AI-Assisted Metagenomic Mining of New Dye-Decolorizing Peroxidases: Systematic Biochemical Validation and Catalytic Structural Diversity

Abstract

Dye-decolorizing peroxidases (DyPs) are promising biocatalysts for lignin valorization and environmental remediation. Here, 427 putative marine Class P DyPs containing the conserved heme-binding motif were retrieved from metagenomic data sets. CatPred and UniKP were used in parallel to rank candidates, while AlphaFold3-derived models served as structural hypotheses for assessing predicted structural diversity and guiding candidate selection. The candidates were grouped into six structural classes, and seven representative enzymes were experimentally examined. Four showed detectable activity, including PhDyP and KpDyP, which exhibited optimal activity at pH 6.0, showed halotolerance, and retained 50% activity at 0 °C. Comparison of predicted and experimental properties showed that current kinetic-prediction tools support prioritization but do not reliably capture absolute kinetic values, assay-condition effects, heme incorporation, or global structural integrity. These findings expand the characterized diversity of Class P DyPs and support their further exploration as robust enzyme scaffolds for green bioremediation.