Prediction of GFP spectral properties using artificial neural network

Identifying structure-absorption relationships and predicting absorption strength of non-fullerene acceptors for organic photovoltaics

Non-fullerene acceptors (NFAs) are excellent light harvesters, yet the origin of their high optical extinction is not well understood. In this work, we investigate the absorption strength of NFAs by building a database of time-dependent density functional theory (TDDFT) calculations of ∼500 π-conjugated molecules. The calculations are first validated by comparison with experimental measurements in solution and solid state using common fullerene and non-fullerene acceptors. We find that the molar extinction coefficient (ε _d,max) shows reasonable agreement between calculation in vacuum and experiment for molecules in solution, highlighting the effectiveness of TDDFT for predicting optical properties of organic π-conjugated molecules. We then perform a statistical analysis based on molecular descriptors to identify which features are important in defining the absorption strength. This allows us to identify structural features that are correlated with high absorption strength in NFAs and could be used to guide molecular design: highly absorbing NFAs should possess a planar, linear, and fully conjugated molecular backbone with highly polarisable heteroatoms. We then exploit a random decision forest algorithm to draw predictions for ε _d,max using a computational framework based on extended tight-binding Hamiltonians, which shows reasonable predicting accuracy with lower computational cost than TDDFT. This work provides a general understanding of the relationship between molecular structure and absorption strength in π-conjugated organic molecules, including NFAs, while introducing predictive machine-learning models of low computational cost.

More Is Not Always Better: Local Models Provide Accurate Predictions of Spectral Properties of Porphyrins

The development of new functional materials based on porphyrins requires fast and accurate prediction of their spectral properties. The available models in the literature for absorption wavelength and extinction coefficient of the Soret band have low accuracy for this class of compounds. We collected spectral data for porphyrins to extend the literature set and compared the performance of global and local models for their modelling using different machine learning methods. Interestingly, extension of the public database contributed models with lower accuracies compared to the models, which we built using porphyrins only. The later model calculated acceptable RMSE = 2.61 for prediction of the absorption band of 335 porphyrins synthesized in our laboratory, but had a low accuracy (RMSE = 0.52) for extinction coefficient. A development of models using only compounds from our laboratory significantly decreased errors for these compounds (RMSE = 0.5 and 0.042 for absorption band and extinction coefficient, respectively), but limited their applicability only to these homologous series. When developing models, one should clearly keep in mind their potential use and select a strategy that could contribute the most accurate predictions for the target application. The models and data are publicly available.

More Is Not Always Better: Local Models Provide Accurate Predictions of Spectral Properties of Porphyrins

The development of new functional materials based on porphyrins requires fast and accurate prediction of their spectral properties. The available models in the literature for absorption wavelength and extinction coefficient of the Soret band have low accuracy for this class of compounds. We collected spectral data for porphyrins to extend the literature set and compared the performance of global and local models for their modelling using different machine learning methods. Interestingly, extension of the public database contributed models with lower accuracies compared to the models, which we built using porphyrins only. The later model calculated acceptable RMSE = 2.61 for prediction of the absorption band of 335 porphyrins synthesized in our laboratory, but had a low accuracy (RMSE = 0.52) for extinction coefficient. A development of models using only compounds from our laboratory significantly decreased errors for these compounds (RMSE = 0.5 and 0.042 for absorption band and extinction coefficient, respectively), but limited their applicability only to these homologous series. When developing models, one should clearly keep in mind their potential use and select a strategy that could contribute the most accurate predictions for the target application. The models and data are publicly available.

More Is Not Always Better: Local Models Provide Accurate Predictions of Spectral Properties of Porphyrins

The development of new functional materials based on porphyrins requires fast and accurate prediction of their spectral properties. The available models in the literature for absorption wavelength and extinction coefficient of the Soret band have low accuracy for this class of compounds. We collected spectral data for porphyrins to extend the literature set and compared the performance of global and local models for their modelling using different machine learning methods. Interestingly, extension of the public database contributed models with lower accuracies compared to the models, which we built using porphyrins only. The later model calculated acceptable RMSE = 2.61 for prediction of the absorption band of 335 porphyrins synthesized in our laboratory, but had a low accuracy (RMSE = 0.52) for extinction coefficient. A development of models using only compounds from our laboratory significantly decreased errors for these compounds (RMSE = 0.5 and 0.042 for absorption band and extinction coefficient, respectively), but limited their applicability only to these homologous series. When developing models, one should clearly keep in mind their potential use and select a strategy that could contribute the most accurate predictions for the target application. The models and data are publicly available.

