Author(s)

Junle Liu

Corresponding Author

Junle Liu

Abstract

With the continuous escalation of global pharmaceutical market demand for levodopa (L-DOPA) (reaching 250 tons annually), the development of efficient dopamine biosynthesis technology has become an urgent need. Traditional Streptomyces screening methods suffer from bottlenecks of low throughput and long cycle time, failing to meet the requirements of industrial production. This study constructed a machine learning-based functional prediction model for tyrosinases, achieving accurate prediction of dopamine-producing capacity in Streptomyces by integrating physicochemical properties, evolutionary features, and three-dimensional structural parameters (such as conservation of copper-binding sites and active site pocket volume) using the Random Forest algorithm. The model demonstrated excellent performance in 10-fold cross-validation (accuracy: 87.3%), and the Pearson correlation coefficient between virtual screening results and experimental data reached 0.82. Combined with flux balance analysis (FBA), this study further revealed that feedback inhibition of DAHP synthase (aroF) in the shikimate pathway represents a key metabolic bottleneck, with in silico knockout of its feedback site simulating a 37.2% increase in tyrosine production. The established "computational prediction-experimental validation" closed-loop paradigm improves screening efficiency by 20-fold, providing an intelligent solution for green biosynthesis of dopamine and other natural products.

Keywords

Dopamine biosynthesis; Streptomyces tyrosinase; Machine learning; Metabolic engineering; Flux balance analysis