Dataset for "Active Learning Identifies Sulfur-Based Enhancers for Fe(III)-Protoporphyrin Catalysis: Recapitulating Features of Natural Oxidase and Beyond"

Sequence-controlled polymers, such as polypeptides, offer a versatile platform for tuning the microenvironment of catalytic centers, drawing inspiration from enzymes while enabling a larger design space, structural flexibility, automated synthesis, and compatibility with closed-loop optimization. Here, we designed an artificial oxidase system by immobilizing Fe(III)-protoporphyrin IX onto a lysine residue in synthetic decapeptides via amide linkage. Using hydrogen peroxide as the oxidant and acetophenone as a model substrate, we applied active learning to iteratively optimize peptide sequences across 234 variants over twenty rounds, yielding a steady increase in catalytic activity. Statistical analysis revealed that sulfur-containing residues—cysteine and methionine—consistently enhanced activity when positioned adjacent to the coordination site. Notably, although sequence optimization began from random inputs, the algorithm quickly converged on cysteine-containing motifs, consistent with features found in natural oxidases. Thioether-containing methionine was also found to promote catalysis, extending the relevance of sulfur-based coordination beyond naturally occurring systems. A hydrophobic micro-environment is equally critical: the activity increases linearly with the peptide hydrophobicity, whereas polar or acidic neighbors negate the sulfur benefit. These findings demonstrate the application of data-driven sequence design for developing tunable, enzyme-inspired catalysts with simplified architectures.