Bioethanol production estimated from volatile compositions in hydrolysates of lignocellulosic biomass by deep learning

• Published in: Journal of bioscience and bioengineering 2020-06, Vol.129 (6), p.723-729
• Main Author: Konishi, Masaaki
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Corn stover; Corncob; Deep learning; Ethanol; Lignocellulose; Neural network
• ISSN: 1389-1723; 1347-4421
• DOI: 10.1016/j.jbiosc.2020.01.006

The cell growth and ethanol production from hydrolysates of various types were estimated from the volatile composition of lignocellulosic biomass by deep neural network (DNN) and the significant compositions estimated by asymmetric autoencoder-decoder (AAE). A six-layer DNN achieved good accuracy with learning and validation losses—0.033 and 0.507, respectively—and estimated overall time courses of yeast growth and ethanol fermentation. The AAE decoded the volatile compositions and represented the features of significant inhibitors via nonlinear dimensionality reduction, which was partly different from those using partial least squares regression reported previously. It revealed the significant features of hydrolysates for bioethanol production, which are lost in conventional approaches. The approach using DNN and AAE is, therefore, useful for bioethanol fermentation and other bioproductions using raw materials. [Display omitted] •A deep neural network estimated well ethanol fermentation from hyderolysate-compositions.•Asymmetric autoencoder-decoder gave significant components.•The significant components included famous inhibitory compounds and the others.