Dr. Chris Marshall runs a structural biology lab at Toronto's University Health Network, solving protein structures by NMR, crystallography and cryo EM. His work centers on the oncogenic and tumor suppressor proteins behind cancer, tracing how mutations change their behavior, and on the earliest stage of drug discovery, screening for molecules that bind them.
He had tried a couple of general purpose AI tools before settling into a routine with EMET, and 25 queries in, he has a clear verdict on where it stands next to them.
"I've never had an answer that I thought had a mistake..while it can take time to respond, I don't mind waiting for a good answer rather than getting a mediocre answer immediately and having to spend more time verifying it."
Untangling 20 protein isoforms
One protein he works with has roughly 20 splice variants, each missing or retaining a different combination of domains, the kind of comparison that is simple to describe and slow to actually build by hand. He asked EMET to lay it out. It returned a clear summary alongside a rough figure mapping the sequence differences across every variant.
"That would have taken all day to assemble”
Choosing the right experiment
The same precision showed up on a technical question with no simple literature answer: which NMR experiments suit a protein with a particular set of features.
"I got a really surprisingly good answer, and the references I needed, that would have taken me all day to find."
A reference for every statement
What he keeps coming back to is how EMET cites its own answers.
"I like that each statement is generally referenced.. other AI gives you a paragraph and then five or six references at the end, and you don't know which one was for the statement you want to validate. That's done well. I like that."
Keeping his own judgment in the loop
Marshall does not send every question to EMET. Quick lookups still go to a plain search engine; EMET is for the questions that would otherwise cost him an afternoon. He has not yet pushed back on an answer or tested it the way he would test a colleague's claim, but he is already planning to, starting with a paper he is reviewing now, where he intends to check its summary of the literature against EMET's own read of the same sources.
That instinct, to treat a fast and detailed answer as a starting point rather than a finished one, is exactly how the tool is meant to be used. For a structural biology lab, the case for an agentic workbench is not that it replaces the reading. It is that it gets a scientist to the reading that matters faster, with the receipts attached.
Want to explore what’s possible?
Try EMETRelated articles
Two grant applications, a visual abstract and training the next generation of scientists: how a neuroscience lab works with EMET
The First Success for EMET as My Co-Investigator: He brought the idea. EMET brought the evidence, and weeks of his time back.
