Automatic recognition of Pd ions in high-resolution mass spectra of multicomponent samples without visible fine isotope structures

Palladium plays a three-fold role in modern science and technology: it serves as a key active center in modern catalytic synthetic methods, occurs as unwanted trace contamination that influences new catalysts development and can lead to false-positive reports, and is an emerging environmental pollutant from automotive catalysts. High activity at ppm/ppb loadings and the necessity to analyze sub-micromolar concentrations of samples are common challenges in these applications, making reliable Pd identification essential. Mass spectrometry (MS) can detect Pd-containing ions in complex mixtures, but post-spectral recognition is difficult when minor Pd signals overlap with signals from other components and fine isotope structures are not visible. The present study introduces a machine-learning-based (ML) approach featuring a graph algorithm for the separation of signals from different multicharged ions and deep-learning classifiers for the automatic recognition of Pd ions in mass spectra of multi-component mixtures down to 10–8 mol L–1. It is applied to Pd intermediates in catalytic synthesis, contaminant catalysis and environmental samples. The scalability of the approach enables extension to Ni, Cu, Ag, Cl and Br. With the developed ML/MS workflow, Pd and other important elements could be automatically revealed; thus, routine MS hardware could be transformed into high-performance scanners. Mass spectrometric detection of Pd-containing ions in complex mixtures poses a challenge due to overlapping spectral features. Here, the authors employ a machine-learning-based graph algorithm to distinguish between various multicharged Pd ions, enabling automatic detection in multicomponent samples.