Combining Group-Contribution Concept and Graph Neural Networks Toward Interpretable Molecular Property Models.

Quantitative structure-property relationships (QSPRs) are important tools to facilitate and accelerate the discovery of compounds with desired properties. While many QSPRs have been developed, they are associated with various shortcomings such as a lack of generalizability and modest accuracy. Albeit various machine-learning and deep-learning techniques have been integrated into such models, another shortcoming has emerged in the form of a lack of transparency and interpretability of such models. In this work, two interpretable graph neural network (GNN) models (attentive group-contribution (AGC) and group-contribution-based graph attention (GroupGAT)) are developed by integrating fundamentals using the concept of group contributions (GC). The interpretability consists of highlighting the substructure with the highest attention weights in the latent representation of the molecules using the attention mechanism. The proposed models showcased better performance compared to classical group-contribution models, as well as against various other GNN models describing the aqueous solubility, melting point, and enthalpies of formation, combustion, and fusion of organic compounds. The insights provided are consistent with insights obtained from the semiempirical GC models confirming that the proposed framework allows highlighting the important substructures of the molecules for a specific property.

Good Reporting Practice for Thermophysical and Thermochemical Property Measurements (IUPAC Technical Report).

Scientific projects frequently involve measurements of thermophysical, thermochemical, and other related properties of chemical compounds and materials. These measured property data have significant potential value for the scientific community, but incomplete and inaccurate reporting often hampers their utilization. The present IUPAC Technical Report summarizes the needs of chemical engineers and researchers as consumers of these data and shows how publishing practices can improve information transfer. In the Report, general principles of Good Reporting Practice are developed together with examples illustrating typical cases of reporting issues. Adoption of these principles will improve the quality, reproducibility, and usefulness of experimental data, bring a better level of consistency to results, and increase the efficiency and impact of research. Closely related to Good Reporting Practice, basic elements of Good Research Practice are also introduced with a goal to reduce the number of ambiguities and unresolved problems within the thermophysical property data domain.

Industrial ion-exchange chromatography development using discontinuous Galerkin methods coupled with forward sensitivity analysis.

In this work, a discontinuous Galerkin method coupled with forward sensitivity analysis (DG-FSA) is presented. The DG-FSA method is used to reduce computational cost required for model-based ion-exchange chromatography development using industrial load samples. As an example, the design of an anion-exchange chromatography step is considered. This step is used to purify an experimental peptide product called Protein G from Novo Nordisk A/S (Bagsværd, Denmark). The results demonstrate, that a fourth order DG-FSA method can reduce computational cost of inverse problems by a factor ×16 compared to a second (low) order DG-FSA method. Furthermore, the fourth-order DG-FSA method enable the computation of probability distributions of optimized processing conditions given uncertainty in model parameters or inputs. This analysis is not possible within a reasonable timeframe when applying the second (low) order DG-FSA method. The design procedure facilitates the optimization of the Protein G purification step. In an experimental validation run, the productivity is increased by 70% while sacrificing 4% yield at a similar purity constraint compared to an experiment with baseline performance.

Power-to-X electroscrubbing parameter analysis for biogas desulfurization.

A new power-to-X desulfurization technology has been examined. The technology uses only electricity to oxidize the hydrogen sulfide (H2S) found in biogas to elemental sulfur. The process works by using a scrubber where the biogas comes into contact with a chlorine containing liquid. This process is capable of removing close to 100% of H2S in biogas. In this paper a parameter analysis of process parameters has been carried out. In addition a long term test of the process has been performed. It has been found that the liquid flow rate has a small but notable influence on the process' performance on removing H2S. The efficiency of the process largely depends on total amount of H2S flowing through the scrubber. As the H2S concentration increases, the amount of chlorine required for the removal process is also increased. A high amount of chlorine in the solvent may lead to unwanted side reactions.

Pair correlation function integrals: computation and use.

We describe a method for extending radial distribution functions obtained from molecular simulations of pure and mixed molecular fluids to arbitrary distances. The method allows total correlation function integrals to be reliably calculated from simulations of relatively small systems. The long-distance behavior of radial distribution functions is determined by requiring that the corresponding direct correlation functions follow certain approximations at long distances. We have briefly described the method and tested its performance in previous communications [R. Wedberg, J. P. O'Connell, G. H. Peters, and J. Abildskov, Mol. Simul. 36, 1243 (2010); Fluid Phase Equilib. 302, 32 (2011)], but describe here its theoretical basis more thoroughly and derive long-distance approximations for the direct correlation functions. We describe the numerical implementation of the method in detail, and report numerical tests complementing previous results. Pure molecular fluids are here studied in the isothermal-isobaric ensemble with isothermal compressibilities evaluated from the total correlation function integrals and compared with values derived from volume fluctuations. For systems where the radial distribution function has structure beyond the sampling limit imposed by the system size, the integration is more reliable, and usually more accurate, than simple integral truncation.

Raw material quality assessment approaches comparison in pectin production.

