Generative AI-driven enzyme engineering unlocks sustainable valorization of broccoli waste into sulforaphane

Abstract

Current production of sulforaphane (SFN), a health-promoting isothiocyanate, relies mainly on energy-intensive plant extraction or chemical synthesis, limiting sustainable large-scale manufacture. Here, a waste-to-value framework integrating generative artificial intelligence (AI), multi-scale modeling, and life-cycle assessment (LCA) was established to engineer a robust microbial biocatalyst for low-carbon valorization of broccoli waste. Genome mining identified a novel Lactobacillus-derived β-thioglucosidase as a promising scaffold. A ProteinMPNN–AlphaFold design pipeline targeting flexible loops near the active site yielded an improved variant with markedly enhanced catalytic performance and thermal robustness, with kcat increasing from 98.5 ± 3.2 to 421.3 ± 9.1 min−1 and Km decreasing from 4.6 ± 0.2 to 0.9 ± 0.1 mM. Molecular dynamics and hybrid quantum mechanics/molecular mechanics (QM/MM) analyses indicated that loop rigidification, improved hydrophobic packing, and a reshaped electrostatic environment promoted transition-state stabilization and reduced the calculated activation barrier for SC bond cleavage from 19.2 to 17.5 kcal mol−1. When applied to homogenized broccoli residues in a heterogeneous solid–liquid system, the engineered enzyme showed high SFN yields and broad tolerance across pH and temperature conditions. Comparative LCA further revealed ∼70% and ∼60% reductions in global warming potential and cumulative energy demand, respectively, relative to conventional routes. As the first study to integrate AI-driven protein engineering with rigorous sustainability metrics for SFN production, this work offers a scalable paradigm for the green biomanufacturing of high-value natural products from agricultural residues.