Ligand-based quantitative structural assessments of SARS-CoV-2 3CLpro inhibitors: An analysis in light of structure-based multi-molecular modeling evidences

Due to COVID-19, the whole world is undergoing a devastating situation, but treatment with no such drug candidates still has been established exclusively. In that context, 69 diverse chemicals with potential SARS-CoV-2 3CL^pro inhibitory property were taken into consideration for building different internally and externally validated linear (SW-MLR and GA-MLR), non-linear (ANN and SVM) QSAR, and HQSAR models to identify important structural and physicochemical characters required for SARS-CoV-2 3CL^pro inhibition. Importantly, 2-oxopyrrolidinyl methyl and benzylester functions, and methylene (hydroxy) sulphonic acid warhead group, were crucial for retaining higher SARS-CoV-2 3CL^pro inhibition. These GA-MLR and HQSAR models were also applied to predict some already repurposed drugs. As per the GA-MLR model, curcumin, ribavirin, saquinavir, sepimostat, and remdesivir were found to be the potent ones, whereas according to the HQSAR model, lurasidone, saquinavir, lopinavir, elbasvir, and paritaprevir were the highly effective SARS-CoV-2 3CL^pro inhibitors. The binding modes of those repurposed drugs were also justified by the molecular docking, molecular dynamics (MD) simulation, and binding energy calculations conducted by several groups of researchers. This current work, therefore, may be able to find out important structural parameters to accelerate the COVID-19 drug discovery processes in the future.

FPredX: Interpretable models for the prediction of spectral maxima, brightness, and oligomeric states of fluorescent proteins

Fluorescent protein (FP) design is among the challenging protein design problems due to the tradeoffs among multiple properties to be optimized. Despite the accumulated efforts in design and characterization, progress has been slow in gaining a full understanding of sequence-property relationships to tackle the multiobjective design problem in FPs. In this study, we approach this problem by developing FPredX, a collection of gradient-boosted decision tree models, which mapped FP sequences to four major design targets of FPs, including excitation maximum, emission maximum, brightness, and oligomeric state. By training using one-hot encoded multiple aligned sequences with hyperparameters optimization in each model, FPredX models showed excellent prediction performance for all target properties compared with existing methods. We further interpreted the FPredX models by comparing the importance of positions along the aligned FP sequence to the predictive performance and suggested positions, which showed differential importance deemed by FPredX models to the prediction of each target property.

Molecular excited states through a machine learning lens

Theoretical simulations of electronic excitations and associated processes in molecules are indispensable for fundamental research and technological innovations. However, such simulations are notoriously challenging to perform with quantum mechanical methods. Advances in machine learning open many new avenues for assisting molecular excited-state simulations. In this Review, we track such progress, assess the current state of the art and highlight the critical issues to solve in the future. We overview a broad range of machine learning applications in excited-state research, which include the prediction of molecular properties, improvements of quantum mechanical methods for the calculations of excited-state properties and the search for new materials. Machine learning approaches can help us understand hidden factors that influence photo-processes, leading to a better control of such processes and new rules for the design of materials for optoelectronic applications.

Machine Learning for Absorption Cross Sections

We present a machine learning (ML) method to accelerate the nuclear ensemble approach (NEA) for computing absorption cross sections. ML-NEA is used to calculate cross sections on vast ensembles of nuclear geometries to reduce the error due to insufficient statistical sampling. The electronic properties-excitation energies and oscillator strengths-are calculated with a reference electronic structure method only for a relatively few points in the ensemble. The KREG model (kernel-ridge-regression-based ML combined with the RE descriptor) as implemented in MLatom is used to predict these properties for the remaining tens of thousands of points in the ensemble without incurring much of additional computational cost. We demonstrate for two examples, benzene and a 9-dicyanomethylene derivative of acridine, that ML-NEA can produce statistically converged cross sections even for very challenging cases and even with as few as several hundreds of training points.

Rational Design of Colchicine Derivatives as anti-HIV Agents via QSAR and Molecular Docking

BACKGROUND

Human immunodeficiency virus (HIV) is an infective agent that causes an acquired immunodeficiency syndrome (AIDS). Therefore, the rational design of inhibitors for preventing the progression of the disease is required.

