AGENTi: APCI-ESI Analyte Ionization Fidelity Pre-Screening Workflow using Multimodel Inference

Abstract Ionization sources convert neutral analytes into ions, enabling their subsequent separation and detection by mass spectrometry. Depending on the technique, this mechanism occurs within a solution or in the gas phase. The two most common methods to generate ions are electrospray ionization (ESI) and atmospheric-pressure chemical ionization (APCI), respectively. Both ESI and APCI can generate the same resultant ion species; however, it has been shown that preferential charge sites can change in response to an ionization source environment. Therefore, an additional quality control is warranted a priori to delineate erroneous analyte identification and/or unequivocal structural assignment. For protonation, protomer fidelity can be assessed by comparing the ion-mobility collision cross section (CCS) values obtained for both ESI and APCI. In this work, we utilized proxy standards (i.e., controls) derived from the available literature to develop a proof-of-concept multimodel inference application called AGENTi to probe differences in ionization for high-throughput protomer shift exploratory analysis. AGENTi uses in sequence the TensorFlow Autoencoder anomaly detection and the YDF Gradient Boosted Tree machine learning frameworks to generate more balanced and robust anomaly scores. We found that using a multimodel screening approach can reduce false positive capture at an acceptable loss of true positives. Moreover, we expect to achieve a better recovery of true positives with an improvement on the current Gradient Boosted Tree regression model for ΔCCS prediction as more ESI-APCI dual data are obtained for training. In all, we demonstrated in this pilot study that our foundation model for a priori probing of ionization pathway fidelity is feasible and efficient with AGENTi. The AGENTi prototype application and documentation are accessible at https://github.com/mitkeng/AGENTi.