Hilda Sandström

The study of aerosol formation and chemistry using machine learning is limited by the lack of molecular descriptors suited to atmospheric compounds. Interpretable models are particularly affected because they often rely on dictionary-based descriptors tied to specific molecular substructures, which currently fail to capture the full range of organic atmospheric compounds, including large, highly oxidized molecules common in the atmosphere. We introduce ATMOMACCS, an interpretable descriptor combining the 166 binary keys of the MACCS fingerprint with motifs inspired by the SIMPOL method for estimating saturation vapor pressures. We show that ATMOMACCS outperforms the RDKit topological fingerprint in kernel ridge regression models, improving predictions of saturation vapor pressures (7%, 8%, 29%, and 43% error reduction), equilibrium partition coefficients (5% and 9% error reduction), glass transition temperatures (22% error reduction), and enthalpies of vaporization (61% error reduction) on six datasets with atmospheric compounds. Feature analysis shows that saturation vapor pressure and partition coefficients are governed by carbon number and oxygen-related features, whereas other phase-transition properties (e.g., enthalpy of vaporization and glass transition temperature) depend on carbon–hydrogen bond types and the presence of heteroatoms other than oxygen. This highlights the generalizability of ATMOMACCS across different datasets and properties as an interpretable molecular descriptor.

This work reveals a striking exception to the well-established rule in chemistry that polar and nonpolar compounds do not spontaneously mix: insertion of methane, ethane, and other small hydrocarbons into the crystal lattice of hydrogen cyanide (HCN), a highly polar molecule. By mixing these components at cryogenic temperatures, we can observe distinct shifts in vibrational modes using Raman spectroscopy. Our computational predictions confirm that cocrystal structures of HCN and ethane, which match our experimental vibrational shifts closely, are thermodynamically and kinetically stable. Given that methane, ethane, and HCN are major components of the atmosphere and surface of Saturn’s moon Titan—where they play key roles in shaping chemistry, weather, and landscape—our findings may prove instrumental for explaining Titan’s chemical and geological evolution.

The formation of aerosol particles in the atmosphere impacts air quality and climate change, but many of the organic molecules involved remain unknown. Machine learning could aid in identifying these compounds through accelerated analysis of molecular properties and detection characteristics. However, such progress is hindered by the current lack of curated datasets for atmospheric molecules and their associated properties. To tackle this challenge, we propose a similarity analysis that connects atmospheric compounds to existing large molecular datasets used for machine learning development. We find a small overlap between atmospheric and non-atmospheric molecules using standard molecular representations in machine learning applications. The identified out-of-domain character of atmospheric compounds is related to their distinct functional groups and atomic composition. Our investigation underscores the need for collaborative efforts to gather and share more molecular-level atmospheric chemistry data. The presented similarity-based analysis can be used for future dataset curation for machine learning development in the atmospheric sciences.

Aerosol formation is a complex process involving numerous molecules, whose identities and environmental variations remain largely uncharted (Bianchi et al., 2019). Computational simulations and property prediction tools have emerged to identify compounds likely to participate in the particle formation process (Elm et al., 2020). In recent years, predictive machine learning models for saturation vapor pressure and partition coefficient estimation have achieved impressive accuracy, with mean absolute errors within one order of magnitude (Besel et al., 2023; Lumiaro et al., 2021); an advancement that enables the categorization of molecules into different volatility regions. However, the interpretability of these models in molecular sciences is often limited unless the molecular descriptor used is easily interpretable. Another challenge is that atmospheric molecules possess unique characteristics that may be overlooked by standard molecular representations developed in other chemical domains (Sandström et al., 2024). We hypothesize that combining sufficiently informative and interpretable descriptors with modern machine learning methods, chemical insight of these largely unknown chemical spaces can be gained in a data-driven way.In this contribution, we introduce a new interpretable molecular descriptor, ATMOMACCS, specifically tailored to atmospheric molecules. We demonstrate its competitive performance in predicting various thermodynamic properties, such as saturation vapor pressure, vaporization enthalpy, partition coefficients, and glass-transition temperature, equaling or surpassing published results for four distinct atmospheric molecular datasets (Besel et al., 2023; Wang et al., 2017; Ferraz-Caetano et al., 2024; Li et al., 2020). Our descriptor is based on enumerating atmospherically relevant structural motifs, making it readily interpretable for atmospheric chemists. Additionally, in our approach, we analyze the relative importance of these motifs with Shapley Additive Explanations (SHAP) values (Lundberg & Lee, 2017), providing insight into the performance improvements observed. Notably, from this analysis, we found that explicitly counting the number of carbon atoms is particularly important for property prediction, though less so for water-gas phase partition coefficients. Moreover, the analysis shows that general structural motifs are roughly equally important as motifs specific to atmospheric organic chemistry, and the combinations of these two types of motifs were pivotal for predictive performance.Our molecular descriptor, ATMOMACCS, can serve as a vital tool for advancing data-driven atmospheric science, addressing the need for more customized and accurate modelling in the field. Furthermore, the descriptor’s inherent interpretability and its strong performance in thermodynamic property prediction, with machine learning, show promise for further research in atmospheric chemistry.This work was supported by the VILMA (Virtual laboratory for molecular level atmospheric transformations) centre of excellence funded by the Academy of Finland under grant 346377. Besel, V. et al. (2023). Sci. Data 10, 1–11.Bianchi, F et al. (2019). Chem. Rev. 119, 3472–3509.Elm, J. et al. (2020) J. Aerosol Sci. 149, 105621.Ferraz-Caetano, J. et al. (2024). Chemosphere, 359, 142257Li, Y. et al. (2020). Atmos. Chem. Phys. 20, 8103–8122.Lumiaro, E. et al. (2021). Atmos. Chem. Phys. 21, 13227–13246.Lundberg, S. M. and Lee S.-I. (2017). Curran Associates Inc. 30, 9781510860964.Sandström, H. et al. (2024). Adv. Sci. 11, 2306235.Wang, C. et al. (2017). Atmos. Chem. Phys. 17, 7529–7540.

