PolyID: Artificial Intelligence for Discovering Performance-Advantaged and Sustainable Polymers.

A necessary transformation for a sustainable economy is the transition from fossil-derived plastics to polymers derived from biomass and waste resources. While renewable feedstocks can enhance material performance through unique chemical moieties, probing the vast material design space by experiment alone is not practically feasible. Here, we develop a machine-learning-based tool, PolyID, to reduce the design space of renewable feedstocks to enable efficient discovery of performance-advantaged, biobased polymers. PolyID is a multioutput, graph neural network specifically designed to increase accuracy and to enable quantitative structure-property relationship (QSPR) analysis for polymers. It includes a novel domain-of-validity method that was developed and applied to demonstrate how gaps in training data can be filled to improve accuracy. The model was benchmarked with both a 20% held-out subset of the original training data and 22 experimentally synthesized polymers. A mean absolute error for the glass transition temperatures of 19.8 and 26.4 °C was achieved for the test and experimental data sets, respectively. Predictions were made on polymers composed of monomers from four databases that contain biologically accessible small molecules: MetaCyc, MINEs, KEGG, and BiGG. From 1.4 × 106 accessible biobased polymers, we identified five poly(ethylene terephthalate) (PET) analogues with predicted improvements to thermal and transport performance. Experimental validation for one of the PET analogues demonstrated a glass transition temperature between 85 and 112 °C, which is higher than PET and within the predicted range of the PolyID tool. In addition to accurate predictions, we show how the model's predictions are explainable through analysis of individual bond importance for a biobased nylon. Overall, PolyID can aid the biobased polymer practitioner to navigate the vast number of renewable polymers to discover sustainable materials with enhanced performance.

Upper-Bound Energy Minimization to Search for Stable Functional Materials with Graph Neural Networks.

The discovery of new materials in unexplored chemical spaces necessitates quick and accurate prediction of thermodynamic stability, often assessed using density functional theory (DFT), and efficient search strategies. Here, we develop a new approach to finding stable inorganic functional materials. We start by defining an upper bound to the fully relaxed energy obtained via DFT as the energy resulting from a constrained optimization over only cell volume. Because the fractional atomic coordinates for these calculations are known a priori, this upper bound energy can be quickly and accurately predicted with a scale-invariant graph neural network (GNN). We generate new structures via ionic substitution of known prototypes, and train our GNN on a new database of 128 000 DFT calculations comprising both fully relaxed and volume-only relaxed structures. By minimizing the predicted upper-bound energy, we discover new stable structures with over 99% accuracy (versus DFT). We demonstrate the method by finding promising new candidates for solid-state battery (SSB) electrolytes that not only possess the required stability, but also additional functional properties such as large electrochemical stability windows and high conduction ion fraction. We expect this proposed framework to be directly applicable to a wide range of design challenges in materials science.

Muconic acid production from glucose and xylose in Pseudomonas putida via evolution and metabolic engineering.

Muconic acid is a bioprivileged molecule that can be converted into direct replacement chemicals for incumbent petrochemicals and performance-advantaged bioproducts. In this study, Pseudomonas putida KT2440 is engineered to convert glucose and xylose, the primary carbohydrates in lignocellulosic hydrolysates, to muconic acid using a model-guided strategy to maximize the theoretical yield. Using adaptive laboratory evolution (ALE) and metabolic engineering in a strain engineered to express the D-xylose isomerase pathway, we demonstrate that mutations in the heterologous D-xylose:H+ symporter (XylE), increased expression of a major facilitator superfamily transporter (PP_2569), and overexpression of aroB encoding the native 3-dehydroquinate synthase, enable efficient muconic acid production from glucose and xylose simultaneously. Using the rationally engineered strain, we produce 33.7 g L-1 muconate at 0.18 g L-1 h-1 and a 46% molar yield (92% of the maximum theoretical yield). This engineering strategy is promising for the production of other shikimate pathway-derived compounds from lignocellulosic sugars.

A quantitative metric for organic radical stability and persistence using thermodynamic and kinetic features.

