Artificial intelligence in predicting extraction of anti-cancer compounds

- Observed various saponins in the particulate Saponaria vaccaria L. seeds.
- ANNs predicted saponins yields and extraction kinetics.
- Yields and extraction kinetics from ANN and diffusional models were comparable.

Saponins from the particulate Saponaria vaccaria L. seeds (0.29-0.84 mm, 15.35-61.40 g H2O/100 g dry mass) were extracted for methanol concentrations (MeOH) of 30, 50, 70, and 90 mL/100 mL H2O and temperatures (T) of, 30, 45, and 60 °C at ten extraction intervals (t) between 1 and 180 min. A calibration equation was developed from the liquid-chromatogram-mass spectroscopy peaks to quantify the extract yields (mg mL−1) for various types of saponins. An artificial neural network (ANN) with three inputs, MeOH, T, and t predicted the extraction kinetics and the yields with less than ca. 12% error. The ANN model not only slightly outperformed the numerical diffusional model, but it also made the prediction simple and faster eliminating the use of the partition coefficient and the effective diffusivity. Therefore an ANN model can be a right approach to predict the yields of saponins and similar products.