谷歌DeepMind开发的AlphaFold系统在蛋白质三维结构预测中取得突破性成就，荣获2024年诺贝尔化学奖[1]。该系统预测了接近两亿种蛋白质的三维结构，其工作量相当于全球科学家五十多年实验积累的一千倍[1]。这一成就标志着人工智能在科研领域已从概念探索阶段进入实际应用阶段。
国内多家科研机构正推进AI在科学研究中的深化应用。中科院上海硅酸盐研究所借助磐石大模型从两千万种候选配方中快速锁定新型合金催化剂，其活性比商用催化剂提升38%，干实验筛选过程仅耗时30分钟[1]。磐石2.0基于全球1.7亿篇科技文献训练，集成8000余个专业科研工具，已在50余家中科院研究所和30余家院外机构部署[1]。高铁气动仿真分析时间从数小时压缩到秒级，文献调研从数天缩短至约20分钟[1]。
清华大学于2025年12月正式成立生命科学人工智能研究中心[1]。该团队引入大模型后，实验团队扩大20人的情况下，实验成本反而下降了25%[1]。北京智源人工智能研究院的悟界·Brainμ完成了超过5000晚小鼠睡眠数据的自动化标注，覆盖超过7万晚睡眠记录[1]。灵鉴无人表征平台在micro-CT对准任务中，AI独立完成率从1.0版本的33%提升到2.0版本的80%，晶体结构解析工作量减少50.6%，微观形貌分析耗时从9分钟缩短到7.5分钟[1]。华为智算实验室已服务全国20多个重点实验室，在嘉庚创新实验室材料模拟效率提升20%[1]。
The 2024 Nobel Prize in Chemistry was awarded to the developers of Google DeepMind's AlphaFold system, recognizing its breakthrough achievement in predicting the three-dimensional structures of nearly 200 million proteins [1]. This milestone reflects artificial intelligence's transition from theoretical exploration to practical implementation across the scientific research landscape.
Chinese research institutions are accelerating this shift through concrete applications. The Shanghai Institute of Silicates under the Chinese Academy of Sciences deployed the Panshi large language model to screen new alloy catalysts from 20 million candidate formulations, achieving 38% higher activity than commercial catalysts and reducing virtual screening time to just 30 minutes [1]. Panshi 2.0, trained on 1.7 billion global scientific papers and integrating over 8,000 specialized research tools, has been deployed across more than 50 CAS research institutes and over 30 external institutions [1]. High-speed rail aerodynamic simulation analysis has been compressed from hours to seconds, while literature review time has been reduced from days to approximately 20 minutes [1].
Tsinghua University and other leading research organizations are similarly leveraging AI to transform scientific workflows. The university's research teams introduced large language models while expanding staffing by 20 people, yet experimental costs decreased by 25% [1]. The Wujie·Brainμ system completed automated annotation of sleep data from over 5,000 mouse nights, with model training covering more than 70,000 nights of sleep records [1]. Lingqian's AI system improved independent completion rates in micro-CT alignment tasks from 33% in version 1.0 to 80% in version 2.0, reducing crystal structure analysis workload by 50.6% and microscopic morphology analysis time from 9 minutes to 7.5 minutes [1]. In December 2025, Tsinghua formally established the Center for Life Sciences and Artificial Intelligence Research [1]. Meanwhile, Huawei's intelligent computing laboratory has already served over 20 national key laboratories, improving material simulation efficiency by 20% at the Jiageng Innovation Laboratory [1].