OBJECTIVE

This study aims to construct quantitative structure-activity relationship (QSAR) models, molecular docking and newly rational design of colchicine and derivatives with anti-HIV activity.

METHODS

A data set of 24 colchicine and derivatives with anti-HIV activity were employed to develop the QSAR models using machine learning methods (e.g. multiple linear regression (MLR), artificial neural network (ANN) and support vector machine (SVM)), and to study a molecular docking.

RESULTS

The significant descriptors relating to the anti-HIV activity included JGI2, Mor24u, Gm and R8p+ descriptors. The predictive performance of the models gave acceptable statistical qualities as observed by correlation coefficient (Q2) and root mean square error (RMSE) of leave-one out cross-validation (LOO-CV) and external sets. Particularly, the ANN method outperformed MLR and SVM methods that displayed LOO-CV 2 Q and RMSELOO-CV of 0.7548 and 0.5735 for LOOCV set, and Ext 2 Q of 0.8553 and RMSEExt of 0.6999 for external validation. In addition, the molecular docking of virus-entry molecule (gp120 envelope glycoprotein) revealed the key interacting residues of the protein (cellular receptor, CD4) and the site-moiety preferences of colchicine derivatives as HIV entry inhibitors for binding to HIV structure. Furthermore, newly rational design of colchicine derivatives using informative QSAR and molecular docking was proposed.

CONCLUSION

These findings serve as a guideline for the rational drug design as well as potential development of novel anti-HIV agents.

AutoWeka: toward an automated data mining software for QSAR and QSPR studies

In biology and chemistry, a key goal is to discover novel compounds affording potent biological activity or chemical properties. This could be achieved through a chemical intuition-driven trial-and-error process or via data-driven predictive modeling. The latter is based on the concept of quantitative structure-activity/property relationship (QSAR/QSPR) when applied in modeling the biological activity and chemical properties, respectively, of compounds. Data mining is a powerful technology underlying QSAR/QSPR as it harnesses knowledge from large volumes of high-dimensional data via multivariate analysis. Although extremely useful, the technicalities of data mining may overwhelm potential users, especially those in the life sciences. Herein, we aim to lower the barriers to access and utilization of data mining software for QSAR/QSPR studies. AutoWeka is an automated data mining software tool that is powered by the widely used machine learning package Weka. The software provides a user-friendly graphical interface along with an automated parameter search capability. It employs two robust and popular machine learning methods: artificial neural networks and support vector machines. This chapter describes the practical usage of AutoWeka and relevant tools in the development of predictive QSAR/QSPR models.

AVAILABILITY

The software is freely available at http://www.mt.mahidol.ac.th/autoweka.

Predictive QSAR modeling of aldose reductase inhibitors using Monte Carlo feature selection

This study explores the chemical space and quantitative structure-activity relationship (QSAR) of a set of 60 sulfonylpyridazinones with aldose reductase inhibitory activity. The physicochemical properties of the investigated compounds were described by a total of 3230 descriptors comprising of 6 quantum chemical descriptors and 3224 molecular descriptors. A subset of 5 descriptors was selected from the aforementioned pool by means of Monte Carlo (MC) feature selection coupled to multiple linear regression (MLR). Predictive QSAR models were then constructed by MLR, support vector machine and artificial neural network, which afforded good predictive performance as deduced from internal and external validation. The investigated models are capable of accounting for the origins of aldose reductase inhibitory activity and could be utilized in predicting this property in screening for novel and robust compounds.

Exploring the chemical space of aromatase inhibitors

Aromatase, a rate-limiting enzyme catalyzing the conversion of androgen to estrogen, is overexpressed in human breast cancer tissue. Aromatase inhibitors (AIs) have been used for the treatment of estrogen-dependent breast cancer in post-menopausal women by blocking the biosynthesis of estrogen. The undesirable side effects in current AIs have called for continued pursuit for novel candidates with aromatase inhibitory properties. This study explores the chemical space of all known AIs as a function of their physicochemical properties by means of univariate (i.e., statistical and histogram analysis) and multivariate (i.e., decision tree and principal component analysis) approaches in order to understand the origins of aromatase inhibitory activity. Such a non-redundant set of AIs spans a total of 973 compounds encompassing both steroidal and non-steroidal inhibitors. Substructure analysis of the molecular fragments provided pertinent information on the structural features important for ligands providing high and low aromatase inhibition. Analyses were performed on data sets stratified according to their structural scaffolds (i.e., steroids and non-steroids) and bioactivities (i.e., actives and inactives). These analyses have uncover a set of rules characteristic to active and inactive AIs as well as revealing the constituents giving rise to potent aromatase inhibition.