Long-lived organic radicals are promising candidates for the development of high-performance energy solutions such as organic redox batteries, transistors, and light-emitting diodes. However, "stable" organic radicals that remain unreactive for an extended time and that can be stored and handled under ambient conditions are rare. A necessary but not sufficient condition for organic radical stability is the presence of thermodynamic stabilization, such as conjugation with an adjacent π-bond or lone-pair, or hyperconjugation with a σ-bond. However, thermodynamic factors alone do not result in radicals with extended lifetimes: many resonance-stabilized radicals are transient species that exist for less than a millisecond. Kinetic stabilization is also necessary for persistence, such as steric effects that inhibit radical dimerization or reaction with solvent molecules. We describe a quantitative approach to map organic radical stability, using molecular descriptors intended to capture thermodynamic and kinetic considerations. The comparison of an extensive dataset of quantum chemical calculations of organic radicals with experimentally-known stable radical species reveals a region of this feature space where long-lived radicals are located. These descriptors, based upon maximum spin density and buried volume, are combined into a single metric, the radical stability score, that outperforms thermodynamic scales based on bond dissociation enthalpies in identifying remarkably long-lived radicals. This provides an objective and accessible metric for use in future molecular design and optimization campaigns. We demonstrate this approach in identifying Pareto-optimal candidates for stable organic radicals.

Bayesian Inference for Integrating Yarrowia lipolytica Multiomics Datasets with Metabolic Modeling.

Optimizing the metabolism of microbial cell factories for yields and titers is a critical step for economically viable production of bioproducts and biofuels. In this process, tuning the expression of individual enzymes to obtain the desired pathway flux is a challenging step, in which data from separate multiomics techniques must be integrated with existing biological knowledge to determine where changes should be made. Following a design-build-test-learn strategy, building on recent advances in Bayesian metabolic control analysis, we identify key enzymes in the oleaginous yeast Yarrowia lipolytica that correlate with the production of itaconate by integrating a metabolic model with multiomics measurements. To this extent, we quantify the uncertainty for a variety of key parameters, known as flux control coefficients (FCCs), needed to improve the bioproduction of target metabolites and statistically obtain key correlations between the measured enzymes and boundary flux. Based on the top five significant FCCs and five correlated enzymes, our results show phosphoglycerate mutase, acetyl-CoA synthetase (ACSm), carbonic anhydrase (HCO3E), pyrophosphatase (PPAm), and homoserine dehydrogenase (HSDxi) enzymes in rate-limiting reactions that can lead to increased itaconic acid production.

Real-time prediction of 1H and 13C chemical shifts with DFT accuracy using a 3D graph neural network.

Nuclear magnetic resonance (NMR) is one of the primary techniques used to elucidate the chemical structure, bonding, stereochemistry, and conformation of organic compounds. The distinct chemical shifts in an NMR spectrum depend upon each atom's local chemical environment and are influenced by both through-bond and through-space interactions with other atoms and functional groups. The in silico prediction of NMR chemical shifts using quantum mechanical (QM) calculations is now commonplace in aiding organic structural assignment since spectra can be computed for several candidate structures and then compared with experimental values to find the best possible match. However, the computational demands of calculating multiple structural- and stereo-isomers, each of which may typically exist as an ensemble of rapidly-interconverting conformations, are expensive. Additionally, the QM predictions themselves may lack sufficient accuracy to identify a correct structure. In this work, we address both of these shortcomings by developing a rapid machine learning (ML) protocol to predict 1H and 13C chemical shifts through an efficient graph neural network (GNN) using 3D structures as input. Transfer learning with experimental data is used to improve the final prediction accuracy of a model trained using QM calculations. When tested on the CHESHIRE dataset, the proposed model predicts observed 13C chemical shifts with comparable accuracy to the best-performing DFT functionals (1.5 ppm) in around 1/6000 of the CPU time. An automated prediction webserver and graphical interface are accessible online at http://nova.chem.colostate.edu/cascade/. We further demonstrate the model in three applications: first, we use the model to decide the correct organic structure from candidates through experimental spectra, including complex stereoisomers; second, we automatically detect and revise incorrect chemical shift assignments in a popular NMR database, the NMRShiftDB; and third, we use NMR chemical shifts as descriptors for determination of the sites of electrophilic aromatic substitution.

