A computational oncology scientist on EMET: "By far the best thing about it was the way it was displaying information”

Dr. Vijyendra Ramesh is a scientist at the Dana-Farber Cancer Institute. His work runs from kidney cancer at the bench through to the computational side, including building pipelines for long read RNA sequencing.

He is not new to AI tooling for science. He had beta tested other agentic science tools already, and had recently started using another. So when he used EMET, he used it on three live use cases rather than on test prompts, and he had something to compare it against. On the things that mattered most to his week, EMET came out ahead of the other AI tools he has used.

"By far the best thing about EMET was the way it was displaying information. And it does that contextually, based on the type of thing you want from it."

Kidney cancer mutations, aggregated from the portal

Dr. Ramesh asked EMET to aggregate the mutations found in the most commonly used clear cell kidney cancer lines. EMET went into cBioPortal, pulled what was there, and returned a table of the relevant genes and the mutation types found in them, rendered inline, with the entries worth closer attention called out.

He didn’t even ask for a table - EMET knew it would deliver value without being prompted.

"It was very good at picking up that context. I didn't ask for it, but I appreciated it doing that."

Reaching into third party portals like cBioPortal and aggregating what comes back into something a scientist can act on is where he saw the clearest gap between EMET and the other tools he has tried.

Long read RNA seq pipelines, drawn as a decision tree

Long read RNA sequencing is an area where published guidance is thin, which makes it a hard question to ask any tool. Dr. Ramesh asked EMET to lay out the workflows that exist and the data transformations needed to reach a usable endpoint.

EMET wrote it out in prose. Then it also drew it.

"It did a good job of visualizing it, like a flowchart, point A to point B, a couple of options I could take at a particular step. That was interesting, rather than just laying it out text wise, which it did."

Related to the same work, he pointed to something that quietly eats hours in computational biology. Accession numbers for public repository datasets are buried inside the body of papers, so a plain web search will not hand them to you. EMET returned accession numbers together with the context of what each dataset actually contains, which is the part that decides whether the dataset is worth an afternoon.

An R script from scratch

Dr. Ramesh asked EMET to write an R script from scratch, using a specific R package to process a particular data type, against an objective he set. What came back was correctly formatted, used the right syntax, and worked through the problem in order.

"I was pretty impressed by how it organized everything."

He also compared response times against the other AI tools on his desk. EMET was the fast one.

The scientist stays in control

None of this replaces the scientist, and Dr. Ramesh did not treat it as though it did. Where he had prior knowledge, he pushed back on EMET and made it show its work. That is the loop the product is built for.

His experience shows what an academic oncology lab can get out of an agentic workbench when it is pointed at real projects: aggregation across portals, the data that is genuinely hard to find, and a first draft of the code, all in the place where the thinking is already happening.