Paolo Elvati

Extracting meaningful features from complex, high-dimensional datasets across scientific domains remains challenging. Current methods often struggle with scalability, limiting their applicability to large datasets, or make restrictive assumptions about feature-property relationships, hindering their ability to capture complex interactions. BoUTS's general and scalable feature selection algorithm surpasses these limitations by identifying both universal features relevant to all datasets and task-specific features predictive for specific subsets. Evaluated on seven diverse chemical regression datasets, BoUTS achieves state-of-the-art feature sparsity while generally maintaining prediction accuracy comparable to specialized methods. Notably, BoUTS's universal features enable domain-specific knowledge transfer between datasets, and we expect these results to be broadly useful to manually-guided inverse problems. Beyond its current application, BoUTS holds potential for elucidating data-poor systems by leveraging information from similar data-rich systems.Scientific Contribution: BoUTS selects nonlinear, universally informative features across multiple datasets. We identify crucial "universal features" across seven real-world chemistry datasets, which enhance cross-dataset interpretability and selection stability. BoUTS is highly scalable and is applicable to tabular data from many domains, and our results identify connections between seemingly unrelated chemical domains.

The physical growth of Polycyclic Aromatic Compounds (PACs) to soot particles plays a significant role in understanding the chemistry of soot formation. Insights into the process can be gained from PACs’ free energy of dimerization landscape. However, because the infeasibly large space of possible PAC dimers cannot be exhaustively simulated, researchers must train machine learning models on a subset of data to impute the rest. To this end, we propose and assess an active learning approach to discovering the optimal PACs for training a machine learning model to predict PACs’ association and dissociation free energies. The comparison between active learning and random sampling showed that active learning has faster loss convergence, requiring fewer training samples to reach the same level of accuracy. The trained model accurately modeled unseen PACs and exhibited robustness against changes in the sampling space used to train the model. More broadly, this work shows how active learning can optimize the design and improve the understanding of more expensive models in specific domains.

Abstract Nanoparticles formed in nonthermal plasmas (NTPs) can have unique properties and applications. However, modeling their growth in these environments presents significant challenges due to the non-equilibrium nature of NTPs, making them computationally expensive to describe. In this work, we address the challenges associated with accelerating the estimation of parameters needed for these models. Specifically, we explore how different machine learning models can be tailored to improve prediction outcomes. We apply these methods to reactive classical molecular dynamics data, which capture the processes associated with colliding silane fragments in NTPs. These reactions exemplify processes where qualitative trends are clear, but their quantification is challenging, hard to generalize, and requires time-consuming simulations. Our results demonstrate that good prediction performance can be achieved when appropriate loss functions are implemented and correct invariances are imposed. While the diversity of molecules used in the training set is critical for accurate prediction, our findings indicate that only a fraction (15%–25%) of the energy and temperature sampling is required to achieve high levels of accuracy. This suggests a substantial reduction in computational effort is possible for similar systems.

Nanoparticles formed in nonthermal plasmas (NTPs) can have unique properties and applications. However, modeling their growth in these environments presents significant challenges due to the non-equilibrium nature of NTPs, making them computationally expensive to describe. In this work, we address the challenges associated with accelerating the estimation of parameters needed for these models. Specifically, we explore how different machine learning models can be tailored to improve prediction outcomes. We apply these methods to reactive classical molecular dynamics data, which capture the processes associated with colliding silane fragments in NTPs. These reactions exemplify processes where qualitative trends are clear, but their quantification is challenging, hard to generalize, and requires time-consuming simulations. Our results demonstrate that good prediction performance can be achieved when appropriate loss functions are implemented and correct invariances are imposed. While the diversity of molecules used in the training set is critical for accurate prediction, our findings indicate that only a fraction (15%-25%) of the energy and temperature sampling is required to achieve high levels of accuracy. This suggests a substantial reduction in computational effort is possible for similar systems.

Nanoparticles (NPs) formed in nonthermal plasmas (NTPs) can have unique properties and applications. However, modeling their growth in these environments presents significant challenges due to the non-equilibrium nature of NTPs, making them computationally expensive to describe. In this work, we address the challenges associated with accelerating the estimation of parameters needed for these models. Specifically, we explore how different machine learning models can be tailored to improve prediction outcomes. We apply these methods to reactive classical molecular dynamics data, which capture the processes associated with colliding silane fragments in NTPs. These reactions exemplify processes where qualitative trends are clear, but their quantification is challenging, hard to generalize, and requires time-consuming simulations. Our results demonstrate that good prediction performance can be achieved when appropriate loss functions are implemented and correct invariances are imposed. While the diversity of molecules used in the training set is critical for accurate prediction, our findings indicate that only a fraction (15-25\%) of the energy and temperature sampling is required to achieve high levels of accuracy. This suggests a substantial reduction in computational effort is possible for similar systems.

