Poster # 165

Dinesh Kumar Saini¹, Anshul Rana¹, Maitiniyazi Maimaitijiang², Jyotirmoy Halder¹, Jinfeng Zhang¹, Subash Thapa¹, Shaukat Ali¹, Karl Glover¹, and Sunish Sehgal¹
1. South Dakota State University, Department of Agronomy, Horticulture & Plant Science, Brookings, South Dakota
2. South Dakota State University, Department of Geography and Geospatial Sciences, Brookings, South Dakota
Corresponding Author: Sunish Sehgal, sunish.sehgal@sdstate.edu

Dinesh Kumar Saini

Fusarium head blight (FHB), caused by a necrotrophic pathogen Fusarium graminearum Schwabe is one of the most destructive fungal diseases of wheat which is known to produce a harmful mycotoxin called deoxynivalenol (DON). DON contamination of wheat grains and/or flours is serious problem for food safety globally. Currently, DON levels in grain or flour samples can be measured using a variety of methods, including enzyme-linked immunosorbent assay (ELISA) and gas chromatography-mass spectrometry (GC-MS) but these methods are time-consuming, costly, and destructive. Therefore, a limited number of samples were analyzed for DON content in the elite and advanced breeding nurseries and majority of the selections in early and preliminary trials were made based on visual score for FHB or Fusarium Damaged Kernels (FDK). Development of rapid and non-destructive and economical method to estimate DON content can help to better facilitate breeding for lower DON. In the present study, GC-MS was used to estimate the DON content in 250 advanced breeding lines from SDSU wheat breeding program that were evaluated in the 2020-21 FHB nursery. The DON content of the 250 selected lines ranged from 0 to 77.6 ppm. The flour samples (3 grams) were analyzed for moisture and protein content on NIR and subjected to close-range hyperspectral imaging (400–1000 nm). Four different machine learning models (PLS regression, Xgboost, RF, and SVM) and one deep learning model (1D CNN) were evaluated on the training (70%) and test (30%) sets to ascertain the most efficient model based on accuracy with the coefficient of determination in prediction (R²) and root mean squared error (RMSE). Overall, advanced hyperspectral imaging coupled with machine and deep learning approaches shows great potential in high-throughput estimation of DON content and will enable breeding for lower DON content.

Acknowledgement and Disclaimer: This material is based upon work supported by the U.S. Department of Agriculture, under Agreement 59-0206-2-153. This is a cooperative project with the U.S. Wheat & Barley Scab Initiative. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture.