##中科院计算所团队提出CarbonNovo，基于AI进行蛋白质结构和序列的端到端从头设计
**北京** – 近日，中国科学院计算所张海仓带领的研究团队提出了 CarbonNovo，一种基于人工智能 (AI) 的蛋白质结构和序列端到端从头设计方法。该研究以「CarbonNovo: Joint Design of Protein Structure and Sequence Using a Unified Energy-based Model」为题发表在机器学习会议 ICML 2024 上。
蛋白质是生物执行功能的重要大分子，蛋白质从头设计旨在创造全新的蛋白质，在药物开发和酶工程中有着广泛的应用。近年来，基于 AI 的蛋白质从头设计快速发展，已被成功应用于抗体设计、小蛋白药物设计等领域，显著提高了设计成功率和效率。
CarbonNovo 的突破在于其采用了一种端到端的联合设计方法，将蛋白质主链结构和序列设计整合到一个统一的能量模型中。这与传统方法中分别进行结构设计和序列设计的“两阶段”框架不同，CarbonNovo 能够更有效地利用结构和序列之间的相互依赖关系，从而生成更稳定、更符合生物学功能的蛋白质。
CarbonNovo 的主要贡献包括：
* **端到端联合设计：** CarbonNovo 首次将蛋白质结构和序列设计整合到一个统一的能量模型中，以端到端的方式进行联合设计。
* **基于能量的生成模型：** CarbonNovo 利用基于能量的生成模型，将蛋白质的结构和序列信息与经典物理模型下的能量联系在一起，从而更准确地预测蛋白质的稳定性和功能。
* **蛋白质语言模型的引入：** CarbonNovo 首次将蛋白质语言模型引入蛋白质结构设计任务，利用海量天然蛋白质序列数据包含的先验信息，提高了设计效率和准确性。
CarbonNovo 的研究成果为蛋白质从头设计领域带来了新的突破，为药物开发、酶工程等领域提供了新的工具和方法。未来，该团队将继续探索蛋白质设计的新方法，推动蛋白质设计领域的发展。
英语如下：
##AI-Powered Protein Design: Breakthrough Achievement by CAS Institute of Computing Technology Team
**Keywords:** AI protein, end-to-end, carbon-based design
## CAS Institute of Computing Technology Team Proposes CarbonNovo, an AI-Based End-to-End De Novo Protein Structure and Sequence Design Method
**Beijing** – Recently, a research team led by Zhang Haicang from the Institute of Computing Technology, Chinese Academy of Sciences, has proposed CarbonNovo, anartificial intelligence (AI)-based end-to-end de novo protein structure and sequence design method. The research was published in the machine learning conference ICML 2024 under the title “CarbonNovo: Joint Design of Protein Structureand Sequence Using a Unified Energy-based Model.”
Proteins are essential macromolecules that execute biological functions. De novo protein design aims to create novel proteins, which has wide applications in drug development and enzyme engineering. In recent years, AI-based de novo protein design has rapidly developed and has been successfully applied in antibody design, small protein drug design, etc., significantly improving the design success rate and efficiency.
The breakthrough of CarbonNovo lies in its adoption of an end-to-end joint design method, integrating protein backbone structure and sequence design into a unifiedenergy model. Unlike the traditional “two-stage” framework where structure design and sequence design are performed separately, CarbonNovo can more effectively leverage the interdependence between structure and sequence, thereby generating more stable and biologically functional proteins.
The main contributions of CarbonNovo include:
* **End-to-End Joint Design:**CarbonNovo is the first to integrate protein structure and sequence design into a unified energy model for end-to-end joint design.
* **Energy-Based Generative Model:** CarbonNovo utilizes an energy-based generative model, linking protein structure and sequence information with energy under classical physical models, thus more accurately predicting proteinstability and function.
* **Introduction of Protein Language Model:** CarbonNovo is the first to introduce a protein language model into protein structure design tasks, leveraging prior information contained in massive natural protein sequence data to improve design efficiency and accuracy.
The research results of CarbonNovo have brought new breakthroughs to the field of de novoprotein design, providing new tools and methods for drug development, enzyme engineering, and other fields. In the future, the team will continue to explore new methods for protein design, promoting the development of the protein design field.
【来源】https://www.jiqizhixin.com/articles/2024-08-21-8
Views: 7