Machine-learning-guided extraction optimization and physicochemical characteristic of GRAS microalgae proteins for sustainable food applications

Microalgae are promising sustainable proteins, but recalcitrant cell envelopes and extraction-induced quality losses limit recovery and application. Six GRAS microalgae were screened, and Chlorella pyrenoidosa , Arthrospira platensis , and Chlorella sorokiniana were selected for ultrasound-assisted alkaline extraction integrating bath-probe benchmarking, optimization, and structure-function assessment. Probe sonication outperformed bath and alkaline-only controls ( p < 0.05), with A. platensis reaching 96.38% recovery and 365.3 mg/g biomass versus 71.93% and 272.7 mg/g in the control; both Chlorella species showed smaller gains consistent with cellulose/algaenan-rich walls. Response surface methodology identified species-specific optima, and machine learning-based Gaussian process regression provided uncertainty-aware modelling that improved cross-validation for the Chlorella datasets and supported robust operating-window selection where responses plateaued. Process intensity was quantified: the optimized probe protocol required 22.5 kJ per batch (∼6.94 kWh/kg biomass) and ∼19 kWh/kg recovered protein (optimized A. platensis example). Protein concentrates contained 63-79% protein (dry basis); SDS-PAGE and FTIR indicated preserved protein signatures without pervasive fragmentation. Functionality was strongly medium/pH dependent: solubility remained < 25% in deionized water but increased sharply in buffer at pH ≥ 7. A. platensis concentrates exhibited high water absorption (∼3.7 g/g), whereas optimized C. pyrenoidosa isolates showed high oil absorption (∼4.2 g/g). Emulsification and foaming were pH sensitive, with optimization improving emulsion stability at alkaline pH and modulating foam performance in a species-dependent manner. Overall, ultrasound-assisted alkaline extraction offers an energy-transparent route to multifunctional microalgal proteins and a framework for multi-response optimization. • Ultrasound-alkaline extraction optimized for protein recovery from GRAS microalgae • GPR-complemented RSM maps predictive uncertainty across the design space. • Energy consumption was 22.5 kJ per batch, equivalent to 6.94 kWh/kg for A. platensis. • Optimized protein isolates reached 63-79% purity with intact secondary structures • Functional assays revealed WAC and OAC comparable to soy, pea, and wheat proteins