Chai-1:解码生命的分子相互作用 Chai-1: Decoding the molecular interactions of life

原始链接: https://www.chaidiscovery.com/blog/introducing-chai-1

简介:Chai-1是一个尖端的、多用途的人工智能模型,专为分子结构预测而设计。 该工具在药物发现的各个方面都超越了当前的行业标准。 它不仅可以预测蛋白质,还可以预测小分子、DNA、RNA、化学修饰等。 用户可以通过用户友好的网站访问 Chai-1,该网站提供商业和非商业选项。 对于那些希望将其用于个人项目的人,源代码也将被发布。 主要特点包括: - 在 PoseBusters 基准测试和 CASP15 数据集等标准化测试中优于其他模型。 - 与许多类似工具不同,无需多重序列比对 (MSA) 即可有效运行。 - 预测复杂蛋白质排列(多聚体)的卓越能力。 额外功能:当提供额外的实验数据时,Chai-1 可以显着提高其性能,在某些情况下可以观察到两位数的改进。 一个例子是“表位调节”,即使添加有限数量的接触或口袋残基细节也能显着改善抗体-抗原结构预测,促进人工智能驱动的抗体设计。 可用性:用户可以通过访问 [www.lab.chaidiscovery.com](http://www.lab.chaidiscovery.com) 或从 GitHub 存储库 [github.com/chaidiscovery] 下载来亲自尝试 Chai-1 /chai-lab](https://github.com/chaidiscovery/chai-lab)。 未来的工作:Chai-1 的创建者旨在通过开发更先进的人工智能模型来彻底改变生物学领域,这些模型能够预测和操纵生物分子(生命的基本单位)之间的相互作用。 进一步的进展将很快分享。 支持与协作:Chai-1 的开发得益于 Dimension、Thrive Capital、OpenAI、Conviction、Neo、Lachy Groom、Amplify Partners、Anna 和 Greg Brockman、Blake Byers、Fred Ehrsam、Julia 和 Kevin 等合作伙伴的支持 哈茨、威尔·盖布里克、大卫·弗兰克尔、R·马丁·查韦斯以及许多其他人。

Introduction: Chai-1 is a cutting-edge, multi-purpose artificial intelligence model designed for molecular structure prediction. This tool surpasses current industry standards in various aspects of drug discovery. It can predict not just proteins but also small molecules, DNA, RNA, chemical modifications, and more. Users can access Chai-1 through a user-friendly website, with both commercial and non-commercial options available. For those who wish to use it for personal projects, the source code will also be released. Key features include: - Outperforming other models on standardized tests like the PoseBusters benchmark and CASP15 dataset. - Able to operate effectively without multiple sequence alignments (MSAs), unlike many similar tools. - Superior ability in predicting complex protein arrangements (multimers). Additional Capabilities: When provided with additional experimental data, Chai-1 can significantly enhance its performance, with double-digit improvements observed in certain cases. One example is the 'epitope conditioning', where adding even a limited number of contact or pocket residue details substantially improves antibody-antigen structure predictions, facilitating AI-driven antibody design. Availability: Users can try Chai-1 out for themselves either by visiting [www.lab.chaidiscovery.com](http://www.lab.chaidiscovery.com) or downloading it from the GitHub repository at [github.com/chaidiscovery/chai-lab](https://github.com/chaidiscovery/chai-lab). Future Work: The creators of Chai-1 aim to revolutionize the field of biology by developing more advanced AI models capable of predicting and manipulating interactions among biological molecules - the basic units of life. Further developments will be shared soon. Support & Collaboration: The development of Chai-1 was made possible thanks to support from partners like Dimension, Thrive Capital, OpenAI, Conviction, Neo, Lachy Groom, Amplify Partners, Anna and Greg Brockman, Blake Byers, Fred Ehrsam, Julia and Kevin Hartz, Will Gaybrick, David Frankel, R. Martin Chavez, and numerous others.


We’re excited to release Chai-1, a new multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of tasks relevant to drug discovery. Chai-1 enables unified prediction of proteins, small molecules, DNA, RNA, covalent modifications, and more.

The model is available for free via a web interface, including for commercial applications such as drug discovery. We are also releasing the model weights and inference code as a software library for non-commercial use.

A frontier model for biomolecular interactions

We tested Chai-1 across a large number of benchmarks, and found that the model achieves a 77% success rate on the PoseBusters benchmark (vs. 76% by AlphaFold3), as well as an Cα LDDT of 0.849 on the CASP15 protein monomer structure prediction set (vs. 0.801 by ESM3-98B).

Unlike many existing structure prediction tools which require multiple sequence alignments (MSAs), Chai-1 can also be run in single sequence mode without MSAs while preserving most of its performance. The model can fold multimers more accurately (69.8%) than the MSA-based AlphaFold-Multimer model (67.7%), as measured by the DockQ acceptable prediction rate. Chai-1 is the first model that’s able to predict multimer structures using single-sequences alone (without MSA search) at AlphaFold-Multimer level quality.

For more information, and a comprehensive analysis of the model, read our technical report.

A natively multi-modal foundation model

In addition to its frontier modeling capabilities directly from sequences, Chai-1 can be prompted with new data, e.g. restraints derived from the lab, which boost performance by double-digit percentage points. We explore a number of these capabilities in our technical report, such as epitope conditioning – using even a handful of contacts or pocket residues (potentially derived from lab experiments) doubles antibody-antigen structure prediction accuracy, making antibody engineering more feasible using AI.

Releasing the model for all

We are releasing Chai-1 via a web interface for free, including for commercial applications such as drug discovery. We are also releasing the code for Chai-1 for non-commercial use as a software library. We believe that when we build in partnership with the research and industrial communities, the entire ecosystem benefits.

Try Chai-1 for yourself by visiting lab.chaidiscovery.com, or run it from our GitHub repository at github.com/chaidiscovery/chai-lab.

What's next?

The team comes from pioneering research and applied AI companies such as OpenAI, Meta FAIR, Stripe, and Google X. Collectively, we have played pivotal roles in the advancement of research in AI for biology. The majority of the team has been Head of AI at leading drug discovery companies, and has collectively helped advance over a dozen drug programs. 

Chai-1 is the result of a few months of intense work, and yet we are only at the starting line. Our broader mission at Chai Discovery is to transform biology from science into engineering. To that end, we'll be building further AI foundation models that predict and reprogram interactions between biochemical molecules, the fundamental building blocks of life. We’ll have more to share on this soon.

We are grateful for the partnership of Dimension, Thrive Capital, OpenAI, Conviction, Neo, Lachy Groom, and Amplify Partners, as well as Anna and Greg Brockman, Blake Byers, Fred Ehrsam, Julia and Kevin Hartz, Will Gaybrick, David Frankel, R. Martin Chavez, and many others.

相关文章
联系我们 contact @ memedata.com