Abstract Motivation Understanding the mechanisms and structural mappings between molecules and pathway classes are critical for design of reaction predictors for synthesizing new molecules. This article studies the problem of prediction of classes of metabolic pathways (series of chemical reactions occurring within a cell) in which a given biochemical compound participates. We apply a hybrid machine learning approach consisting of graph convolutional networks used to extract molecular shape features as input to a random forest classifier. In contrast to previously applied machine learning methods for this problem, our framework automatically extracts relevant shape features directly from input SMILES representations, which are atom-bond specifications of chemical structures composing the molecules. Results Our method is capable of correctly predicting the respective metabolic pathway class of 95.16% of tested compounds, whereas competing methods only achieve an accuracy of 84.92% or less. Furthermore, our framework extends to the task of classification of compounds having mixed membership in multiple pathway classes. Our prediction accuracy for this multi-label task is 95.62%. We analyze the relative importance of various global physicochemical features to the pathway class prediction problem and show that simple linear/logistic regression models can predict the values of these global features from the shape features extracted using our framework. Availability and implementation https://github.com/baranwa2/MetabolicPathwayPrediction.

Extracting meaningful features from complex, high-dimensional datasets across scientific domains remains challenging. Current methods often struggle with scalability, limiting their applicability to large datasets, or make restrictive assumptions about feature-property relationships, hindering their ability to capture complex interactions. BoUTS's general and scalable feature selection algorithm surpasses these limitations to identify both universal features relevant to all datasets and task-specific features predictive for specific subsets. Evaluated on seven diverse chemical regression datasets, BoUTS achieves state-of-the-art feature sparsity while maintaining prediction accuracy comparable to specialized methods. Notably, BoUTS's universal features enable domain-specific knowledge transfer between datasets, and suggest deep connections in seemingly-disparate chemical datasets. We expect these results to have important repercussions in manually-guided inverse problems. Beyond its current application, BoUTS holds immense potential for elucidating data-poor systems by leveraging information from similar data-rich systems. BoUTS represents a significant leap in cross-domain feature selection, potentially leading to advancements in various scientific fields.

Finite-size effects in the static structure factor S(k) are analyzed for an amorphous substance. As the number of particles is reduced, S(0) increases greatly, up to an order of magnitude. Meanwhile, there is a decrease in the height of the first peak S_{peak}. These finite-size effects are modeled accurately by the Binder formula for S(0) and our empirical formula for S_{peak}. Procedures are suggested to correct for finite-size effects in S(k) data and in the hyperuniformity index H≡S(0)/S_{peak}. These principles generally apply to S(k) obtained from particle positions in noncrystalline substances. The amorphous substance we simulate is a two-dimensional liquid, with a soft Yukawa interaction modeling a dusty plasma experiment.

<p>A diverse range of polycyclic <a href="https://www.sciencedirect.com/topics/chemical-engineering/aromatic-compound" target="_blank">aromatic compounds</a> (PACs) is thought to exist in flame environments before and during soot inception. This work seeks to develop a machine learning (ML)-based soot inception model that considers detailed and diverse PAC properties such as <a href="https://www.sciencedirect.com/topics/chemical-engineering/oxygenation" target="_blank">oxygenation</a>, aliphatic content, radical character, size, and shape. To this end, temporal rates of change of PAC properties were computed by the stochastic modelling code SNapS2 and used as input to an ML model that predicts soot inception rate. The model is trained using experimentally-derived soot inception rates for three atmospheric pressure laminar premixed ethylene/air flames. An ML model (kernel ridge regression with a linear kernel) was developed to predict the soot inception rate in the three <a href="https://www.sciencedirect.com/topics/engineering/premixed-flame" target="_blank">premixed flames</a>. The soot inception rate predictions from this SNapS2-informed ML model outperformed the predictions from both the advanced soot <a href="https://www.sciencedirect.com/topics/engineering/computational-fluid-dynamic-modeling" target="_blank">modelling CFD</a> code CoFlame and an ML model which used CFD-determined inputs (temperature and species concentrations). The final model had an R^2 value of approximately 0.71 and a <a href="https://www.sciencedirect.com/topics/engineering/mean-absolute-error" target="_blank">mean absolute error</a> approximately 25% of the target values. The performance of the SNapS2-informed model suggests that detailed PAC properties are important to consider in inception modelling. While expanding this approach to other types of flames and fuels is crucial for future improvement to the model’s accuracy and generality, this methodology provides a successful framework for the current system. The success of this method demonstrates that ML can offer improvements in accuracy compared to current <a href="https://www.sciencedirect.com/topics/engineering/computational-fluid-dynamics" target="_blank">CFD</a> inception models and the highlights the potential for ML in soot predictions.</p>

