Dr. Adam Willits, a senior research associate at The University of Kansas Medical Center, has been trying to build a multiplex antibody panel for immunofluorescence. Multiplexing means several targets stained and imaged together, so the panel has to work as a set, not as a list of individually plausible antibodies. When it is wrong, you find out slowly, and after you have paid.
He had already put other general purpose AI platforms on the problem, including training them on his own material first. Each of them gave him a panel.
"Other platforms have provided antibody panels that somewhat match what I needed, but they haven't worked when I purchased and tried them."
The refusal
EMET's answer was different. It told him the panel was not achievable given his requirements. Then it pointed him toward different assays that could answer the same biological question, and supplied protocols and reagents for them.
"The fact that it chose to tell me no and provide an experimental alternative that can answer the same question was a pleasant surprise."
That is two things, and the second is what makes the first useful. A tool that only declines sends the scientist back to the beginning. One that reroutes the question has done the harder half of the thinking.
Why the no was worth taking seriously
A refusal is only worth as much as the judgment behind it, and he had a basis for weighing this one. On the same projects, EMET had found gaps in the published literature on his topic and explained the underlying concepts at a level that the other tools did not reach, without first being trained on his material.
"It has been able to find holes in the literature and explain scientific concepts at a much deeper level without training compared to [other platforms] with training I’ve supplied."
So when it declined, it was not a system hedging at the edge of what it knew. It was the tool that had been going deeper than the others, telling him the design did not hold.
"It is the most scientifically knowledgeable AI platform out there."
Learning QuPath for histology analysis
Declining is not a default setting. Alongside the panel work, he has been learning QuPath, open source software for quantitative pathology and image analysis, well enough to trust his own numbers. EMET is what he asked when he got stuck.
"It has been particularly helpful at teaching me science-related software like QuPath for analyzing histology, better than other AI platforms."
Which is why EMET now has a defined place in his week.
"I tend to use EMET in questions related to a concept in the field, established resources, products, protocols, etc."
The bench is still the judge
None of this rests on taking EMET's word for anything. The earlier panels failed because he bought the antibodies and ran them. His comparison is not between outputs on a screen; it is between what happened afterward in his tissue. EMET's no carries weight precisely because he had already paid for the alternative.
That is where EMET is different. Not only answering faster, but being right often enough, and honest often enough, that a scientist can plan around it.
Want to explore what’s possible?
Try EMET