Different opportunities are explored to evaluate quality variation in raw materials from biological origin. Assessment of raw materials attributes is an important step in a bio-based production as fluctuations in quality are a major source of process disturbance. This can be due to a variety of biological, seasonal, and supply scarcity reasons. The final properties of a product are invariably linked with the initial properties of the raw material. Thus, the operational conditions of a process can be tuned to drive the product to the required specification based on the quality assessment of the raw material being processed. Process analytical technology tools which enable this assessment in a far more informative and rapid manner than current industrial practices that rely on rule-of-thumb decisions are assessed. An example with citrus peels is used to demonstrate the conceptual and performance differences of distinct quality assessment approaches. The analysis demonstrates the advantage of characterization through multivariate data analysis coupled with a complementary spectroscopic technique, near-infrared spectroscopy. The quantitative comparative analysis of three different approaches, discriminant classification based on expert-knowledge, unsupervised classification, and spectroscopic correlation with reference physicochemical variables, is performed in the same dataset context. © 2018 Her Majesty the Queen in Right of Canada © 2018 American Institute of Chemical Engineers Biotechnol. Prog., 35: e2762, 2019.

Lumped parameter model for prediction of initial breakthrough profiles for the chromatographic capture of antibodies from a complex feedstock.

A simple mathematical model to predict initial breakthrough profiles from preparative chromatographic separations of biological macromolecules has been developed. A lumped parameter approach was applied, employing Langmuirian adsorption kinetics to describe the rate of mass transfer (MT) from the bulk liquid in the column to the bound state. Equilibrium and kinetic adsorption data were determined for six different packed bed chromatographic adsorbents: two derivatised with rProtein A; and four functionalised with synthetic low molecular weight ligands. All adsorption isotherms were well described by the Langmuir model, whereas the data fitting to kinetic batch experiments showed that the model was inadequate after the first approximately 25 min of adsorption for four of the six adsorbents. The model underestimated the dynamic Ig breakthough on packed beds of rProtein A Sepharose FF, MabSelect, MBI HyperCel, and MabSorbent A1P, applying a feedstock of 20-100% (v/v) clarified rabbit antiserum. However, when employing a maximum adsorption capacity 25% greater than that determined in batch binding studies, excellent agreement was obtained at all antiserum strengths for most adsorbents. Useful insights into scale-up and process design can be obtained by applying the model, without determining tentative parameters specific for each adsorbent and target protein concentration. However, the model parameters are solvent dependent so a prerequisite for its true applicability is that binding is both Langmuirian and essentially independent of the ionic strength of the feedstock applied.

Evaluation of binary solvent mixtures for efficient monoacylglycerol production by continuous enzymatic glycerolysis.

This study was aimed at evaluating different binary solvent mixtures for efficient industrial monoacylglycerol (MAG) production by enzymatic glycerolysis. Of all investigated cases, the binary mixture of tert-butanol:tert-pentanol (TB:TP) 80:20 vol % was the most suitable organic medium for continuous enzymatic glycerolysis, ensuring high MAG formation in a short time, reasonable solvent price, and easy handling during distillation/condensation processing. A minimum solvent dosage of 44-54 wt % of the reaction mixture was necessary to achieve high MAG yields of 47-56 wt %, within 20 min. The melting and boiling points of the TB:TP mixture were estimated to be 7 and 85 degrees C, respectively, using thermodynamic models. These predictions were in good agreement with experimentally determined values. In spite of the high reaction efficiency in the binary TB:TP system, the mixture of glycerol and sunflower oil (containing 97.1% triacylglycerol) yielded surprisingly a liquid/liquid phase split behavior even at high temperatures (>80 degrees C). This in contrast to thermodynamic model calculations suggested full miscibility in all proportions. These findings suggest that enhanced reaction efficiency in organic solvent also depends upon aspects other than the system homogeneity such as reduced viscosity, reduced mass transfer limitations, and the accessibility of the substrate to the active site of the enzyme.

Group-contribution based property estimation and uncertainty analysis for flammability-related properties.

This study presents new group contribution (GC) models for the prediction of Lower and Upper Flammability Limits (LFL and UFL), Flash Point (FP) and Auto Ignition Temperature (AIT) of organic chemicals applying the Marrero/Gani (MG) method. Advanced methods for parameter estimation using robust regression and outlier treatment have been applied to achieve high accuracy. Furthermore, linear error propagation based on covariance matrix of estimated parameters was performed. Therefore, every estimated property value of the flammability-related properties is reported together with its corresponding 95%-confidence interval of the prediction. Compared to existing models the developed ones have a higher accuracy, are simple to apply and provide uncertainty information on the calculated prediction. The average relative error and correlation coefficient are 11.5% and 0.99 for LFL, 15.9% and 0.91 for UFL, 2.0% and 0.99 for FP as well as 6.4% and 0.76 for AIT. Moreover, the temperature-dependence of LFL property was studied. A compound specific proportionality constant (K(LFL)) between LFL and temperature is introduced and an MG GC model to estimate K(LFL) is developed. Overall the ability to predict flammability-related properties including the corresponding uncertainty of the prediction can provide important information for a qualitative and quantitative safety-related risk assessment studies.

Protein dynamics in organic media at varying water activity studied by molecular dynamics simulation.

In nonaqueous enzymology, control of enzyme hydration is commonly approached by fixing the thermodynamic water activity of the medium. In this work, we present a strategy for evaluating the water activity in molecular dynamics simulations of proteins in water/organic solvent mixtures. The method relies on determining the water content of the bulk phase and uses a combination of Kirkwood-Buff theory and free energy calculations to determine corresponding activity coefficients. We apply the method in a molecular dynamics study of Candida antarctica lipase B in pure water and the organic solvents methanol, tert-butyl alcohol, methyl tert-butyl ether, and hexane, each mixture at five different water activities. It is shown that similar water activity yields similar enzyme hydration in the different solvents. However, both solvent and water activity are shown to have profound effects on enzyme structure and flexibility.