Application of PLS and Back-Propagation Neural Networks for the estimation of soil properties

Two different multivariate calibration methods, Partial Least Squares (PLS) and Back-Propagation Neural Networks (BP-ANN), were applied to microbial community DNA to predict soil properties (%Sand, %Silt, %Clay, %Nitrogen, %Organic Carbon, DNA) in environmental soil samples. The microbial community DNA was extracted from 48 environmental soil samples. After amplification of bacterial ribosomal RNA by polymerase chain reaction (PCR), the products were separated by gel electrophoresis. Characteristic complex band patterns were obtained, indicating high bacterial diversity. Two hundred and fifty-six DNA band patterns were used to predict the soil properties. Based on the brightness of the bands, densitometric curves of DNA band patterns were extracted from the gel images. The curves were smoothed using the Savitsky-Golay method and scaled to DNA standard markers. The predictive powers of the two methods (PLS and BP-ANN) are presented and compared. The Back-Propagation Neural Networks method combined with principal component analysis (PCA) was found to have the most predictive power with the independent test set.