Questions, answered.

Single-cell data: a standard .h5ad AnnData file, rows = cells, columns = genes, gene symbols as HGNC. Raw counts, normalised or log1p are all fine — the server detects which.

Spatial data: a 10x Space Ranger ≥ 3.0 output folder (zipped), with the H&E hi-res image present. Spatial bins should be ≤ 50 μm; LingoCell does not currently aggregate larger bins.

A results page at the job's own link, which only your account can open, and an e-mail when the job has ended. The e-mail gives the link to that page; it never contains file names or results. You can switch these e-mails off under Account.

On the results page you find a PDF report and the result files of the task you ran, each with its size and checksum, ready to download. The report states what was run on which file, the checks made on the input, every warning, the settings and the software release that were used, and the limits of the method.

Result files stay on the server for 30 days after the job finished, large data files for 7 days; both dates are shown on the results page and in the e-mail. The PDF report is yours to keep. LingoCell is for research use only: results must not be used for diagnosis or treatment decisions.

任务结束后，您会得到一个结果页面（即任务链接，只有您的账号可以打开）和一封通知邮件。邮件中给出该页面的链接，不包含文件名和结果内容；可在账号页面关闭此类邮件。结果页面提供 PDF 报告和该任务的结果文件，均可下载，并列出大小和校验值。报告写明对哪个文件运行了什么分析、对输入所做的检查、全部警告、实际使用的参数与软件版本，以及方法的局限。

结果文件在任务完成后保留 30 天，大型数据文件保留 7 天，两个到期日都会显示在结果页面和邮件中；PDF 报告下载后可长期保存。LingoCell 仅供科研使用，结果不得用于诊断或治疗决策。

One model with three trained stages. Stage 1 aligns gene expression and text via CLIP and a TinyLlama decoder. Stage 1.5 bridges the scRNA-seq ↔ Visium HD distribution gap. Stage 2 adds a spatial branch and learns niche embeddings while preserving Stage 1 capabilities.

The same checkpoint serves all four tasks on the server.

786 cell types, taken from a curated subset of CellxGene's ontology. Cells outside this set still receive a valid e_cell vector; the type prediction will just be the nearest neighbour inside the vocab. See cell_type_code_mapping.json in Resources.

For retrieval: top-1 cosine ≥ 0.55 is reliable for 50K-candidate indices; 0.40 – 0.55 is borderline and warrants manual review; below 0.40 should be treated as no match.

For descriptions: turn on save_attn and check that the top-attended genes are biologically consistent with the claim. We document an attention-grounded reading protocol in the Describe walkthrough.

Cell-type annotation: ~ 26 ms / cell on an RTX 4090 (fp16). 100 K cells finish in ~ 45 s.

Spatial-niche on a 150 K-cell Visium HD slide: ~ 11 min including H&E feature extraction.

Description generation: ~ 180 ms / cell with greedy sampling.

Yes. Inference fits in 12 GB vRAM for scRNA-seq and 24 GB for Visium HD with the default configuration. Training is gated by 4× H100 — we provide checkpoints, not a one-click trainer.

CPU inference works for small tests; expect ~ 15 min per 10 K cells.

The pipeline runs an alias-resolution step (HGNC + GENCODE) before tokenising. As long as 60 % or more of your symbols map cleanly, results stay within ±1 % of fully-mapped accuracy. Unmapped genes are silently zero-filled, and the report lists which ones.

Code is MIT-licensed. Pretrained weights are CC-BY-NC 4.0 — free for academic use, contact us for a commercial agreement. Generated cell descriptions inherit the same non-commercial restriction.

Yes, for as long as it is needed to run your jobs and no longer. Your file is stored on our cloud server in Hong Kong and is copied to the laboratory's GPU machine in Japan while a job runs on it. It is used only to run the jobs you submit; it is not used to train or improve models.