<p>A diverse range of polycyclic <a href="https://www.sciencedirect.com/topics/chemical-engineering/aromatic-compound" target="_blank">aromatic compounds</a> (PACs) is thought to exist in flame environments before and during soot inception. This work seeks to develop a machine learning (ML)-based soot inception model that considers detailed and diverse PAC properties such as <a href="https://www.sciencedirect.com/topics/chemical-engineering/oxygenation" target="_blank">oxygenation</a>, aliphatic content, radical character, size, and shape. To this end, temporal rates of change of PAC properties were computed by the stochastic modelling code SNapS2 and used as input to an ML model that predicts soot inception rate. The model is trained using experimentally-derived soot inception rates for three atmospheric pressure laminar premixed ethylene/air flames. An ML model (kernel ridge regression with a linear kernel) was developed to predict the soot inception rate in the three <a href="https://www.sciencedirect.com/topics/engineering/premixed-flame" target="_blank">premixed flames</a>. The soot inception rate predictions from this SNapS2-informed ML model outperformed the predictions from both the advanced soot <a href="https://www.sciencedirect.com/topics/engineering/computational-fluid-dynamic-modeling" target="_blank">modelling CFD</a> code CoFlame and an ML model which used CFD-determined inputs (temperature and species concentrations). The final model had an R^2 value of approximately 0.71 and a <a href="https://www.sciencedirect.com/topics/engineering/mean-absolute-error" target="_blank">mean absolute error</a> approximately 25% of the target values. The performance of the SNapS2-informed model suggests that detailed PAC properties are important to consider in inception modelling. While expanding this approach to other types of flames and fuels is crucial for future improvement to the model’s accuracy and generality, this methodology provides a successful framework for the current system. The success of this method demonstrates that ML can offer improvements in accuracy compared to current <a href="https://www.sciencedirect.com/topics/engineering/computational-fluid-dynamics" target="_blank">CFD</a> inception models and the highlights the potential for ML in soot predictions.</p>

Amyloid nanofibers are abundant in microorganisms and are integral components of many biofilms, serving various purposes, from virulent to structural. Nonetheless, the precise characterization of bacterial amyloid nanofibers has been elusive, with incomplete and contradicting results. The present work focuses on the molecular details and characteristics of PSMα1-derived functional amyloids present in Staphylococcus aureus biofilms, using a combination of computational and experimental techniques, to develop a model that can aid the design of compounds to control amyloid formation. Results from molecular dynamics simulations, guided and supported by spectroscopy and microscopy, show that PSMα1 amyloid nanofibers present a helical structure formed by two protofilaments, have an average diameter of about 12 nm, and adopt a left-handed helicity with a periodicity of approximately 72 nm. The chirality of the self-assembled nanofibers, an intrinsic geometric property of its constituent peptides, is central to determining the fibers’ lateral growth.

A growing body of work has linked key biological activities to the mechanical properties of cellular membranes, and as a means of identification. Here, we present a computational approach to simulate and compare the vibrational spectra in the low-THz region for mammalian and bacterial membranes, investigating the effect of membrane asymmetry and composition, as well as the conserved frequencies of a specific cell. We find that asymmetry does not impact the vibrational spectra, and the impact of sterols depends on the mobility of the components of the membrane. We demonstrate that vibrational spectra can be used to distinguish between membranes and, therefore, could be used in identification of different organisms. The method presented, here, can be immediately extended to other biological structures (e.g., amyloid fibers, polysaccharides, and protein-ligand structures) in order to fingerprint and understand vibrations of numerous biologically-relevant nanoscale structures.

A growing body of work has linked key biological activities to the mechanical properties of cellular membranes, and as a means of identification. Here, we present a computational approach to simulate and compare the vibrational spectra in the low-THz region for mammalian and bacterial membranes, investigating the effect of membrane asymmetry and composition, as well as the conserved frequencies of a specific cell. We find that asymmetry does not impact the vibrational spectra, and the impact of sterols depends on the mobility of the components of the membrane. We demonstrate that vibrational spectra can be used to distinguish between membranes and, therefore, could be used in identification of different organisms. The method presented, here, can be immediately extended to other biological structures (e.g., amyloid fibers, polysaccharides, and protein-ligand structures) in order to fingerprint and understand vibrations of numerous biologically-relevant nanoscale structures.