Skip to main content

Stanford Developing Mass Spec, Machine Learning System for Cystic Fibrosis Diagnosis

Premium

NEW YORK – Something as simple as moving a glass slide across the skin to collect perspiration may eventually enable a mass spectrometry test for cystic fibrosis and other diseases, according to a preliminary study published on Monday.

In the study published in the Proceedings of the National Academy of Sciences, researchers from Stanford University described a test, using sweat as a patient sample, that combined machine learning and mass spectrometry to diagnose cystic fibrosis, an autosomal recessive condition.

Specifically, the test, which takes two minutes to complete, leverages electrospray ionization mass spectrometry with a machine-learning algorithm based on the use of decision trees to analyze metabolic patterns. In an analysis of 277 samples, the test demonstrated an accuracy of between 96 percent and 100 percent, said Richard Zare, a Stanford University professor of chemistry and one of test's developer.

"Although this is in its infancy, the approach is really quite simple and appears to have accuracy that is better than the best methods that are now being used to diagnose cystic fibrosis," he said in an interview.

Cystic fibrosis is the result of mutations in the CF transmembrane conductance regulator (CFTR) gene. If the condition can be diagnosed early, CFTR modulator drugs and intensive therapy regimens can improve the outlook for patients.

However, the traditional method to detect the condition — including use of a decades' old cholinergic pilocarpine iontophoresis test to measure chloride concentration in sweat — has well-known challenges, said Carlos Milla, a study coauthor, Stanford University professor of pediatrics, and director of the Stanford Cystic Fibrosis Center.

The chloride concentration test requires that clinicians chemically induce sweating from babies or from others receiving CF testing. Most children with cystic fibrosis are diagnosed before the age of two and, according to the Stanford researchers, babies are most frequently tested.

The clinicians obtain sweat samples and test them for high concentrations of salt, which is a surrogate marker that warns when lung cells are not working properly. Because of the sampling challenges, the test cannot be completed for about 10 percent of infants and small children, Milla said.

Further, there can be significant ambiguity in results. "In California, for every two infants the test clearly identifies as having cystic fibrosis, another three will fall in a grey area where doctors can't say for sure whether the child will ever develop the disease," Milla said.

Zare said that the mass spec-based approach they are developing builds on a study he and his colleagues completed and published in 2017 in Analytical Chemistry in which they had classified genders, ethnicities, and ages of study subjects using fingerprint analysis by combining desorption electrospray ionization mass spectrometry and machine learning.

He added that their approach differs from methods that are focused on using specific disease biomarkers. The Stanford University approach analyzes patterns of metabolites revealed by mass spectrometry and compares the newly collected patterns with those previously recorded for patients with cystic fibrosis and people who are healthy.

Unlike with older testing methods for CF, collecting the sweat sample is also simple and noninvasive, Zare said. The investigators swipe the glass slide across the forehead or nose to collect perspiration naturally present on the skin. The method doesn't require stimulating sweat glands, and it eliminates the influence on test results of secretory rate. No additional sample preparation is required, and perspiration molecules enter the mass spectrometer with the aid of a solvent spray, Zare said.

In the study, a machine-learning model employing decision trees recognized the metabolic patterns that the test system used to differentiate samples. The model made predictions based on a series of tests of the intensities of pattern peaks, and it filtered out molecules from creams and lotions picked up by the mass spectrometer by recognizing a large variance associated with non-target molecules across different subjects, he added.

The investigators used a mathematical approach, called a statistical bootstrap, to provide an estimate of the uncertainty associated with the machine-learning model's predictions.

Other researchers are currently using metabolic biomarkers from different sources — including blood, sweat, breath, sputum, and stool — to develop approaches to diagnose cystic fibrosis. Among companies developing biomarker tests that provide alternative options to mainstream testing, Reeuwijk, Netherlands-based Breathomix has collaborated with clinical research groups using its breath profiling platform in preliminary studies to diagnose cystic fibrosis and other conditions.

Seattle-based NanoString recently added a panel designed to detect biomarkers of fibrotic diseases, such as NASH and cystic fibrosis, to its menu. And Menlo Park, California-based BillionToOne recently inked an exclusive distribution agreement with Heidelberg, Germany-based Eluthia to launch BillionToOne's Unity noninvasive prenatal test in certain European countries. Unity enables screening for the autosomal recessive conditions, including cystic fibrosis, from fetal DNA gathered from maternal blood.

As the Stanford group continues to conduct research and expand applications for the approach, a tie-up with a company familiar with the commercialization of diagnostic tests may be needed to move the method they've developed closer to market, Zare said. If the approach can be developed for clinical practice, its time-to-result will need to include time to transport samples to a laboratory that uses mass spectrometers and that can run the analysis. However, for diagnosis of a condition such as cystic fibrosis, achieving an immediate time-to-result may not be as critical as for other conditions, such as cancers and infectious diseases, Zare said. Additionally, if more mass spectrometers are adopted in clinical applications and prices for instruments and tests continue to drop, the new approach may find acceptance, he said.

For its work, the Stanford team used Thermo Scientific's LTQ-Orbitrap XL mass spectrometer.

Zare added he anticipates that eventually a point-of-care mass spectrometer may become available that would complement his group's diagnostic system and enable more immediate results. He noted, however, that he doesn't anticipate becoming involved in developing a point-of-care system and prefers instead to focus on developing new applications for the existing system.