Thermodynamic and Kinetic Modeling of Co-utilization of Glucose and Xylose for 2,3-BDO Production by Zymomonas mobilis.

Prior engineering of the ethanologen Zymomonas mobilis has enabled it to metabolize xylose and to produce 2,3-butanediol (2,3-BDO) as a dominant fermentation product. When co-fermenting with xylose, glucose is preferentially utilized, even though xylose metabolism generates ATP more efficiently during 2,3-BDO production on a BDO-mol basis. To gain a deeper understanding of Z. mobilis metabolism, we first estimated the kinetic parameters of the glucose facilitator protein of Z. mobilis by fitting a kinetic uptake model, which shows that the maximum transport capacity of glucose is seven times higher than that of xylose, and glucose is six times more affinitive to the transporter than xylose. With these estimated kinetic parameters, we further compared the thermodynamic driving force and enzyme protein cost of glucose and xylose metabolism. It is found that, although 20% more ATP can be yielded stoichiometrically during xylose utilization, glucose metabolism is thermodynamically more favorable with 6% greater cumulative Gibbs free energy change, more economical with 37% less enzyme cost required at the initial stage and sustains the advantage of the thermodynamic driving force and protein cost through the fermentation process until glucose is exhausted. Glucose-6-phosphate dehydrogenase (g6pdh), glyceraldehyde-3-phosphate dehydrogenase (gapdh) and phosphoglycerate mutase (pgm) are identified as thermodynamic bottlenecks in glucose utilization pathway, as well as two more enzymes of xylose isomerase and ribulose-5-phosphate epimerase in xylose metabolism. Acetolactate synthase is found as potential engineering target for optimized protein cost supporting unit metabolic flux. Pathway analysis was then extended to the core stoichiometric matrix of Z. mobilis metabolism. Growth was simulated by dynamic flux balance analysis and the model was validated showing good agreement with experimental data. Dynamic FBA simulations suggest that a high agitation is preferable to increase 2,3-BDO productivity while a moderate agitation will benefit the 2,3-BDO titer. Taken together, this work provides thermodynamic and kinetic insights of Z. mobilis metabolism on dual substrates, and guidance of bioengineering efforts to increase hydrocarbon fuel production.

Importance of Engineered and Learned Molecular Representations in Predicting Organic Reactivity, Selectivity, and Chemical Properties.

Machine-readable chemical structure representations are foundational in all attempts to harness machine learning for the prediction of reactivities, selectivities, and chemical properties directly from molecular structure. The featurization of discrete chemical structures into a continuous vector space is a critical phase undertaken before model selection, and the development of new ways to quantitatively encode molecules is an active area of research. In this Account, we highlight the application and suitability of different representations, from expert-guided "engineered" descriptors to automatically "learned" features, in different prediction tasks relevant to organic and organometallic chemistry, where differing amounts of training data are available. These tasks include statistical models of stereo- and enantioselectivity, thermochemistry, and kinetics developed using experimental and quantum chemical data.The use of expert-guided molecular descriptors provides an opportunity to incorporate chemical knowledge, domain expertise, and physical constraints into statistical modeling. In applications to stereoselective organic and organometallic catalysis, where data sets may be relatively small and 3D-geometries and conformations play an important role, mechanistically informed features can be used successfully to obtain predictive statistical models that are also chemically interpretable. We provide an overview of several recent applications of this approach to obtain quantitative models for reactivity and selectivity, where topological descriptors, quantum mechanical calculations of electronic and steric properties, along with conformational ensembles, all feature as essential ingredients of the molecular representations used.Alternatively, more flexible, general-purpose molecular representations such as attributed molecular graphs can be used with machine learning approaches to learn the complex relationship between a structure and prediction target. This approach has the potential to out-perform more traditional representation methods such as "hand-crafted" molecular descriptors, particularly as data set sizes grow. One area where this is particularly relevant is in the use of large sets of quantum mechanical data to train quantitative structure-property relationships. A general approach toward curating useful data sets and training highly accurate graph neural network models is discussed in the context of organic bond dissociation enthalpies, where this strategy outperforms regression using precomputed descriptors.Finally, we describe how graph neural network predictions can be incorporated into mechanistically informed statistical models of chemical reactivity and selectivity. Once trained, this approach avoids the expensive computational overhead associated with quantum mechanical calculations, while maintaining chemical interpretability. We illustrate examples for which fast predictions of bond dissociation enthalpy and of the identities of radicals formed through cleavage of a molecule's weakest bond are used in simple physical models of site-selectivity and reactivity.

