Sigma profiles in deep learning: towards a universal molecular descriptor
Sigma profiles are shown to be universal molecular descriptors in the prediction of material properties using powerful deep learning methodologies.
SEARCH · Search NASA
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Sigma profiles are shown to be universal molecular descriptors in the prediction of material properties using powerful deep learning methodologies.
Machine learning has proven to be a powerful tool for accelerating biofuel development. Although numerous models are available to predict a range of properties using chemical descriptors, there is a trade-off between interpretability and performance. Neural networks provide predictive models with high accuracy at the expense of some interpretability, while simpler models such as linear regression often lack in accuracy. In addition to model architecture, feature selection is also critical for developing interpretable and accurate predictive models. We present a method for systematically selecting molecular descriptor features and developing interpretable machine learning models without sacrificing accuracy. Our method simplifies the process of selecting features by reducing feature multicollinearity and enables discoveries of new relationships between global properties and molecular descriptors. To demonstrate our approach, we developed models for predicting melting point, boiling point, flash point, yield sooting index, and net heat of combustion with the help of the Tree-based Pipeline Optimization Tool (TPOT). For training, we used publicly available experimental data for up to 8351 molecules. Our models accurately predict various molecular properties for organic molecules (mean absolute percent error (MAPE) ranges from 3.3% to 10.5%) and provide a set of features that are well-correlated to the property. This method enables researchers to explore sets of features that significantly contribute to the prediction of the property, offering new scientific insights. To help accelerate early stage biofuel research and development, we also integrated the data and models into a open-source, interactive web tool.
In this work we report on the development of the Signature Molecular Descriptor (or Signature) for use in the solution of inverse design problems as well as in highthroughput screening applications. The ultimate goal of using Signature is to identify novel and non-intuitive chemical structures with optimal predicted properties for a given application. We demonstrate this in three studies: green solvent design, glucocorticoid receptor ligand design and the design of inhibitors for Factor XIa. In many areas of engineering, compounds are designed and/or modified in incremental ways which rely upon heuristics or institutional knowledge. Often multiple experiments are performed and the optimal compound is identified in this brute-force fashion. Perhaps a traditional chemical scaffold is identified and movement of a substituent group around a ring constitutes the whole of the design process. Also notably, a chemical being evaluated in one area might demonstrate properties very attractive in another area and serendipity was the mechanism for solution. In contrast to such approaches, computer-aided molecular design (CAMD) looks to encompass both experimental and heuristic-based knowledge into a strategy that will design a molecule on a computer to meet a given target. Depending on the algorithm employed, the molecule which is designed might be quite novel (re: no CAS registration number) and/or non-intuitive relative to what is known about the problem at hand. While CAMD is a fairly recent strategy (dating to the early 1980s), it contains a variety of bottlenecks and limitations which have prevented the technique from garnering more attention in the academic, governmental and industrial institutions. A main reason for this is how the molecules are described in the computer. This step can control how models are developed for the properties of interest on a given problem as well as how to go from an output of the algorithm to an actual chemical structure. This report provides details on a technique to describe molecules on a computer, called Signature, as well as the computer-aided molecule design algorithm built around Signature. Two applications are provided of the CAMD algorithm with Signature. The first describes the design of green solvents based on data in the GlaxoSmithKline (GSK) Solvent Selection Guide. The second provides novel non-steroidal glucocorticoid receptor ligands with some optimally predicted properties. In addition to using the CAMD algorithm with Signature, it is demonstrated how to employ Signature in a high-throughput screening study. Here, after classifying both active and inactive inhibitors for the protein Factor XIa using Signature, the model developed is used to screen a large, publicly-available database called PubChem for the most active compounds.
The ability Gram-negative pathogens have at adapting and protecting themselves against antibiotics has increasingly become a public health threat. Data-driven models identifying molecular properties that correlate with outer membrane (OM) permeation and growth inhibition while avoiding efflux could guide the discovery of novel classes of antibiotics. Here we evaluate 174 molecular descriptors in 1260 antimicrobial compounds and study their correlations with antibacterial activity in Gram-negative Pseudomonas aeruginosa. The descriptors are derived from traditional approaches quantifying the compounds’ intrinsic physicochemical properties, together with, bacterium-specific from ensemble docking of compounds targeting specific MexB binding pockets, and all-atom molecular dynamics simulations in different subregions of the OM model. Using these descriptors and the measured inhibitory concentrations, we design a statistical protocol to identify predictors of OM permeation/inhibition. We find consistent rules across most of our data highlighting the role of the interaction between the compounds and the OM. An implementation of the rules uncovered in our study is shown, and it demonstrates the accuracy of our approach in a set of previously unseen compounds. Our analysis sheds new light on the key properties drug candidates need to effectively permeate/inhibit P. aeruginosa, and opens the gate to similar data-driven studies in other Gram-negative pathogens.
