Skip to main content
ARS Home » Southeast Area » Stoneville, Mississippi » Crop Genetics Research » Research » Publications at this Location » Publication #138630
Title: CLASSIFIER DEVELOPMENT OF NIR (NEAR INFRARED SPECTROSCOPY) DERIVED DATA TO PREDICT SCN RESISTANCE

Author
item CAO, M - UNIV OF MO
item SLEPER, D - UNIV OF MO
item Arelli, Prakash
item ROBERTS, C - UNIV OF MO
item SHYU, C - UNIV OF MO

Submitted to: Crop Science
Publication Type: Abstract Only
Publication Acceptance Date: 11/10/2002
Publication Date: 11/10/2002
Citation: Cao, M., Sleper, D.A., Arelli, P.R., Roberts, C.A., Shyu, C.R. 2002. Classifier development of nir (near infrared spectroscopy) derived data to predict scn resistance. Crop Science. p. 207

Interpretive Summary:

Technical Abstract: Soybean cyst nematode (Heterodera glycines Ichinohe) (SCN) is the most damaging pest of soybean today. Current methods for selecting SCN resistant genotypes are labor and resource intensive. Near infrared spectroscopy (NIR) was initially used to determine the resistance to SCN in our lab, but more accurate calibration equations remain to be developed for the prediction. Because of the complexity of spectrum data, appropriate dimension reduction and multi-collinearity must be addressed to make stable predictions. Machine learning techniques (computational statistics) are suitable for dealing with high dimensional data, non-homogeneous data, imprecise questions with weak (or no) assumptions which are beyond the computation limits of traditional statistics. The objective of this study was to use machine learning techniques to classify SCN resistant and susceptible soybean genotypes. Various supervised learning approaches (forward selection, principal component regression and neural networks) were estimated based on the data collected from NIR analysis with bioassay results providing class labels for the resistant and susceptible genotypes. Our preliminary results indicated that forward selection had better prediction, evidenced by the cross-validation. The advantages and disadvantages of these methods will be discussed. Heterogeneous samples will be used to test the validity of the classifiers.