Advances in integrative structural biology: Towards understanding protein complexes in their cellular context.

Microorganisms rely on protein interactions to transmit signals, react to stimuli, and grow. One of the best ways to understand these protein interactions is through structural characterization. However, in the past, structural knowledge was limited to stable, high-affinity complexes that could be crystallized. Recent developments in structural biology have revolutionized how protein interactions are characterized. The combination of multiple techniques, known as integrative structural biology, has provided insight into how large protein complexes interact in their native environment. In this mini-review, we describe the past, present, and potential future of integrative structural biology as a tool for characterizing protein interactions in their cellular context.

Quantum chemical calculations for over 200,000 organic radical species and 40,000 associated closed-shell molecules.

The stabilities of radicals play a central role in determining the thermodynamics and kinetics of many reactions in organic chemistry. In this data descriptor, we provide consistent and validated quantum chemical calculations for over 200,000 organic radical species and 40,000 associated closed-shell molecules containing C, H, N and O atoms. These data consist of optimized 3D geometries, enthalpies, Gibbs free energy, vibrational frequencies, Mulliken charges and spin densities calculated at the M06-2X/def2-TZVP level of theory, which was previously found to have a favorable trade-off between experimental accuracy and computational efficiency. We expect this data to be useful in the further development of machine learning techniques to predict reaction pathways, bond strengths, and other phenomena closely related to organic radical chemistry.

Software and Methods for Computational Flux Balance Analysis.

As genetic engineering of organisms has grown easier and more precise, computational modeling of metabolic systems has played an increasingly important role in both guiding experimental interventions and in understanding the results of metabolic perturbations.

Publisher Correction: Prediction of organic homolytic bond dissociation enthalpies at near chemical accuracy with sub-second computational cost.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

A perspective on biomass-derived biofuels: From catalyst design principles to fuel properties.

The hazards to health and the environment associated with the transportation sector include smog, particulate matter, and greenhouse gas emissions. Conversion of lignocellulosic biomass into biofuels has the potential to provide significant amounts of infrastructure-compatible liquid transportation fuels that reduce those hazardous materials. However, the development of these technologies is inefficient, due to: (i) the lack of a priori fuel property consideration, (ii) poor shared vocabulary between process chemists and fuel engineers, and (iii) modern and future engines operating outside the range of traditional autoignition metrics such as octane or cetane numbers. In this perspective, we describe an approach where we follow a "fuel-property first" design methodology with a sequence of (i) identifying the desirable fuel properties for modern engines, (ii) defining molecules capable of delivering those properties, and (iii) designing catalysts and processes that can produce those molecules from a candidate feedstock in a specific conversion process. Computational techniques need to be leveraged to minimize expenses and experimental efforts on low-promise options. This concept is illustrated with current research information available for biomass conversion to fuels via catalytic fast pyrolysis and hydrotreating; outstanding challenges and research tools necessary for a successful outcome are presented.

Prediction of organic homolytic bond dissociation enthalpies at near chemical accuracy with sub-second computational cost.

Bond dissociation enthalpies (BDEs) of organic molecules play a fundamental role in determining chemical reactivity and selectivity. However, BDE computations at sufficiently high levels of quantum mechanical theory require substantial computing resources. In this paper, we develop a machine learning model capable of accurately predicting BDEs for organic molecules in a fraction of a second. We perform automated density functional theory (DFT) calculations at the M06-2X/def2-TZVP level of theory for 42,577 small organic molecules, resulting in 290,664 BDEs. A graph neural network trained on a subset of these results achieves a mean absolute error of 0.58 kcal mol-1 (vs DFT) for BDEs of unseen molecules. We further demonstrate the model on two applications: first, we rapidly and accurately predict major sites of hydrogen abstraction in the metabolism of drug-like molecules, and second, we determine the dominant molecular fragmentation pathways during soot formation.

Performance-advantaged ether diesel bioblendstock production by a priori design.

