Exploring the Complexities of Dissolved Organic Matter Photochemistry from the Molecular Level by Using Machine Learning Approaches

Photochemicalreactions are essential components alteringdissolved organic matter (DOM) chemistry. We first used machine learningapproaches to compatibly integrate existing irradiation experimentsand provide novel insights into the estuarine DOM transformation processes. Dissolved organic matter (DOM) sustainsa substantial part of theorganic matter transported seaward, where photochemical reactionssignificantly affect its transformation and fate. The irradiationexperiments can provide valuable information on the photochemicalreactivity (photolabile, photoresistant, and photoproduct) of molecules.However, the inconsistency of the fate of irradiated molecules amongdifferent experiments curtailed our understanding of the roles thephotochemical reactions have played, which cannot be properly addressedby traditional approaches. Here, we conducted irradiation experimentsfor samples from two large estuaries in China. Molecules that occurredin irradiation experiments were characterized by the Fourier transformion cyclotron resonance mass spectrometry and assigned probabilisticlabels to define their photochemical reactivity. These molecules withprobabilistic labels were used to construct a learning database forestablishing a suitable machine learning (ML) model. We further appliedour well-trained ML model to "un-matched" (i.e., notdetected in our irradiation experiments) molecules from five estuariesworldwide, to predict their photochemical reactivity. Results showedthat numerous molecules with strong photolability can be capturedsolely by the ML model. Moreover, comparing DOM photochemical reactivityin five estuaries revealed that the riverine DOM chemistry largelydetermines their subsequent photochemical transformation. We offeran expandable and renewable approach based on ML to compatibly integrateexisting irradiation experiments and shed insight into DOM transformationand degradation processes.