Predicting fish by-product proteolysis status by RGB laser-scattering imaging combined with machine learning procedures

This work aimed to control the protein hydrolysis process of fish by-products through RGB laser-scattering imaging. By-products from mackerel and sardine waste were hydrolysed with papain, pepsin and Protamex®, and the degree of hydrolysis was measured six times during 6 h. Imaging information was captured at those sampling times to obtain non-destructive data across the process. The used enzymes reached a different degree of hydrolysis depending on the type of fish. After a dimensional reduction of the imaging, data evolved along the process, although the observed kinetics did not fit with those observed in the measurements of the degree of hydrolysis. When imaging data was supervised by a multivariate regression method, a high prediction of the degree of hydrolysis capacity was observed for all cases of enzyme/fish. Moreover, successful prediction models were obtained from all data within each fish. The effect of the type of enzyme was not significant for the prediction capacity. Thus, a common model including data from all cases was developed. In that case, the effect of the fish matrix made intervening data with pre-treatments and variable selection procedures necessary to obtain prediction coefficients up to 0.95. The results evidenced the capacity of this imaging technique to capture the variability generated by the hydrolysis of proteases with the independence of the type of enzyme and fish.