夜雨聆风
2026年5月7日，陈-扎克伯格基金资助的Theofanis Karaletsos团队在《Science》发表重磅成果，推出跨物种单细胞生成式基础模型TranscriptFormer。它像一位“生命进化翻译官”，用12个物种、1.12亿个细胞、15.3亿年进化数据训练，首次读懂跨越亿万年的细胞“通用语言”，为生命科学研究翻开新篇章。
什么是TranscriptFormer？简单说就是“细胞界超级解码器”
传统科研中，不同物种细胞差异巨大，比如人类和石珊瑚（分化6.85亿年）的细胞几乎没法直接对比，像鸡同鸭讲。而TranscriptFormer把基因表达当成“细胞句子”，用AI自监督学习，从海量跨物种细胞数据里，自学细胞运作的底层逻辑，不用人工标注，就能捕捉不同物种细胞的共性规律。
三大能力颠覆传统认知
1.跨物种“零障碍”识别：对和人类分开6.85亿年的石珊瑚，仍能精准分类细胞类型，性能远超现有模型；还能直接识别人类细胞的疾病状态，无需额外训练。
2.自动“脑补”生命规律：模型没学过发育过程、物种亲缘关系，却能自动“浮现”出细胞生长轨迹、物种进化脉络，就像自学成才读懂了生命之树 。
3.通用框架打通比较生物学：首次证明，从真菌到人类，所有生物的细胞组织都有“通用密码”，一套规律能解释亿万物种的细胞运作。
为啥重要？关乎疾病攻克与生命本质探索
1.医学革命加速器：未来研究人类疾病，不用只盯着人类细胞，可通过研究酵母、果蝇等简单生物的细胞，快速找到疾病机理；还能提前预测细胞病变，助力癌症、遗传病的早期诊断 。
2.生命本质的“解码器”：它证实生命演化虽复杂，但细胞底层逻辑高度统一，让我们从“看单个物种”升级到“俯瞰整个生命史”，重新理解生命起源与演化 。
3.科研效率“倍增器”：传统跨物种研究要数年，现在用TranscriptFormer可快速模拟实验、预测结果，大幅降低科研成本，加速全球生命科学研究进程 。
未来已来，AI重塑生命科学研究
TranscriptFormer不是终点，而是AI解码生命的新起点。作为开源工具，它将成为全球科学家的“虚拟实验室”，后续结合基因编辑、纳米技术，有望攻克更多医学难题，甚至实现“人工设计细胞”的愿景。
从解读15亿年细胞进化，到破解生命通用密码，TranscriptFormer证明，AI不仅能理解人类语言，更能读懂生命本身的“语言”，为人类探索生命奥秘、对抗疾病开辟全新路径。
Shocking Breakthrough! AI Model Decodes 1.5-Billion-Year Life History and Cracks the Universal Code of Cells
On May 7, the team led by Theofanis Karaletsos from the Chan Zuckerberg Initiative (CZI) published a landmark achievement in Science, unveiling a cross-species generative foundation model for single cells — TranscriptFormer. Acting as a "translator of evolutionary life", this model was trained on 112 million cells across 12 species covering 1.53 billion years of evolutionary history. For the first time, it decodes the universal "language" of cells spanning hundreds of millions of years, opening a new chapter in life science research.
What Is TranscriptFormer? Simply the "GPT of the Cell World"
In traditional research, cells vary drastically between different species. For example, human cells and stony coral cells — separated by 685 million years of evolution — can barely be compared directly, like speaking entirely different languages. Treating gene expression as "cell sentences", TranscriptFormer leverages AI self-supervised learning to independently grasp the underlying logic of cellular operation from massive cross-species cell data, capturing common patterns across species without manual labeling.
Three Capabilities Reshaping Traditional Cognition
1. Seamless cross-species recognition: It accurately classifies cell types even in stony corals that diverged from humans 685 million years ago, outperforming existing models by a large margin. It can also identify disease states in human cells without additional training.
2. Spontaneously deducing life laws: Though never trained on developmental processes or species kinship, the model spontaneously reveals cell developmental trajectories and evolutionary lineages, mastering the tree of life through self-learning.
3. Unified framework for comparative biology: It proves for the first time that cellular organization across all living beings, from fungi to humans, follows a universal set of rules.
Why This Matters: Advancing Disease Treatment and Exploring the Essence of Life
1. Accelerating medical innovation: Researchers can study simple organisms such as yeast and fruit flies to uncover mechanisms of human diseases, enabling early diagnosis of cancer and genetic disorders.
2. Decoding the essence of life: It confirms that despite complex evolution, all cells share unified underlying logic, upgrading research from single-species observation to a panoramic view of the entire history of life.
3. Boosting research efficiency: Cross-species studies that once took years can now be simulated and predicted rapidly, cutting research costs and speeding up global life science progress.
The Future Is Here，AI Reshapes Life Science Research
TranscriptFormer marks a new starting point for AI decoding life. As an open-source tool, it serves as a virtual laboratory for scientists worldwide. Combined with gene editing and nanotechnology, it is expected to conquer more medical challenges and even realize the vision of artificially designing cells.
From interpreting 1.5 billion years of cellular evolution to cracking life’s universal code, TranscriptFormer proves AI can understand not only human language but also the inherent language of life, blazing a new path for humanity to explore biological mysteries and fight diseases.
