Phenotyping technology for assessing protein content in seaweed by field spectroscopy and a machine learning algorithm

Abstract

Determining seaweed protein concentration and the associated phenotype is critical for food industries that require precise tools to moderate concentration fluctuations and attenuate risks. Algal protein extraction and profiling have been widely investigated, but content determination involves a costly, time-consuming, and high-energy, laboratory-based fractionation technique. The present study examines the potential of field spectroscopy technology as a precise, high-throughput, non-destructive tool for on-site detection of red seaweed protein concentration. By using information from a large dataset of 144 Gracilaria sp. specimens, studied in a land-based cultivation set-up, under six treatment regimes during two cultivation seasons, and an artificial neural network, machine learning algorithm and diffuse visible-near infrared reflectance spectroscopy, predicted protein concentrations in the algae were obtained. The prediction results were highly accurate (R2 = 0.95; RMSE = 0.84), exhibiting a high correlation with the analytically determined values. External validation of the model derived from a separate trial, exhibited even better results (R2 = 0.99; RMSE = 0.45). This model, trained to convert phenotypic spectral measurements and pigment intensity into accurate protein content predictions, can be adapted to include diversified algae species and usages. HighlightNon-destructive determination of protein content in the edible red seaweed Gracilaria sp. by in-situ, VIS-NIR spectroscopy and a machine learning algorithm.