An account holds one file at a time. The file is deleted automatically 24 hours after the last job that used it has ended (72 hours if that job stopped because of a fault on our side or because it reached a time, memory or output limit, so that it can be run again), or 24 hours after the upload if no job used it. Uploading a new file replaces the earlier one, and you can delete the file yourself at any time. An upload that was never finished is discarded 24 hours after its last part arrived.

Results belong to your account, not to a public link: only you can open them, after signing in. Result files are deleted 30 days after the job finished, large data files after 7 days. The Privacy Policy has the full list, including sign-in records and backups.

会保存，但只保存到完成您的任务所需的时间。文件存放在位于中国香港的云服务器上，任务运行时会复制到位于日本的实验室 GPU 计算服务器。文件只用于运行您提交的任务，不会用于训练或改进模型。

每个账号同一时间只保存一个文件。文件在最后一个使用它的任务结束 24 小时后自动删除（如果该任务因我方故障、或因达到时间、内存或输出大小上限而中止，则保留 72 小时，以便重新运行）；上传后未被任何任务使用的文件在 24 小时后删除。上传新文件会替换原有文件，您也可以随时自行删除。未完成的上传在最后一个分片到达 24 小时后被清除。

结果归属于您的账号，而不是一个公开链接：只有您本人登录后才能打开。结果文件在任务完成 30 天后删除，大型数据文件在 7 天后删除。完整清单（包括登录记录和备份）见《隐私政策》。

Yes. Uploading data and running jobs need an account; reading this site does not. Signing in for the first time creates the account: there is no separate registration and no password.

Enter your e-mail address on the Sign in page and we send you a one-time code, valid for 10 minutes. Where the Google or GitHub buttons are shown you can continue with one of them instead; if the address it reports already has a LingoCell account, we may first send a code to that address to confirm that it is yours.

At the first sign-in we ask for your full name, your institution or organisation and your country or region, for your agreement to the Terms of Use and the Privacy Policy, and separately for your consent to the processing of your data outside your own country. Results and notifications are tied to the account's e-mail address, which cannot be changed later.

需要。上传数据和运行任务必须登录；浏览本网站不需要。首次登录即自动创建账号，无需另行注册，也无需设置密码。在登录页面输入邮箱地址，我们会发送一次性验证码，10 分钟内有效。页面上显示 Google 或 GitHub 按钮时，也可以用它们登录；如果它们提供的邮箱已经有 LingoCell 账号，我们可能会先向该邮箱发送一次验证码，以确认邮箱确属您本人。

首次登录时需要填写姓名、所属单位、国家或地区，同意《使用条款》和《隐私政策》，并单独同意在您所在国家或地区之外处理您的数据。结果和通知都与账号邮箱绑定，邮箱之后无法更改。

Yes, with the e-mail code. Google and GitHub sign-in usually cannot be reached from mainland China, so use the e-mail code there. Before you upload, please also read the entry on data from human samples collected in mainland China further down this page.

可以，请使用邮箱验证码登录。中国大陆网络通常无法使用 Google 或 GitHub 登录。上传数据之前，请务必阅读本页下方关于在中国境内采集的人体样本数据的说明。

Wait a little first: some institutional mail servers hold messages from new senders. Look in the Spam or Junk folder, and check the address you typed for typos. Add no-reply@mail.lingocell.com to your contacts or allow-list, then ask for a new code. That is possible 60 seconds after the last one, and codes we sent earlier keep working until they expire.

Enter the code in the same browser window where you asked for it. If nothing has arrived after 10 minutes, try another mailbox or contact the lab for help.

请先稍等片刻：部分单位邮箱的安全网关会延迟或拦截来自新发件人的邮件。请查看“垃圾邮件”文件夹，并确认邮箱地址没有输错。将 no-reply@mail.lingocell.com 加入通讯录或白名单后重新获取验证码：距上一次发送 60 秒后即可重新获取，此前收到的验证码在过期前仍然有效。

请在发起登录的同一浏览器窗口中输入验证码。10 分钟后仍未收到的，可换一个邮箱再试，或联系实验室。

Jobs run one at a time on a single GPU machine in the laboratory, which is also used for other work. A submitted job waits in the queue until the machine is free, and its page shows how many jobs are ahead of yours. The order is fair between users, so one user with many jobs does not hold up everyone else.