Lignocellulosic biomass offers a renewable carbon source which can be anaerobically digested to produce short-chain carboxylic acids. Here, we assess fuel properties of oxygenates accessible from catalytic upgrading of these acids a priori for their potential to serve as diesel bioblendstocks. Ethers derived from C2 and C4 carboxylic acids are identified as advantaged fuel candidates with significantly improved ignition quality (>56% cetane number increase) and reduced sooting (>86% yield sooting index reduction) when compared to commercial petrodiesel. The prescreening process informed conversion pathway selection toward a C11 branched ether, 4-butoxyheptane, which showed promise for fuel performance and health- and safety-related attributes. A continuous, solvent-free production process was then developed using metal oxide acidic catalysts to provide improved thermal stability, water tolerance, and yields. Liter-scale production of 4-butoxyheptane enabled fuel property testing to confirm predicted fuel properties, while incorporation into petrodiesel at 20 vol % demonstrated 10% improvement in ignition quality and 20% reduction in intrinsic sooting tendency. Storage stability of the pure bioblendstock and 20 vol % blend was confirmed with a common fuel antioxidant, as was compatibility with elastomeric components within existing engine and fueling infrastructure. Technoeconomic analysis of the conversion process identified major cost drivers to guide further research and development. Life-cycle analysis determined the potential to reduce greenhouse gas emissions by 50 to 271% relative to petrodiesel, depending on treatment of coproducts.

Bayesian inference of metabolic kinetics from genome-scale multiomics data.

Modern biological tools generate a wealth of data on metabolite and protein concentrations that can be used to help inform new strain designs. However, learning from these data to predict how a cell will respond to genetic changes, a key need for engineering, remains challenging. A promising technique for leveraging omics measurements in metabolic modeling involves the construction of kinetic descriptions of the enzymatic reactions that occur within a cell. Parameterizing these models from biological data can be computationally difficult, since methods must also quantify the uncertainty in model parameters resulting from the observed data. While the field of Bayesian inference offers a wide range of methods for efficiently estimating distributions in parameter uncertainty, such techniques are poorly suited to traditional kinetic models due to their complex rate laws and resulting nonlinear dynamics. In this paper, we employ linear-logarithmic kinetics to simplify the calculation of steady-state flux distributions and enable efficient sampling and inference methods. We demonstrate that detailed information on the posterior distribution of parameters can be obtained efficiently at a variety of problem scales, including nearly genome-scale kinetic models trained on multiomics datasets. These results allow modern Bayesian machine learning tools to be leveraged in understanding biological data and in developing new, efficient strain designs.

Message-passing neural networks for high-throughput polymer screening.

Machine learning methods have shown promise in predicting molecular properties, and given sufficient training data, machine learning approaches can enable rapid high-throughput virtual screening of large libraries of compounds. Graph-based neural network architectures have emerged in recent years as the most successful approach for predictions based on molecular structure and have consistently achieved the best performance on benchmark quantum chemical datasets. However, these models have typically required optimized 3D structural information for the molecule to achieve the highest accuracy. These 3D geometries are costly to compute for high levels of theory, limiting the applicability and practicality of machine learning methods in high-throughput screening applications. In this study, we present a new database of candidate molecules for organic photovoltaic applications, comprising approximately 91 000 unique chemical structures. Compared to existing datasets, this dataset contains substantially larger molecules (up to 200 atoms) as well as extrapolated properties for long polymer chains. We show that message-passing neural networks trained with and without 3D structural information for these molecules achieve similar accuracy, comparable to state-of-the-art methods on existing benchmark datasets. These results therefore emphasize that for larger molecules with practical applications, near-optimal prediction results can be obtained without using optimized 3D geometry as an input. We further show that learned molecular representations can be leveraged to reduce the training data required to transfer predictions to a new density functional theory functional.

Approaches to Computational Strain Design in the Multiomics Era.

