瑞士洛桑联邦理工学院（EPFL）的研究人员近日在《Nature Communications》上发表了一篇研究论文，提出了一个全新的基于Transformer的逆向蛋白质序列设计方法。该方法利用深度学习技术，能够以比传统方法快10倍的速度设计出具有高热稳定性和催化活性的酶。
这项研究的核心是CARBonAra模型，它是一种基于几何Transformer的深度学习方法，能够根据蛋白质主链支架预测氨基酸序列。研究人员首先使用PDB数据库中的结构数据对CARBonAra模型进行训练，然后它可以预测给定主链支架所有位置的氨基酸置信度。这意味着CARBonAra能够处理任何类型的分子复合物，包括其他蛋白质、小分子等，从而在蛋白质设计过程中自然地考虑非蛋白质实体。
这项技术不仅能够提高蛋白质设计流程的多功能性，还能够实现所需的功能，为生物学、医学、生物技术和材料科学等领域带来重大影响。在蛋白质治疗药物的设计方面，定制蛋白质可能比小分子药物更具竞争力，有助于革新许多健康问题的治疗方式，从自身免疫疾病到癌症，提供更有效和个性化的治疗方案。
此外，CARBonAra模型还能够解决传统蛋白质设计模型在处理非蛋白质实体时的局限性，为酶功能的设计提供了一个全新的视角。通过设计新酶或改造现有酶，可以创造出促进自然界中罕见或不存在反应的催化剂，这对制药业和环保技术等多个行业都有深远影响。
总之，EPFL的研究人员通过CARBonAra模型展示了深度学习在蛋白质设计领域的巨大潜力，为未来的生物技术和医学研究开辟了新的道路。随着这一技术的不断成熟和应用，我们有理由相信，它将会在疾病的预防和治疗中发挥重要作用。
英语如下：
News Title: “New Breakthrough: Swiss Team Develops Fast Algorithm for Protein Sequence Design”
Keywords: Protein Design, Deep Learning, Enzyme Activity
News Content:
A research team from the École Polytechnique Fédérale de Lausanne (EPFL) recently published a study in Nature Communications, proposing a novel Transformer-based inverse protein sequence design method. This method utilizes deep learning technology to design enzymes with high thermal stability and catalytic activity ten times faster than traditional methods.
The core of this study is the CARBonAra model, a deep learning method based on geometric Transformers capable of predicting amino acid sequences based on protein backbone structures. The researchers trained the CARBonAra model using structural data from the Protein Data Bank (PDB) and then it can predict the confidence scores for all amino acid positions on a given backbone. This means CARBonAra can handle any type of molecular complex, including other proteins, small molecules, etc., thus naturally considering non-protein entities in the protein design process.
This technology not only enhances the multi-functionality of the protein design workflow but also realizes the desired functions, bringing significant impacts to fields such as biology, medicine, biotechnology, and materials science. In the design of protein-based therapeutic drugs, custom proteins may be more competitive than small molecules, contributing to the innovation of many treatment methods for health issues, ranging from autoimmune diseases to cancer, offering more effective and personalized treatment options.
Moreover, the CARBonAra model can address the limitations of traditional protein design models in handling non-protein entities, providing a new perspective for enzyme function design. By designing new enzymes or modifying existing ones, catalysts that promote reactions rare or non-existent in nature can be created, having profound impacts on industries such as pharmaceuticals and environmental technology.
In summary, the EPFL researchers have demonstrated the vast potential of deep learning in the field of protein design through the CARBonAra model, opening new paths for future biotechnology and medical research. As this technology matures and is applied, there is reason to believe that it will play a significant role in the prevention and treatment of diseases.
【来源】https://www.jiqizhixin.com/articles/2024-08-05-7
Views: 1