Atmospheric particles impact our climate and adversely affect air quality and human health (IPCC, 2022; Pozzer et al., 2023). Molecular emissions in the atmosphere can react with ozone and radicals, forming a diverse array of organic compounds that can drive particle formation (Bianchi et al., 2019). However, due to the vast number of potential reactions and precursors, the identities of many of these particle-forming products remain largely unknown. Electron ionization mass spectrometry (EI-MS) is a widely used tool for identifying organic compounds in aerosol particle samples (Franklin et al., 2022; Worton et al., 2017; Hamilton et al., 2004). High-confidence identification relies on matching recorded EI-MS spectra to reference spectra in mass spectral libraries, which contain reference data for known compounds (Laskin et al., 2018). However, the identification of many atmospheric compounds is limited by a lack of reference data for these species (Nozière et al., 2015; Sandström et al., 2024).In this presentation, I will introduce a simulated reference mass spectrometry dataset for atmospheric organic compounds. Using quantum chemistry and machine learning-based EI-MS simulation tools (Wei et al., 2019; Koopman & Grimme, 2021), we have simulated mass spectra for organic atmospheric compounds from the Master Chemical Mechanism (MCM v3.2, http://mcm.leeds.ac.uk/MCM, Wang et al., 2017). This simulated mass spectral dataset will be made publicly available to support future efforts to identify atmospheric organic compounds and advance our understanding of organic particle formation processes. This work was supported by the VILMA (Virtual laboratory for molecular level atmospheric transformations) centre of excellence funded by the Research Council of Finland under grant 346377. Bianchi, F., et al. (2019). Chemical Reviews, 119, 3472–3509.Franklin, E. B., et al. (2022). Atmospheric Measurement Techniques, 15, 3779–3803.Hamilton, J. F., et al. (2004). Atmospheric Chemistry and Physics, 4, 1279–1290.IPCC. (2022). Climate Change 2022: Impacts, Adaptation, and Vulnerability. Cambridge University Press.Koopman, J., & Grimme, S. (2021). Journal of the American Society for Mass Spectrometry, 32, 1735–1751.Laskin, J., et al. (2018). Analytical Chemistry, 90, 166–189.Nozière, B., et al. (2015). Chemical Reviews, 115, 3919–3983.Pozzer, A., et al. (2023). GeoHealth, 7, 24711403.Sandström, H., et al. (2024). Advanced Science, 11, 2306235.Wang, C., et al. (2017). Atmospheric Chemistry and Physics, 17, 7529–7540.Wei, J. N., et al. (2019). ACS Central Science, 5, 700–708.Worton, D. R., et al. (2017). Analyst, 142, 2395–2403.

New particle formation (NPF) is a significant source of atmospheric aerosols. The formation rate (J) quantifies the rate at which new particles form, and can be empirically determined from the particle number size distribution (PNSD). However, identifying the specific causes of NPF in urban areas remains challenging.It is believed that in urban environments, most NPF occurs through the clustering of sulfuric acid (H2SO4) with bases. However, in a chamber study, it was observed that also the clustering of H2SO4 with highly oxygenated organic molecules (HOMs) may greatly promote NPF (Riccobono, 2014, Science, 344(6185), 717-721).This research project employs AI data-driven approaches to predict J and examine the role of HOMs in NPF events.The data were collected in August 2022 in two close sites in Leipzig, Germany: an urban background (Leibniz Institute for Tropospheric Research) and a roadside (Eisenbahnstraße). The data include HOMs, H2SO4, and bases from the nitrate CIMS, PNSD measurements, pollutants such as BC, and meteorological variables. Previous research indicated a higher concentration of OOMs at the roadside (Brean, 2024, Environ. Sci. Technol. 58, 10664−10674), suggesting its potential impact on J.Preliminary results show that our data-driven model successfully predicted J values on a logarithmic scale with a mean absolute error of 0.33 at the urban background site and 0.63 at the roadside. Further analysis reveals the most significant contributors to predicting J, indicating that alongside H2SO4, various HOMs play a crucial role.