Amine-based solvents are likely to play an important role in CO{sub 2} capture applications in the future, and the identification of amines with superior performance will facilitate their use in CO{sub 2} capture. While some improvements in performance will be achieved through process modifications, modifying the CO{sub 2} capture performance of an amine also implies in part an ability to modify the reactions between the amine and CO{sub 2} through development of new functionalized amines. We present a computational study of trends in the reactions between CO{sub 2} and functionalized amines with a focus on identifying molecular descriptors that determine trends in reactivity. We examine the formation of bicarbonate and carbamate species on three classes of functionalized amines: alkylamines, alkanolamines, and fluorinated alkylamines including primary, secondary and tertiary amines in each class. These functional groups span electron-withdrawing to donating behavior, hydrogen-bonding, extent of functionalization, and proximity effects of the functional groups. Electron withdrawing groups tend to destabilize CO{sub 2} reaction products, whereas electron-donating groups tend to stabilize CO{sub 2} reaction products. Hydrogen bonding stabilizes CO{sub 2} reaction products. Electronic structure descriptors based on electronegativity were found to describe trends in the bicarbonate formation energy. A chemical correlation was observed between the carbamate formation energy and the carbamic acid formation energy. The local softness on the reacting N in the amine was found to partially explain trends carbamic acid formation energy.
Ready biodegradability is a key property for evaluating the long-term effects of chemicals on the environment and human health. As such, it is used as a screening test for the assessment of persistent, bioaccumulative and toxic substances. Regulators encourage the use of non-testing methods, such as in silico models, to save money and time. A dataset of 757 chemicals was collected to assess the performance of four freely available in silico models that predict ready biodegradability. They were applied to develop a new consensus method that prioritizes the use of each individual model according to its performance on chemical subsets driven by the presence or absence of different molecular descriptors. This consensus method was capable of almost eliminating unpredictable chemicals, while the performance of combined models was substantially improved with respect to that of the individual models. - Highlights: • Consensus method to predict ready biodegradability by prioritizing multiple QSARs. • Consensus reduced the amount of unpredictable chemicals to less than 2%. • Performance increased with the number of QSAR models considered. • The absence of 2D atom pairs contributed significantly to the consensus model.
Density functional reactivity theory (DFRT) employs the electron density of a molecule and its related quantities such as gradient and Laplacian to describe its structure and reactivity properties. Proper descriptions at both molecular (global) and atomic (local) levels are equally important and illuminating. In this work, we make use of Bader's zero-flux partition scheme and consider atomic contributions for a few global reactivity descriptors in DFRT, including the density-based quantification of steric effect and related indices. Earlier, we proved that these quantities are intrinsically correlated for atomic and molecular systems [S. B. Liu, J. Chem. Phys. 126, 191107 (2007); ibid. 126, 244103 (2007)]. In this work, a new basin-based integration algorithm has been implemented, whose reliability and effectiveness have been extensively examined. We also investigated a list of simple hydrocarbon systems and different scenarios of bonding processes, including stretching, bending, and rotating. Interesting changing patterns for the atomic and molecular values of these quantities have been revealed for different systems. This work not only confirms the strong correlation between these global reactivity descriptors for molecular systems, as theoretically proven earlier by us, it also provides new and unexpected changing patterns for their atomic values, which can be employed to understand the origin and nature of chemical phenomena.
While machine learning has found increasing use in predicting the properties of polymeric materials with only a knowledge of chain architecture, determining the molecular factors underpinning properties (“interpretable AI”) has remained less well explored. We show that encoding chain chemistry in commonly employed formats, e.g., binary-valued fingerprints, leads to uniqueness issues during the hashing process to save storage space. This is because the hashing algorithm can map several chemical moieties into the same bit. These issues carry over into the ML algorithms, especially for “inverse” design and interpretable AI, and cannot be avoided by changing the length of the fingerprint. Using MACCS key featurizations of monomer repeats resolves some of these issues, and we show that a few substructures consistently appear in top features for maximizing permeability across several gases and ML models. These are carbon–carbon double bonds (as in polyacetylenes) especially when they are associated with methyl groups (found in branching architectures). Here these results, derived from the limited data set of ~ 500 polymers with experimental gas permeation data, are in agreement with physical insight and thus provide a robust foundation which could further enable study of these material classes through detailed experiments and simulations.
