The objectives of this study were to identify molecular features characteristic to arctic DOM from the Kolyma River basin and to elucidate structural imprints induced by a choice of the sorption technique. To achieve this goal, DOM was isolated from the Kolyma River basin with a use of three nonionic sorbents: Amberlite XAD-8 resin, PPL- and C18 - SPE cartridges, and one anion exchanging resin–diethylaminoethyl (DEAE) -cellulose. The structural studies were conducted with a use of electrospray ionization Fourier Transform Ion Cyclotron Resonance (ESI FT-ICR) mass spectrometry and liquid state ¹H NMR spectroscopy. The DOM isolates obtained with a use of PPL and C18 cartridges were characterized with higher content of aliphatic compounds as compared to XAD-8 and DEAE-isolates. In total, for all arctic DOM isolates we observed predominance of hydrogen saturated compounds with high H/C values of identified formulas from FT-ICR MS data. ¹H NMR spectroscopy studies have confirmed this trend and revealed high contribution of alkyl-chain protons into the spectral density of the arctic DOM reaching 43% for PPL isolates.

Retention time prediction, facilitated by advances in machine learning, has become a useful tool in untargeted LC-MS applications. State-of-the-art approaches include graph neural networks and 1D-convolutional neural networks that are trained on the METLIN small molecule retention time dataset (SMRT). These approaches demonstrate accurate predictions comparable with the experimental error for the training set. The weak point of retention time prediction approaches is the transfer of predictions to various systems. The accuracy of this step depends both on the method of mapping and on the accuracy of the general model trained on SMRT. Therefore, improvements to both parts of prediction workflows may lead to improved compound annotations. Here, we evaluate capabilities of message-passing neural networks (MPNN) that have demonstrated outstanding performance on many chemical tasks to accurately predict retention times. The model was initially trained on SMRT, providing mean and median absolute cross-validation errors of 32 and 16 s, respectively. The pretrained MPNN was further fine-tuned on five publicly available small reversed-phase retention sets in a transfer learning mode and demonstrated up to 30% improvement of prediction accuracy for these sets compared with the state-of-the-art methods. We demonstrated that filtering isomeric candidates by predicted retention with the thresholds obtained from ROC curves eliminates up to 50% of false identities.