Use of RSM and CHAID data mining algorithm for predicting mineral nutrition of hazelnut

Abstract

Defining optimal mineral-salt concentrations for in vitro plant development is challenging, due to the many chemical interactions in growth media and genotype variability among plants. Statistical approaches that are easier to interpret are needed to make optimization processes practical. Response Surface Methodology (RSM) and the Chi-Squared Automatic Interaction Detection (CHAID) data mining algorithm were used to analyze the growth of shoots in a hazelnut tissue-culture medium optimization experiment. Driver and Kuniyuki Walnut medium (DKW) salts (NH4NO3, Ca(NO3)2·4H2O, CaCl2·2H2O, MgSO4·7H2O, KH2PO4 and K2SO4) were varied from 0.5× to 3× DKW concentrations with 42 combinations in a IV-optimal design. Shoot quality, shoot length, multiplication and callus formation were evaluated and analyzed using the two methods. Both analyses indicated that NH4NO3 was a predominant nutrient factor. RSM projected that low NH4NO3 and high KH2PO4 concentrations were significant for quality, shoot length, multiplication and callus formation in some of the hazelnut genotypes. CHAID analysis indicated that NH4NO3 at ≤1.701× DKW and KH2PO4 at >2.012× DKW were the most critical factors for shoot quality. NH4NO3 at ≤0.5× DKW and Ca(NO3)2 at ≤1.725× DKW were essential for good multiplication. RSM results were genotype dependent while CHAID included genotype as a factor in the analysis, allowing development of a common medium rather than several genotype specific media. Overall, CHAID results were more specific and easier to interpret than RSM graphs. The optimal growth medium for Corylus avellana L. cultivars should include: 0.5× NH4NO3, 3× KH2PO4, 1.5× Ca(NO3)2.