A novel strategy combining full Fourier transform infrared spectroscopy with machine learning enhanced the accuracy for analyzing cell wall polymers in maize stover

Cellulose, hemicellulose and lignin are three major cell wall polymers extracted from crop feedstocks to produce valuable biofuels and bio-products. Due to the labor-intensive and environmentally hazardous characteristics of chemical methods for plant cell wall fractionation and polymer measurement, there exists an urgent necessity to develop high-throughput and environmentally sustainable techniques. In this study, we combined full Fourier Transform Infrared (FTIR) spectroscopy with advanced machine learning algorithms in 200 maize samples to create excellent models for predicting different cell wall polymers. The full FTIR spectral features were utilized as input dataset in comparison to 14 special peak features. All four algorithms demonstrated exceptional performance, achieving a high correlation coefficient above 0.92 in the training set and exceeding 0.85 in the test set. Notably, Random Forest model emerged as the most effective in predicting cell wall content, achieving an accuracy ranging from from 0.89–0.95 in the test set and demonstrating strong applicability for external validation with considerable robustness. Thus, this novel strategy offers a much simpler, accurate and high-throughput approach for predicting cell wall composition in large-scale maize samples than reported. The advancement of these models would enhance the selection of elite maize germplasms and improve the comprehensive utilization of maize feedstocks for bioenergy and biomass products. • There is a large variation of cell wall components among 200 maize sample stalks. • The full FTIR spectral features were utilized as input dataset for model building. • All four algorithms demonstrated enhanced accuracy based on full FTIR features. • Random Forest model performed strong applicability by external validation.