You can close the page: the job keeps its place, and we e-mail you when it ends. While the GPU machine is offline or busy, jobs stay in the queue and start by themselves when it is back. A job that has still not started after 7 days is closed, and you are told by e-mail. You can cancel a waiting or running job on its page.

任务在实验室的一台 GPU 计算服务器上逐个运行，这台机器同时也承担其他工作。提交后的任务先进入队列，待机器空闲时开始；任务页面会显示排在您前面的任务数。排队顺序在用户之间保持公平，不会因为某位用户提交了很多任务而让其他人一直等待。

您可以关闭页面：任务会保留在队列中，结束时我们会发邮件通知。GPU 计算服务器离线或忙于其他工作时，任务继续排队，恢复后自动开始。排队超过 7 天仍未开始的任务将被关闭，并邮件通知您。排队中或运行中的任务可以在任务页面取消。

One .h5ad file per account at a time, up to 200 MB. Uploading again replaces the earlier file once the new upload has finished; a larger dataset has to be split into several files and jobs. An upload can be paused and resumed, and an interrupted one continues where it stopped if you choose the same file again within 24 hours.

Each account can have 3 jobs waiting or running at once and can submit 10 jobs a day. A job that runs longer than the time limit of its task, at most 4 hours, is stopped.

每个账号同一时间只能保存一个 .h5ad 文件，大小不超过 200 MB。再次上传时，新文件上传完成后即替换原有文件；更大的数据集需要拆分为多个文件、分别提交任务。上传可以暂停和续传；上传中断后，24 小时内重新选择同一文件即可从中断处继续。

每个账号最多可同时有 3 个排队或运行中的任务，每天最多提交 10 个任务。任务运行时间超过该任务的时限（最长 4 小时）将被中止。

Your file: use Delete next to it on the Server page or under Account. A file that a waiting or running job still needs can be deleted once that job has ended or been cancelled.

A job and its result files: open the job page and delete the job there.

Everything: delete your account under Account and confirm with your e-mail address. This removes your profile, sign-in methods, file, jobs and results from our servers and cancels running jobs; it cannot be undone. What remains is a record of the consents you gave and a minimal record of each job (its id, task, status and dates), both without your e-mail address, and encrypted database backups, in which deleted data can persist for up to 14 days. If you cannot sign in, contact the lab and we will delete the account for you.

文件：在 Server 页面或账号页面使用文件旁的删除按钮即可。仍被排队或运行中的任务使用的文件，需等该任务结束或取消后才能删除。任务及其结果文件：打开任务页面，在那里删除该任务。

全部数据：在账号页面删除账号，并输入您的邮箱地址确认。这将从我们的服务器上删除您的个人资料、登录方式、文件、任务和结果，并取消运行中的任务，且无法撤销。删除后仍会保留的只有：您作出同意的记录和每个任务的最简记录（编号、任务类型、状态和日期），两者均不含您的邮箱地址；以及加密的数据库备份，已删除的数据在备份中最长还会保留 14 天。如果无法登录，请联系实验室，我们会为您删除账号。

Please note: data generated from human samples collected in mainland China cannot be uploaded to this service unless the dataset is already lawfully public. This is a legal restriction on providing such data to an institution outside China.

Before every upload you are asked to confirm that your dataset is not of this kind. If you are not sure whether your data is affected, contact the lab before you upload.

请注意：利用在中国境内采集的人体样本产生的数据（已依法公开的数据除外）不能上传至本服务。这是关于向境外机构提供人类遗传资源信息的法律限制。

每次上传前，您都需要确认数据不属于此类。如不确定您的数据是否受此限制，请在上传前联系实验室。