Pests significantly impact crop yields, leading to food insecurity. Pesticides are substances, or a mixture of substances, made to eliminate or control pests, or to regulate the growth of crops.Currently, more than 1000 pesticides are available in the market. However, their long-lasting environmental impact necessitates strict regulation, especially regarding their presence in food (FAO, 2022). Pesticides play also a role in the atmosphere as their volatilization can produce oxidized products through photolysis or OH reactions and they can be transported over large distances.The fundamental properties and behaviours of these compounds are still not well understood. Because of their complex structure, even low DFT level computations can be extremely expensive. This project applies machine learning (ML) tools to chemical ionization mass spectra to ultimately develop a technique capable of predicting spectra’s peak intensities and the chemical ionization mass spectrometry (CIMS) sensitivity to pesticides. The primary challenge is to develop a ML model that comprehensively explains ion-molecule interactions while minimizing computational costs.Our data set comprises different standard mixtures containing, in total, 716 pesticides measured with an orbitrap atmospheric pressure CIMS, with a multi-scheme chemical ionization inlet (MION) and five different concentrations (Rissanen et al, 2019; Partovi et al, 2023). The reagents of the ionization methods are CH2Br2, H2O, O2 and (CH3)2CO, generating respectively Br- , H3O+, O2- and [(CH3)2 CO + H]+ ions.The project follows a general ML workflow: after an exploratory analysis, the data are preprocessed and fed to the ML algorithm, which classifies the ionization method able to detect the molecule, and, therefore, predicts the peak intensity of each pesticide; the accuracy of the prediction can be retrieved after measuring the performance of the model.A random forest classifier was chosen to perform the classification of the ionization methods, to predict which one was able to detect each pesticide. The regression was performed with a kernel ridge regressor. Each algorithm was run with different types of molecular descriptors (topological fingerprint, MACCS keys and many-body tensor representation), to test which one was able to represent the molecular structure most accurately.The results of the exploratory analysis highlight different trends between the positive and negative ionization methods, suggesting that different ion-molecule mechanisms are involved (Figure 1). The classification reaches around 80% accuracy for each ionization method with all four molecular descriptors tested, while the regression can predict fairly well the logarithm of the intensities of each ionization method, reaching 0.5 of error with MACCS keys for (CH3)2CO reagent (Figure 2).Figure 1: Distribution of pesticide peak intensities for each reagent ion at five different concentrations.Figure 2: Comparison of the KRR performance on (CH3)2CO reagent data with four different molecular descriptors.

Chemical ionization mass spectrometry (CIMS) is widely used in atmospheric chemistry studies. However, due to the complex interactions between reagent ions and target compounds, chemical understanding remains limited and compound identification difficult. In this study, we apply machine learning to a reference dataset of pesticides in two standard solutions to build a model that can provide insights from CIMS analyses in atmospheric science. The CIMS measurements were performed with an orbitrap mass spectrometer coupled to a thermal desorption multi-scheme chemical ionization inlet unit (TD-MION-MS) with both negative and positive ionization modes utilizing Br^-, O₂^-, H₃O⁺ and (CH₃)₂COH⁺ (AceH⁺) as reagent ions. We then trained two machine learning methods on this data: 1) random forest (RF) for classifying if a pesticide can be detected with CIMS, and 2) kernel ridge regression (KRR) for predicting the expected CIMS signals. We compared their performance on five different representations of the molecular structure: the topological fingerprint (TopFP), the molecular access system keys (MACCS), a custom descriptor based on standard molecular properties (RDKitPROP), the Coulomb matrix (CM) and the many-body tensor representation (MBTR). The results indicate that MACCS outperforms the other descriptors. Our best classification model reaches a prediction accuracy of 0.85 ± 0.02 and a receiver operating characteristic curve area of 0.91 ± 0.01. Our best regression model reaches an accuracy of 0.44 ± 0.03 logarithmic units of the signal intensity. Subsequent feature importance analysis of the classifiers reveals that the most important structural fragments are NH and OH for the negative ionization schemes and nitrogen-containing groups for the positive ionization schemes.

The formation of aerosol particles in the atmosphere impacts air quality and climate change, but many of the organic molecules involved remain unknown. Machine learning could aid in identifying these compounds through accelerated analysis of molecular properties and detection characteristics. However, such progress is hindered by the current lack of curated datasets for atmospheric molecules and their associated properties. To tackle this challenge, we propose a similarity analysis that connects atmospheric compounds to existing large molecular datasets used for machine learning development. We find a small overlap between atmospheric and non-atmospheric molecules using standard molecular representations in machine learning applications. The identified out-of-domain character of atmospheric compounds is related to their distinct functional groups and atomic composition. Our investigation underscores the need for collaborative efforts to gather and share more molecular-level atmospheric chemistry data. The presented similarity based analysis can be used for future dataset curation for machine learning development in the atmospheric sciences.