**我国研究团队推出首个蛋白质预训练语言模型ESM-AA**
近日，来自清华大学AIR实验室、北京大学和南京大学的研究团队取得了蛋白质语言建模领域的重大突破，提出了首个从氨基酸到原子尺度的蛋白质预训练语言模型——ESM-AA模型。该模型的出色性能展示了多尺度统一建模在克服现有局限和挖掘新能力方面的巨大潜力。
ESM-AA模型不仅能处理氨基酸信息，还能处理原子信息，为蛋白质研究提供了全新的视角。其强大的整合多尺度信息能力，为研究不同生物结构间的相互作用开辟了新的道路。当前论文已被ICML 2024录用，引起了学界广泛关注。
蛋白质作为生命活动的关键执行者，其深入理解和建模对于生物科学具有重要意义。如何更好地理解与建模蛋白质也成为了目前AI4Science领域的一个研究热点。包括Deepmind、华盛顿大学Baker组在内的前沿研究机构也针对蛋白质全原子建模问题展开研究。
ESM-AA模型的推出，标志着我国在蛋白质语言建模领域取得了重要进展，为未来的蛋白质研究及相关领域应用提供了强有力的工具。基于ESM-AA模型，研究人员有望开发出与AlphaFold3、RoseTTAFold All-Atom相竞争的模型，进一步推动生物科学的发展。
英语如下：
News Title: “Tsinghua AIR and Other Research Teams Breakthrough the Boundaries of Protein Language Modeling: ESM-AA Model Leads a New Revolution”
Keywords: ESM-AA Model, Protein Language Modeling, Multi-scale Unified Modeling
News Content: **Chinese Research Team Launches the First Protein Pre-training Language Model ESM-AA**
Recently, research teams from Tsinghua University’s AIR Lab, Peking University, and Nanjing University have made significant breakthroughs in the field of protein language modeling, proposing the first protein pre-training language model ESM-AA, which covers from amino acid to atomic scale. The outstanding performance of this model demonstrates the enormous potential of multi-scale unified modeling in overcoming existing limitations and exploring new capabilities.
The ESM-AA model is not only capable of processing amino acid information but also atomic information, providing a brand-new perspective for protein research. Its powerful ability to integrate multi-scale information opens up new paths for studying the interactions between different biological structures. The current paper has been accepted by ICML 2024 and has attracted widespread attention in the academic community.
Proteins, as the key executors of life activities, their deep understanding and modeling are of great significance to the biological sciences. How to better understand and model proteins has become a research hotspot in the field of AI4Science. Leading research institutions including Deepmind and Baker group at the University of Washington have also conducted research on protein full-atom modeling.
The launch of the ESM-AA model marks significant progress in the field of protein language modeling in China, providing a powerful tool for future protein research and related applications. Based on the ESM-AA model, researchers are expected to develop models that compete with AlphaFold3 and RoseTTAFold All-Atom, further promoting the development of biological sciences.
【来源】https://www.jiqizhixin.com/articles/2024-06-28-2
Views: 4