In this study, we tried to assess the utility of the structural information of drugs for predicting human acute toxicity from in vitro basal cytotoxicity, and to interpret the informative quality and the pharmacokinetic meaning of each structural descriptor. For this, human acute toxicity data of 67 drugs were taken from literature with their basal cytotoxicity data, and used to develop predictive models. A series of multiple linear regression analyses were performed to construct feasible regression models by combining molecular descriptors and cytotoxicity data. We found that although the molecular descriptors alone had only moderate correlation with human acute toxicity, they were highly useful for explaining the discrepancy between in vitro cytotoxicity and human acute toxicity. Among many possible models, we selected the most explanatory models by changing the number and the type of combined molecular descriptors. The results showed that our selected models had high predictive power (R{sup 2}: between 0.7 and 0.87). Our analysis indicated that those successful models increased the prediction accuracies by providing the information on human pharmacokinetic parameters which are the major reason for the difference between human acute toxicity and cytotoxicity. In addition, we performed a clustering analysis on selected molecular descriptors to assess their informative qualities. The results indicated that the number of single bonds, the number of hydrogen bond donors and valence connectivity indices are closely related to linking cytotoxicity to acute toxicity, which provides insightful explanation about human toxicity beyond cytotoxicity.
High-potential molecules derived from biomass sources may suitably replace or supplement traditional nonrenewable hydrocarbon fuels to reduce pollution and fuel processing cost. Experimental property testing of these bioproducts is usually conducted years after initial bench-scale experiments, due to high experimental costs and/or high volume requirements. However, neglecting to conduct property testing early in the pathway development cycle can lead to investments spent on scaling-up production of bioproducts and biofuels that do not perform as expected. Instead, machine-learning techniques can be used to develop quantitative structure–property relationships for molecules using a relatively large training set of molecular descriptor data. For this study, we compiled measured properties, IR spectra, and molecular descriptors of bio-based molecules from databases and published studies for training models of bioproduct properties. We trained regression models with molecular descriptors and will compare results of different estimators. This study describes the first steps towards a performance prediction tool for bio-based alternative fuels. Keywords: Machine learning, biofuels, jet fuels, fuel properties
The combination of machine learning (ML) models with chemistry-related tasks requires the description of molecular structures in a machine-readable way. The nature of these so-called molecular descriptors has a direct and major impact on the performance of ML models and remains an open problem in the field. Structural descriptors like SMILES strings or molecular graphs lack size-independence and can be memory intensive. Machine-learned descriptors can be of low dimensionality and constant size but lack physical significance and human interpretability. Sigma profiles, which are unnormalized histograms of the surface charge distributions of solvated molecules, combine physical significance with low dimensionality and size-independence, making them a suitable candidate for a universal molecular descriptor. However, their widespread adoption in ML applications requires open access to sigma profile generation, which is currently not available. This work details the development of OpenSPGen – an open-source tool for generating sigma profiles. Also presented are studies on the effect of different settings on the efficacy of the generated sigma profiles at predicting thermophysical material properties when used as inputs to a Gaussian process as a simple surrogate ML model. We find that a higher level of theory does not translate to more accurate results. We also provide further recommendations for sigma profile calculation and use in ML models.
A mechanistic understanding of metal–organic framework (MOF) synthesis and scale-up remains underexplored due to the complex nature of the interactions of their building blocks. In this work, we investigate the collective assembly of building units at the early stages of MOF nucleation, using MIL-101(Cr) as a prototypical example. Using large-scale molecular dynamics simulations, we observe that the choice of solvent (water and N,N-dimethylformamide), the introduction of ions (Na+ and F–) and the relative populations of MIL-101(Cr) half-secondary building unit (half-SBU) isomers have a strong influence on the cluster formation process. Additionally, the shape, size, nucleation and growth rates, crystallinity, and short and long-range order largely vary depending on the synthesis conditions. We evaluate these properties as they naturally emerge when interpreting the self-assembly of MOF nuclei as the time evolution of an undirected graph. Solution-induced conformational complexity and ionic concentration have a dramatic effect on the morphology of clusters emerging during assembly. While pure solvents lead to the rapid formation of a small number of large clusters, the presence of ions in aqueous solutions results in smaller clusters and slower nucleation. This diversity is captured by the key features of the graph representation. Principle component analysis on graph properties reveals that only a small number of molecular descriptors is needed to deconvolute MOF self-assembly. Furthermore, descriptors such as the average coordination number between half-SBUs and fractal dimension are of particular interest as they can be can be followed experimentally by techniques like by time-resolved spectroscopy. Ultimately, graph theory emerges as an approach that can be used to understand complex processes revealing molecular descriptors accessible by both simulation and experiment.
