AI Could Boost Identification of Chron's Disease

Samara Rosenfeld
OCTOBER 02, 2019
stomach

Using artificial intelligence (AI), researchers at Rutgers University discovered new genes linked to Chron’s disease, according to the findings of a study published in the journal Genome Medicine.

The AI highlighted known Chron’s disease genes, including NOD2, and new potential genes. What’s more, the algorithm identified 16% of Chron’s disease patients at strict cutoff at 99% precision. At default cutoff, the AI identified 58% of Chron’s disease patients with 82% precision.

“We believe that we can use the knowledge gained from this study to similarly model other genetically linked diseases,” said senior author Yana Bromberg, Ph.D., an associate professor in the biochemistry and microbiology department at Rutgers University-New Brunswick.

Researchers evaluated patients from four panels: Chron’s disease-train, Chron’s disease-test, WTCCC panel and Genotype-Tissue Expression Project (GTEx) panel.

The Chron’s disease-train panel consisted of 111 people — 64 had Chron’s disease, 47 were healthy. The Chron’s disease-test panel included 51 patients with Chron’s disease and 15 health patients. The WTCCC panel contained 2,678 individuals with Chron’s disease, while the GTEx panel had data from 635 deceased individuals who had no indication of Chron’s disease and were considered healthy.

The research team used a machine-learning technique called AVA,Dx (Analysis of Variation for Association with Disease) to identify genes whose functions changed more in patients with Chron’s disease than in health people, and vice versa.

Accounting for batch effects, the researchers accurately predicted Chron’s disease for thousands of previously unseen individuals from other panels.

The method used less than 5% of the people normally involved in a genome-wide association study to identify disease genes and make accurate Chron’s disease predictions for unseen patients.

AVA,Dx is optimized for high precision and only misclassified a few healthy individuals as sick, the study authors noted.

The genes AVA,Dx identified appear to be relevant to Chron’s disease because of the matches of pathways to known work, the authors added.

“Our method is not a clinical diagnosis tool, but it generates interesting observations that need to be followed up,” Bromberg said.

Additional research could reveal molecular reasons behind certain forms of Chron’s disease and lead to better treatments of the disease, Bromberg concluded.

Get the best insights inside digital health directly to your inbox.

Related
SonarMD Cuts Crohn's Disease Hospitalizations, Costs
UCB and Dearhealth Partner
Retailers Continue Push into Care Coordination

SHARE THIS SHARE THIS
4
Become a contributor