人工智能科学初创公司 SimpleX 于 2024 年 11 月 10 日正式上线其首个技术产品 Tetra-1 的预览版 1。这款模型被定义为一种融合自然语言与蛋白质结构的推理系统,旨在通过解读“生命的语言”来加速科学发现进程 1。
在解决复杂生物学问题方面,Tetra-1 展现出显著的效率优势。以糖尿病靶点 PTP1B 的变构口袋发现为例,人类科学家从 1999 年确认靶点到 2004 年发现口袋耗时约五年,而 Tetra-1 仅需数小时即可完成类似的推理任务 1。此外,在 2024 年 9 月的测试中,该模型凭借推理能力超越了 AlphaFold3 和 IsoDDE 的得分,特别是在处理与已知结构最不相似的题目时表现尤为突出 1。
SimpleX 的核心团队由 AI 研究员、生物科学家以及 AI for Science 领域的专家共同组成 1。创始人将这一愿景概括为“We code life”,意在表达将生命转化为可被机器读懂的代码的目标 1。
AI for Science startup SimpleX released the preview version of its first technical product, Tetra-1, on November 10 1. The model is designed as a reasoning system that integrates natural language with protein structures, aiming to accelerate scientific discovery by understanding what the company describes as the "language of life" 1. Core team members include experts from three distinct fields: artificial intelligence research, biological science, and AI for Science applications 1.
Tetra-1 demonstrates significant advantages in solving complex biological problems, capable of completing reasoning tasks in hours that would take human scientists years 1. For instance, while it took researchers approximately five years—from confirming the target in 1999 to discovering the pocket in 2004—to identify the allosteric pocket of the diabetes target PTP1B, Tetra-1 accomplished this task in just a few hours 1. In performance tests conducted in September 2024, the model surpassed scores achieved by existing tools such as AlphaFold3 and IsoDDE through reasoning, showing particular strength on questions least similar to known structures 1. The founder’s vision for the technology is encapsulated in the phrase: "We code life," meaning they aim to turn life into readable code 1.
评论
还没有评论,欢迎留下第一条。