The pregnane X receptor (PXR) has a key role in regulating the metabolism and transport of structurally diverse endogenous and exogenous compounds. Activation of PXR has the potential to initiate adverse effects, causing drug–drug interactions, and perturbing normal physiological functions. Therefore, identification of PXR ligands would be valuable information for pharmaceutical and toxicological research. In the present study, we developed a quantitative structure–activity relationship (QSAR) model for the identification of PXR ligands using data based on a human PXR binding assay. A total of 631 molecules, representing a variety of chemical structures, constituted the training set of the model. Cross-validation of the model showed a sensitivity of 82%, a specificity of 85%, and a concordance of 84%. The developed model provided knowledge about molecular descriptors that may influence the binding of molecules to PXR. The model was used to screen a large inventory of environmental chemicals, of which 47% was found to be within domain of the model. Approximately 35% of the chemicals within domain were predicted to be PXR ligands. The predicted PXR ligands were found to be overrepresented among chemicals predicted to cause adverse effects, such as genotoxicity, teratogenicity, estrogen receptor activation and androgen receptor antagonism compared to chemicals not causing these effects. The developed model may be useful as a tool for predicting potential PXR ligands and for providing mechanistic information of toxic effects of chemicals. -- Highlights: ► Global QSAR model for the identification of PXR ligands was developed. ► Molecular descriptors that may influence PXR binding were identified. ► 35% of a large set of environmental chemicals were predicted to be PXR ligands. ► Predicted PXR binding was associated with various adverse effects.
We investigate possible improvements in the accuracy of semiempirical quantum chemistry (SQC) methods through the use of machine learning (ML) models for the parameters. For a given class of compounds, ML techniques require sufficiently large training sets to develop ML models that can be used for adapting SQC parameters to reflect changes in molecular composition and geometry. The ML-SQC approach allows the automatic tuning of SQC parameters for individual molecules, thereby improving the accuracy without deteriorating transferability to molecules with molecular descriptors very different from those in the training set. The performance of this approach is demonstrated for the semiempirical OM2 method using a set of 6095 constitutional isomers C7H10O2, for which accurate ab initio atomization enthalpies are available. The ML-OM2 results show improved average accuracy and a much reduced error range compared with those of standard OM2 results, with mean absolute errors in atomization enthalpies dropping from 6.3 to 1.7 kcal/mol. They are also found to be superior to the results from specific OM2 reparameterizations (rOM2) for the same set of isomers. The ML-SQC approach thus holds promise for fast and reasonably accurate high-throughput screening of materials and molecules.
Ensemble learning approach based decision treeboost (DTB) and decision tree forest (DTF) models are introduced in order to establish quantitative structure–toxicity relationship (QSTR) for the prediction of toxicity of 1450 diverse chemicals. Eight non-quantum mechanical molecular descriptors were derived. Structural diversity of the chemicals was evaluated using Tanimoto similarity index. Stochastic gradient boosting and bagging algorithms supplemented DTB and DTF models were constructed for classification and function optimization problems using the toxicity end-point in T. pyriformis. Special attention was drawn to prediction ability and robustness of the models, investigated both in external and 10-fold cross validation processes. In complete data, optimal DTB and DTF models rendered accuracies of 98.90%, 98.83% in two-category and 98.14%, 98.14% in four-category toxicity classifications. Both the models further yielded classification accuracies of 100% in external toxicity data of T. pyriformis. The constructed regression models (DTB and DTF) using five descriptors yielded correlation coefficients (R{sup 2}) of 0.945, 0.944 between the measured and predicted toxicities with mean squared errors (MSEs) of 0.059, and 0.064 in complete T. pyriformis data. The T. pyriformis regression models (DTB and DTF) applied to the external toxicity data sets yielded R{sup 2} and MSE values of 0.637, 0.655; 0.534, 0.507 (marine bacteria) and 0.741, 0.691; 0.155, 0.173 (algae). The results suggest for wide applicability of the inter-species models in predicting toxicity of new chemicals for regulatory purposes. These approaches provide useful strategy and robust tools in the screening of ecotoxicological risk or environmental hazard potential of chemicals. - Graphical abstract: Importance of input variables in DTB and DTF classification models for (a) two-category, and (b) four-category toxicity intervals in T. pyriformis data. Generalization and predictive abilities of the constructed (c) DTB and (d) DTF regression models to predict the T. pyriformis toxicity of diverse chemicals. - Highlights: • Ensemble learning (EL) based models constructed for toxicity prediction of chemicals • Predictive models used a few simple non-quantum mechanical molecular descriptors. • EL-based DTB/DTF models successfully discriminated toxic and non-toxic chemicals. • DTB/DTF regression models precisely predicted toxicity of chemicals in multi-species. • Proposed EL based models can be used as tool to predict toxicity of new chemicals.