10x Genomics ships Claude agent for single-cell analysis 'last mile'

10x Genomics launched Sentira Single Cell on Nov. 2, a cloud application that runs a Claude-based AI agent (a large language model from Anthropic) to complete the analysis step its software left unfinished last year: turning raw single-cell data into answers about which cell types responded to a treatment.

The product targets a specific bottleneck in the single-cell workflow. 10x's existing software could already group cells by similarities in their gene expression. But comparing treated and untreated samples — the question a biologist most often wants answered — required custom code written with Seurat, an R package, or Scanpy, a Python library. Most bench researchers either waited for a bioinformatician or the question went unasked.

"Now, with the capabilities emerging with LLMs, it's actually possible to automate that last mile in an agentic workflow," Michael Schnall-Levin, chief technology officer at 10x Genomics, told R&D World on Nov. 4. "It means people who don't write Python or R can analyze their own data and have a lot more direct access to it."

10x described this as the "last mile" problem when it first introduced Claude into its cloud analysis stack a year ago.

How the workflow actually runs

Inside Sentira Single Cell, a researcher types a question in plain English. The agent proposes an analysis plan, which the researcher can revise before execution. The agent then runs the analysis on 10x Cloud and returns results together with the underlying Python code, written with packages such as Scanpy. Each run produces an editable notebook recording methods and code, so the analysis remains reproducible and auditable rather than a black box.

Researchers can bring their own hypothesis or ask Sentira to suggest one from the data, and they can ask follow-up questions after seeing results. Because analyses run in the cloud, a lab can pursue several questions in parallel rather than queuing them through a single bioinformatician.

Henry Long, a scientific director at Dana-Farber Cancer Institute, emphasized speed in the launch announcement. "What stands out to me is how easily Sentira Single Cell lets us ask questions of our data directly and get answers quickly," Long said. "The questions themselves aren't new, but being able to ask them in natural language changes how fast we can get answers."

The claim is a customer testimonial in a vendor launch, not an independent benchmark. No sample sizes, dataset types or turnaround times were disclosed.

What is measured, and what is not

10x has run internal validation: the team tested the agent on datasets that a computational biologist had already analyzed by hand, then compared Sentira's outputs against known answers. As of Nov. 4, 10x had published no accuracy figures from those tests.

Schnall-Levin said the company plans to publish the results and the evaluation framework behind them, "so people can see what we have evaluated, what we haven't, and the accuracy level we've had." Until that publication appears, R&D managers weighing the tool against retaining bioinformatics staff have vendor statements only.

Pricing and the spatial companion

10x prices Sentira Single Cell by consumption, with promotional credits for new users; the company disclosed no dollar figures.

A companion application, Sentira Spatial, has been in free early access since Sept. 29 for researchers using 10x's Xenium and Atera instruments. It runs on NVIDIA GPUs with rapids-singlecell, a package developed by scverse — the group behind Scanpy — together with NVIDIA. 10x says the GPUs cut analysis cycles from hours to minutes. That claim is a projection from the vendor; no benchmark details were published. A paid launch is expected in early 2027.

For labs already on the Chromium platform, the near-term decision is narrow: whether consumption-priced, agent-run analysis beats the current queue for a bioinformatician's time. The accuracy publication 10x has promised will be the document to watch.