Modern omics analyses are able to effectively characterize the genetic, regulatory, and metabolic phenotypes of engineered microbes, yet designing genetic interventions to achieve a desired phenotype remains challenging. With recent developments in genetic engineering techniques, timelines associated with building and testing strain designs have been greatly reduced, allowing for the first time an efficient closed loop iteration between experiment and analysis. However, the scale and complexity associated with multi-omics datasets complicates manual biological reasoning about the mechanisms driving phenotypic changes. Computational techniques therefore form a critical part of the Design-Build-Test-Learn (DBTL) cycle in metabolic engineering. Traditional statistical approaches can reduce the dimensionality of these datasets and identify common motifs among high-performing strains. While successful in many studies, these methods do not take full advantage of known connections between genes, proteins, and metabolic networks. There is therefore a growing interest in model-aided design, in which modeling frameworks from systems biology are used to integrate experimental data and generate effective and non-intuitive design predictions. In this mini-review, we discuss recent progress and challenges in this field. In particular, we compare methods augmenting flux balance analysis with additional constraints from fluxomic, genomic, and metabolomic datasets and methods employing kinetic representations of individual metabolic reactions, and machine learning. We conclude with a discussion of potential future directions for improving strain design predictions in the omics era and remaining experimental and computational hurdles.

Prediction of reaction knockouts to maximize succinate production by Actinobacillus succinogenes.

Succinate is a precursor of multiple commodity chemicals and bio-based succinate production is an active area of industrial bioengineering research. One of the most important microbial strains for bio-based production of succinate is the capnophilic gram-negative bacterium Actinobacillus succinogenes, which naturally produces succinate by a mixed-acid fermentative pathway. To engineer A. succinogenes to improve succinate yields during mixed acid fermentation, it is important to have a detailed understanding of the metabolic flux distribution in A. succinogenes when grown in suitable media. To this end, we have developed a detailed stoichiometric model of the A. succinogenes central metabolism that includes the biosynthetic pathways for the main components of biomass-namely glycogen, amino acids, DNA, RNA, lipids and UDP-N-Acetyl-α-D-glucosamine. We have validated our model by comparing model predictions generated via flux balance analysis with experimental results on mixed acid fermentation. Moreover, we have used the model to predict single and double reaction knockouts to maximize succinate production while maintaining growth viability. According to our model, succinate production can be maximized by knocking out either of the reactions catalyzed by the PTA (phosphate acetyltransferase) and ACK (acetyl kinase) enzymes, whereas the double knockouts of PEPCK (phosphoenolpyruvate carboxykinase) and PTA or PEPCK and ACK enzymes are the most effective in increasing succinate production.

Metabolic Engineering of Actinobacillus succinogenes Provides Insights into Succinic Acid Biosynthesis.

Actinobacillus succinogenes, a Gram-negative facultative anaerobe, exhibits the native capacity to convert pentose and hexose sugars to succinic acid (SA) with high yield as a tricarboxylic acid (TCA) cycle intermediate. In addition, A. succinogenes is capnophilic, incorporating CO2 into SA, making this organism an ideal candidate host for conversion of lignocellulosic sugars and CO2 to an emerging commodity bioproduct sourced from renewable feedstocks. In this work, we report the development of facile metabolic engineering capabilities in A. succinogenes, enabling examination of SA flux determinants via knockout of the primary competing pathways-namely, acetate and formate production-and overexpression of the key enzymes in the reductive branch of the TCA cycle leading to SA. Batch fermentation experiments with the wild-type and engineered strains using pentose-rich sugar streams demonstrate that the overexpression of the SA biosynthetic machinery (in particular, the enzyme malate dehydrogenase) enhances flux to SA. Additionally, removal of competitive carbon pathways leads to higher-purity SA but also triggers the generation of by-products not previously described from this organism (e.g., lactic acid). The resultant engineered strains also lend insight into energetic and redox balance and elucidate mechanisms governing organic acid biosynthesis in this important natural SA-producing microbe.IMPORTANCE Succinic acid production from lignocellulosic residues is a potential route for enhancing the economic feasibility of modern biorefineries. Here, we employ facile genetic tools to systematically manipulate competing acid production pathways and overexpress the succinic acid-producing machinery in Actinobacillus succinogenes Furthermore, the resulting strains are evaluated via fermentation on relevant pentose-rich sugar streams representative of those from corn stover. Overall, this work demonstrates genetic modifications that can lead to succinic acid production improvements and identifies key flux determinants and new bottlenecks and energetic needs when removing by-product pathways in A. succinogenes metabolism.