Notes of jasmine? Hints of citrus? Computers can be trained to smell like a human, say scientists at Philly’s Monell Center.

Scientists discovered long ago how to represent sounds and colors with raw numbers, paving the way for smartphones, flat-screen TVs, and other devices of the electronic world.

But odor and fragrance have remained quaintly stuck in a pre-digital era, characterized only with the subjective tool of human language.

That might be about to change, according to a new study coauthored by scientists at Philadelphia’s Monell Chemical Senses Center.

Using a machine-learning model developed at Google Research, the authors mapped out how the smells of 5,000 known chemical substances were related to their underlying structure. They then used this mathematical tool to predict the smells of 323 new, lab-made chemicals that no one had smelled before, finding that it performed just as well as a panel of trained human sniffers.

The study, published in the journal Science, is just a first step, as the tool was used only to predict the smells of individual chemicals, not mixtures. But with additional refinements, the tool could be used in such products as food, fragrances, and cleaning supplies, enabling manufacturers to safely screen thousands of chemicals for maximum odor appeal, Monell scientist Joel Mainland said.

It could even be deployed in “electronic noses” to screen people for cancer, detecting faint fingerprints of disease that are too subtle to be detected by the human nose.

“The model never smells it,” he said. “It just looks at the molecular structure.”

The math behind the model defies easy description, but it allows scientists to characterize the types of atoms and chemical bonds in any molecule with a complex system of numerical coordinates, Mainland said. This result can then be translated into a plain-English description of its smell, using some combination of 55 standard olfactory adjectives such as jasmine, vanilla, buttery, or metallic.

Mainland, a neuroscientist who also is affiliated with the University of Pennsylvania, was a senior author of the study along with Alexander B. Wiltschko, chief executive officer of a Cambridge, Mass.-based tech start-up called Osmo. Other authors included Emily Mayhew, a former postdoctoral fellow at Monell who is now an assistant professor at Michigan State University.

In a blog post about the results, Osmo’s Wiltschko described how he got hooked on the problem when studying neuroscience in college, 15 years ago.

“We know why an apple looks red, we know why the crash of a cymbal is bright and loud, but why don’t we know why a molecule smells of cut grass, or cinnamon, or nothing at all?” he wrote. “I couldn’t quite fathom how something that seems so simple would still be so mysterious to all of neuroscience.”

The study goes a long way toward solving one of the oldest problems in perceptual psychology, said Harvard neuroscientist Venkatesh Murthy, who was not involved with the study.

“I don’t think this is going to be a final answer,” said Murthy, the director of Harvard’s Center for Brain Science. “But it is a significant advance.”

Wiltschko helped develop the model while at Google Research, then launched Osmo in January after publishing preliminary results.

The smell evaluations from the computer model do not necessarily agree with those from every human being, as the sense of smell varies from person to person. But the results were a close match to the average ratings by a panel of 15 trained sniffers at Monell, an independent research institute known for its landmark golden statue of a giant nose and mouth on the sidewalk by sculptor Arlene Love.

In addition to screening people for cancer, the tool also could be useful in treating people with an impaired sense of smell, allowing their doctors to measure how well they respond to therapy, Monell’s Mainland said.

Someday, he said, the tool might even allow people to share digitized versions of smells on social media, just like they do with images and music.

Murthy, the Harvard scientist, agreed, though he said the recipient would then need some sort of device to turn that digitized version into an actual fragrance — analogous to how a smartphone screen converts a digital signal into a visible image.

But with the foundation laid by the new study, he said the concept carries more than a whiff of possibility.