Two products had the highest estimated health risks, while three brands showed much lower readings in a study from Arak, Iran.
Researchers measured arsenic, cadmium, mercury, lead and copper in 11 pasta brands, Ash-Novin pasta and wheat germ powder using a laboratory technique called inductively coupled plasma atomic emission spectrometry. They then estimated daily intake, hazard quotients and consumption limits for the products.
Tak had the highest hazard quotient, at 1.657, driven by mercury. Zhik had a hazard quotient of 0.928, driven by arsenic. By contrast, Zar, Mak and Jahan had hazard quotients below 0.01. The researchers also developed a machine-learning system that predicted the hazard quotient from the measurements.
Which products had higher risks
Measured concentrations ranged from 0 to 0.834 milligrams per kilogram for arsenic, 0.0002 to 0.014 for cadmium, 0 to 0.502 for mercury, 0 to 2.0 for lead and 0 to 1.203 for copper.
The highest estimated hazard quotient was found in Tak pasta, at 1.657, with mercury identified as the main contributor. Zhik pasta had a hazard quotient of 0.928, mainly associated with arsenic; the researchers described these results as indicating potential non-carcinogenic health risks. Zar, Mak and Jahan had hazard quotients below 0.01.
A hybrid machine-learning framework combining Random Forest and Gradient Boosting predicted the hazard quotient with a reported R² of 0.97 and an RMSE of 0.042. It identified mercury as contributing 34% of the modeled risk and arsenic 28%, consistent with the study’s direct calculations.
Evidence and caveats
This was a laboratory analysis and health-risk assessment of 13 cereal-based products obtained from Arak, Iran. The researchers measured metal concentrations and used calculated daily intake and hazard quotients to estimate potential non-carcinogenic risk; the study did not report cases of illness or directly measure health effects in consumers.
The findings apply to the products and sampling described in the study and should not automatically be generalized to all cereal products in Iran. The abstract does not provide details about the number of samples per product, sampling dates or how representative the products were of the wider market. It also reports the machine-learning model’s accuracy but does not describe its independent validation in the abstract.
// Source
Scientific Reports · 2026 · DOI: 10.1038/s41598-026-67271-w
Authors: Reza Pourimani, Mohadese Feyzi, Hossein Sadeghi, SeyyedMohsen Mortazavi-Shahroudi
Institutions: Arak University