Beyond Partial Readouts: Toward Integrative Mechanism Inference in Catalytic Oxidation

Conspectus

Oxidative reactive species (ORSs) provide a shared mechanism language for catalytic oxidation in both natural systems and industrial applications. Mechanism investigation seeks to explain how these processes proceed and how they can be regulated, conventionally focusing on identifying the dominant ORS and quantifying its contribution to substrate transformation. Yet ORSs are short-lived, present at low concentrations, and are rapidly interconverting, making their mechanism roles inseparable from the complex environment in which they form, transform, and decay. Disagreements among established identification methods have consequently become a recurring source of controversy across well-studied oxidation systems.

Typically, ORS participations are examined through chemical probing, quenching, or direct spectroscopic detection. The widely used chemical approaches assume that each reagent reacts selectively with a single target species, allowing an experimental response to be read as a mechanism assignment. Such selectivity is scarcely attainable where multiple reactive species coexist and interconvert, so refining reagents or instruments alone cannot reconstruct the underlying chemistry once each response is assigned to one species. Our previous work on spin-trapping electron paramagnetic resonance (EPR) exposed this vulnerability, revealing that even this widely trusted probe can enter the reaction chemistry. Reconstructing these signals into kinetic profiles successfully resolved when each reporting product forms, evolves, and decays, although temporal resolution alone cannot trace the reactions linking these stages.

To organize this chemistry, we cast catalytic oxidation as a directed reactive species network, mapping each reactive species onto a node and every route of generation, interconversion, transformation, and dissipation onto an edge. External substrates or reporters couple to these nodes through distinct reactions weighted by their rate constants. Therefore, ORS dominance is not an intrinsic property of any node but a condition-dependent property of the network, set by how reaction conditions distribute reactive flux and by the substrate or reporter through which the network is read. Whether through probing, quenching, or direct detection, each measurement returns a low-dimensional projection of this high-dimensional network, so conflicting assignments may reflect one dynamic mechanism rather than incompatible chemistries. Recovering the hidden organization behind these partial projections thus becomes an inverse problem.

In this Account, we revisit current ORS identification strategies, examine how chemical, kinetic, and spectroscopic evidence constrains mechanism interpretation, and present a network basis for reconstructing the hidden chemistry behind the resulting observations. Realizing this route appeals to AI-assisted inference under chemist-centered judgment and to structured databases built under this network view through automated experiments, literature mining, and modeling orchestrated by agentic workflows. Thus, it offers not only a reinterpretation of the mechanism puzzles accumulated over decades but a new point of departure for the work ahead.