulate structures composed of proteins, DNA, RNA, and a range of ligands, ions, and chemical modifications.

AlphaFold Server helps scientists generate novel hypotheses and test them in the lab, speeding up workflow and enabling further innovation. The platform provides researchers with a convenient way to generate predictions, regardless of whether they have computing resources or expertise in machine learning.

Uhlmann likes what he has seen of the server so far, it is simpler and faster than the AlphaFold2 version he had previously used. "You upload it, and 10 minutes later, you have the structure," he said.

However, currently the number of users’ access to AlphaFold Server is limited. Currently, scientists can only make 10 predictions per day, which makes it unlikely to obtain protein structures that bind possible drugs.

Video link:
https://mp.weixin.qq.com/s/IcKOIAqEM4B2fdOsE5cfVA

AlphaFold Server:
https://golgi.sandbox.google.com/about

Responsible Harnessing the power of AlphaFold3

With each version of AlphaFold, DeepMind works with the research and security communities to understand the technology’s broad impact. They take a science-led approach and conduct extensive assessments to mitigate potential risks and share broad benefits to biology and humans.

Building on DeepMind’s external consultation for AlphaFold2, in addition to professional third parties in biosecurity, research and industry, they have now engaged with more than 50 domain experts to understand the capabilities and capabilities of subsequent AlphaFold models. any potential risks. DeepMind also participated in community-wide forums and discussions ahead of the release of AlphaFold3.

AlphaFold Server reflects DeepMind’s ongoing commitment to sharing the benefits of AlphaFold, including a free database of 200 million protein structures.

They will also work with EMBL-EBI and in partnership with organizations in the Global South to expand free AlphaFold educational online courses to provide scientists with the tools they need to accelerate adoption and research, including in underfunded areas such as neglected diseases and food security. DeepMind will continue to work with the scientific community and policymakers to develop and deploy AI technologies responsibly.

Unlocking the future of AI-driven cell biology

AlphaFold3 brings the biological world into HD. It enables scientists to understand cellular systems in all their complexity, structure, interactions and modifications.

This new window into the molecules of life reveals how they are interconnected and helps understand how these connections influence biological functions - such as the action of drugs, the production of hormones, and the health-protecting process of DNA repair.

The potential of AlphaFold3 has just begun to be tapped. What is the future of life sciences?

Paper link:
https://www.nature.com/articles/s41586-024-07487-w

関連コンテンツ:
https://blog.google/technology/ai/google-deepmind-isomorphic-alphafold-3-ai-model/

The above is the detailed content of AlphaFold 3 is launched, comprehensively predicting the interactions and structures of proteins and all living molecules, with far greater accuracy than ever before. For more information, please follow other related articles on the PHP Chinese website!