Linn
One-Loop Lab

Make age a choice.

Turning back the clock of the cell.

Linn Biosciences builds AI virtual cells that propose interventions to move aged cells toward a younger state. Outsourced CRO labs test those candidates; results feed back and the model learns. One person runs the loop, with AI agents handling much of the research and coordination. The target is biological age. Skin is the first window where evidence can become visible.

Follow the research
一人闭环实验室

让年龄，成为选择。

把细胞的时钟往回拨。

林恩生物（Linn Biosciences）用 AI 虚拟细胞提出可能让老化细胞回到更年轻状态的干预方案；外包 CRO 实验室做验证；结果回流，模型更新。一个人统筹这条闭环——大量研究与协调由 AI 智能体完成，湿实验外包完成。我们要改变的是细胞的生物学年龄；皮肤是第一个看得见证据的窗口。

关注研究进展

You have two ages. We ask if one can move.

One age is on your ID. The other is carried by your cells. Scientists can estimate the second from chemical marks on DNA called methylation—often called methylation age (Horvath, 2013). It tracks calendar age closely, but the two are not always the same. Linn’s research question is whether that cellular clock can be turned back—starting with human dermal fibroblasts, where readouts are accessible and evidence can appear on a useful timescale.

Source: Horvath S. Genome Biology 2013;14(10):R115. genomebiology.biomedcentral.com

你有两个年龄。我们问：其中一只钟能不能动？

一个写在身份证上；另一个藏在细胞里。科学家已经可以根据 DNA 上的化学标记（甲基化）估算细胞的「甲基化年龄」（Horvath, 2013）。它和日历年龄高度相关，但并不总是完全一致。林恩生物要回答的研究问题是：细胞里的这只钟，能不能往回拨？我们从人皮肤成纤维细胞开始——读数周期短，证据更容易看见。

出处：Horvath S. Genome Biology 2013;14(10):R115. genomebiology.biomedcentral.com

How we validate

Three layers, designed to keep prediction honest before anything reaches a wet lab:

  1. 1. Virtual cell Models predict how a genetic intervention changes the transcriptome of aged human dermal fibroblasts—toward or away from a younger expression state.
  2. 2. Held-out accuracy RejuvBench scores those predictions on donors and ages the model has not seen (leave-one-donor-out), so we measure generalization, not memorization.
  3. 3. Wet-lab check Promising candidates go to outsourced CRO labs. Results return, update the model, and set the next round.

Skin fibroblasts are the evidence window—not a beauty product line. Behind the window is biological age itself. Candidates that erase cell identity or drive uncontrolled growth are screened out by design.

我们怎么验证

三层，先保证预测诚实，再进入湿实验：

  1. 1. 虚拟细胞 模型预测基因干预如何改变老年皮肤成纤维细胞的转录组——是更接近、还是更远离年轻状态。
  2. 2. 留出精度 RejuvBench 在模型没见过的供体与年龄上打分（留一供体交叉验证），测的是泛化，不是背题。
  3. 3. 湿实验复核 有希望的候选交给外包 CRO；结果回流，更新模型，进入下一轮。

皮肤成纤维细胞是证据窗口——不是美容产品线。窗户后面，才是生物学年龄本身。让细胞丢掉身份或失控增殖的候选，在设计上就会被筛掉。

What we are building now

  • Models on public fibroblast data — Training and evaluating virtual-cell pipelines on aged human dermal fibroblasts from public datasets, with leave-one-donor-out evaluation as the primary discipline.
  • RejuvBench — An open benchmark (in development) that asks: how accurately can a model predict the effect of genetic perturbations on transcriptomic age in skin fibroblasts from unseen donors?
  • First CRO round — Internal wet-lab outsourcing specs are ready; external validation starts once candidates clear the computational gates.
  • Broader donors, then tissues — Expanding donor coverage beyond the current five-donor public set, then extending the same question to other aging-related fibroblasts (lung, then heart) as separate tracks.

我们正在建什么

  • 公开成纤维数据上的模型 — 在公开数据集的人皮肤老年成纤维细胞上训练与评测虚拟细胞流程，以留一供体交叉验证作为主纪律。
  • RejuvBench — 正在建设的开放评测基准：模型能否准确预测基因扰动对未见过供体皮肤成纤维细胞转录组年龄的影响？
  • 首轮 CRO — 内部湿实验外包规格已就绪；计算侧过门后，启动外部验证。
  • 更多供体，再扩组织 — 先把供体覆盖从当前五供体公开集扩大，再把同一问题扩展到其他衰老相关成纤维（肺，然后心），各自独立轨道。

One person. A network behind the loop.

“One-person company” means one person accountable for the loop—not one person alone at the bench. AI agents draft literature reviews, experiment plans, and much of the coordination. Advisors contribute part-time expertise. CRO labs run the wet work. Before anything goes to a CRO, a human reviews it.

一个人，闭环背后是一张网

这里的「一人公司」是指：一个人对闭环负责——不是一个人独自站在实验台前。AI 智能体承担大量文献、实验计划与协调草稿；顾问提供兼职专业意见；湿实验由 CRO 完成。任何发给 CRO 的内容，发出前都要经过人工复核。

Where we are

Linn is in early research: building models, benchmarks, and the first CRO validation path. We are not selling a product, and we make no medical claims. Performance numbers will appear here when official evaluation results are ready to share.

我们现在在哪一步

林恩生物处在早期研究阶段：正在建设模型、评测基准与首轮 CRO 验证路径。我们不销售产品，也不做医疗宣称。官方评测结果就绪后，才会在这里公布性能数字。