Alberta: Computer
system that creates personalized metabolic profiles could help doctors
predict diseases like Alzheimer's, cancer and diabetes before they
develop. A breakthrough in machine learning
has also brought about a “game changer” for the science of
metabolomics—and will hasten the development of diagnostic and
predictive tests for Alzheimer’s, cancer, diabetes and numerous other
conditions, leading to improved prevention and treatment. And that’s
just the tip of the iceberg.
Siamak Ravanbakhsh, who recently completed his PhD in computing
science at the University of Alberta and whose research was recently
published in the scientific journal PLOS ONE, said Bayesil, the computer application resulting from this breakthrough, is pretty easy to explain in basic terms.
“There is this technology called NMR spectrometry, which uses some of the same physical principles as MRI. This technology is very cool, because it can determine the concentration of certain compounds in your body,” he said. “Until now you couldn’t do NMR spectrometry automatically. Now you can do it automatically with Bayesil, so you can use it for medical purposes.”
The longer explanation is a bit more complex. NMR spectrometry can be used to determine the molecular composition of a biofluid like blood or cerebrospinal fluid, but it relies on a highly skilled user to manually correct and interpret the resulting spectrum, a visual representation of how much of each compound, or metabolite, is present. This manual process is time-consuming and produces differing interpretations of the results from expert to expert.
“A system that can quickly, accurately and autonomously produce a person’s metabolic profile would enable efficient and reliable prediction of many diseases from a single sample, which could significantly improve the way medicine is practised,” Ravanbakhsh wrote in his paper.
The “high-throughput method” Ravanbakhsh developed enables quick, highly accurate NMR spectral profiling. It will also accelerate the creation of a library of metabolic profiles for different biofluids that can then be used for diagnostic purposes.
David Wishart, a professor in the departments of biological sciences and computing science who co-supervised Ravanbakhsh’s research, said the ability to accurately screen for multiple compounds in biofluids at once has profound implications for the future of medical science.
Wishart, who is also director of the Metabolomics Innovation Centre, said colon cancer is an example of a condition that is difficult to test for—and often, by the time symptoms are present, it’s too advanced to treat effectively. Bayesil could enable the development of a urine test that would detect the precursors of colon cancer four or five years before the cancer develops, he said.
“The whole point of this technology is to develop predictive rather than diagnostic tools. If you can make testing really simple, routine and cost-effective, you’ll identify these things before they happen.”
Bayesil is now available online for people who have results from NMR spectrometers to try out for free. Ravanbakhsh said the Bayesil website has already seen visits and feedback from users around the world, which should help speed the evolution of the tool.
“The next step is producing libraries [of metabolic profiles] for different compounds. We need to popularize the idea.”
Currently Bayesil only works for spectra from blood serum and cerebrospinal fluid, but its proponents foresee a time when it can be used with spectra from saliva and urine to do non-invasive screening for various diseases. A similar tool already in use in Europe shows the potential to use Bayesil for quality control, labelling compliance and product enhancement in fluids like wine and fruit juice.
Russell Greiner, professor of computing science at the U of A and principal investigator at the Alberta Innovates Centre for Machine Learning, co-supervised Ravanbakhsh’s research and says the exciting phase of discovering Bayesil’s capabilities and potential is just beginning.
“Right now the tool is just producing profiles, which is an amazing accomplishment, a game-changer,” Greiner said. “Machine learning is about finding patterns in data, and I’m excited that we now have this high-throughput mechanism that can collect data from thousands of individuals—and I get to play with it, to find those patterns.
“The real question is, what can’t this technology be used for?”
“There is this technology called NMR spectrometry, which uses some of the same physical principles as MRI. This technology is very cool, because it can determine the concentration of certain compounds in your body,” he said. “Until now you couldn’t do NMR spectrometry automatically. Now you can do it automatically with Bayesil, so you can use it for medical purposes.”
The longer explanation is a bit more complex. NMR spectrometry can be used to determine the molecular composition of a biofluid like blood or cerebrospinal fluid, but it relies on a highly skilled user to manually correct and interpret the resulting spectrum, a visual representation of how much of each compound, or metabolite, is present. This manual process is time-consuming and produces differing interpretations of the results from expert to expert.
“A system that can quickly, accurately and autonomously produce a person’s metabolic profile would enable efficient and reliable prediction of many diseases from a single sample, which could significantly improve the way medicine is practised,” Ravanbakhsh wrote in his paper.
The “high-throughput method” Ravanbakhsh developed enables quick, highly accurate NMR spectral profiling. It will also accelerate the creation of a library of metabolic profiles for different biofluids that can then be used for diagnostic purposes.
David Wishart, a professor in the departments of biological sciences and computing science who co-supervised Ravanbakhsh’s research, said the ability to accurately screen for multiple compounds in biofluids at once has profound implications for the future of medical science.
“The analogy is that we’ve been looking at
the world through a keyhole and now we’re looking through a picture
window.” —David Wishart
“The analogy is that we’ve been looking at the world through a
keyhole and now we’re looking through a picture window,” he said. “For
most of the last century we’ve just measured a specific compound, like
glucose for diabetes. But if you could measure a hundred compounds at a
time, you would see there are a number of other things that are changing
that are quite significant, all of which can tell you how to treat or
manage a given condition.Wishart, who is also director of the Metabolomics Innovation Centre, said colon cancer is an example of a condition that is difficult to test for—and often, by the time symptoms are present, it’s too advanced to treat effectively. Bayesil could enable the development of a urine test that would detect the precursors of colon cancer four or five years before the cancer develops, he said.
“The whole point of this technology is to develop predictive rather than diagnostic tools. If you can make testing really simple, routine and cost-effective, you’ll identify these things before they happen.”
Bayesil is now available online for people who have results from NMR spectrometers to try out for free. Ravanbakhsh said the Bayesil website has already seen visits and feedback from users around the world, which should help speed the evolution of the tool.
“The next step is producing libraries [of metabolic profiles] for different compounds. We need to popularize the idea.”
Currently Bayesil only works for spectra from blood serum and cerebrospinal fluid, but its proponents foresee a time when it can be used with spectra from saliva and urine to do non-invasive screening for various diseases. A similar tool already in use in Europe shows the potential to use Bayesil for quality control, labelling compliance and product enhancement in fluids like wine and fruit juice.
Russell Greiner, professor of computing science at the U of A and principal investigator at the Alberta Innovates Centre for Machine Learning, co-supervised Ravanbakhsh’s research and says the exciting phase of discovering Bayesil’s capabilities and potential is just beginning.
“Right now the tool is just producing profiles, which is an amazing accomplishment, a game-changer,” Greiner said. “Machine learning is about finding patterns in data, and I’m excited that we now have this high-throughput mechanism that can collect data from thousands of individuals—and I get to play with it, to find those patterns.
“The real question is, what can’t this technology be used for?”