Neural Network Prediction of Corn Stover Saccharification Based on Its Structural Features

Abstract

The classic assay for a large population biomass is time-consuming, labor intensive, and chemically expensive. This paper would find out a rapid assay for predicting biomass digestibility from biomass structural features without hydrolysis. We examined the 62 representative corn stover accessions that displayed a diverse cell-wall composition and varied biomass digestibility. Correlation analysis was firstly to detect effects of cell-wall compositions and wall polymer features on corn stover digestibility. Based on the dependable relationship of structural features and digestibility, a neural networks model has been developed and successfully predicted the corn stover saccharification based on the features without enzymatic hydrolysis. The actual measured and net-simulated predicted corn stover saccharification had good results as mean square error of 1.80E-05, coefficient of determination of 0.942 and average relative deviation of 3.95. The trained networks satisfactorily predicted the saccharification results based on the features of corn stover. Predicting the corn stover saccharification without hydrolysis will reduce capital and operational costs for corn stover purchasing and storage.