QSAR MODEL OF THE QUORUM-QUENCHING N-ACYL-HOMOSERINE LACTONE LACTONASE ACTIVITY

A quantitative structure-activity relationship (QSAR) study was performed to model the lactonolysis activity of N-acyl-homoserine lactone lactonase. A data set comprising of 20 homoserine lactones and related compounds was taken from the work of Wang et al. Quantum chemical descriptors were calculated using the semiempirical AM1 method. Partial least squares regression was utilized to construct a predictive model. This computational approach reliably reproduced the lactonolysis activity with high accuracy as illustrated by the correlation coefficient in excess of 0.9. It is demonstrated that the combined use of quantum chemical descriptors with partial least squares regression are suitable for modeling the AHL lactonolysis activity.

Molecular Docking of Aromatase Inhibitors

Aromatase is an enzyme that plays a critical role in the development of estrogen receptor positive breast cancer. As aromatase catalyzes the aromatization of androstenedione to estrone, a naturally occurring estrogen, it is a promising drug target for therapeutic management. The undesirable effects found in aromatase inhibitors (AIs) that are in clinical use necessitate the discovery of novel AIs with higher selectivity, less toxicity and improving potency. In this study, we elucidate the binding mode of all three generations of AI drugs to the crystal structure of aromatase by means of molecular docking. It was demonstrated that the docking protocol could reliably reproduce the interaction of aromatase with its substrate with an RMSD of 1.350 Å. The docking study revealed that polar (D309, T310, S478 and M374), aromatic (F134, F221 and W224) and non-polar (A306, A307, V370, L372 and L477) residues were important for interacting with the AIs. The insights gained from the study herein have great potential for the design of novel AIs.

QSPR study of absorption maxima of organic dyes for dye-sensitized solar cells based on 3D descriptors

A quantitative structure-property relationship (QSPR) study was performed for the prediction of the absorption maxima (lambda(max)) of organic dyes for dye-sensitized solar cells (DSSCs). The entire set of 70 dyes was divided into a training set of 53 dyes and a test set of 17 dyes according to Kennard and Stones algorithm. Three-dimensional (3D) descriptors were calculated to represent the dye molecules. A ten-descriptor model, with a squared correlation coefficient (R(2)) of 0.9543 and a standard error of estimation (s) of 14.7 nm, was produced by using the stepwise multilinear regression analysis (MLRA) on the training set. The reliability of the proposed model was further illustrated using various evaluation techniques: leave-one-out cross-validation procedure, randomization tests, and validation through the external test set. All descriptors involved in the model were derived solely from the chemical structure of the dye molecules, which makes the model very useful to estimate the lambda(max) of dyes before they are actually synthesized.

Elucidating the structure-activity relationships of the vasorelaxation and antioxidation properties of thionicotinic acid derivatives

Nicotinic acid, known as vitamin B₃, is an effective lipid lowering drug and intense cutaneous vasodilator. This study reports the effect of 2-(1-adamantylthio)nicotinic acid (6) and its amide 7 and nitrile analog 8 on phenylephrine-induced contraction of rat thoracic aorta as well as antioxidative activity. It was found that the tested thionicotinic acid analogs 6-8 exerted maximal vasorelaxation in a dose-dependent manner, but their effects were less than acetylcholine (ACh)-induced nitric oxide (NO) vasorelaxation. The vasorelaxations were reduced, apparently, in both N^G-nitro-L-arginine methyl ester (L-NAME) and indomethacin (INDO). Synergistic effects were observed in the presence of L-NAME plus INDO, leading to loss of vasorelaxation of both the ACh and the tested nicotinic acids. Complete loss of the vasorelaxation was noted under removal of endothelial cells. This infers that the vasorelaxations are mediated partially by endothelium-induced NO and prostacyclin. The thionicotinic acid analogs all exhibited antioxidant properties in both 2,2-diphenyl-1-picrylhydrazyl (DPPH) and superoxide dismutase (SOD) assays. Significantly, the thionicotinic acid 6 is the most potent vasorelaxant with ED₅₀ of 21.3 nM and is the most potent antioxidant (as discerned from DPPH assay). Molecular modeling was also used to provide mechanistic insights into the vasorelaxant and antioxidative activities. The findings reveal that the thionicotinic acid analogs are a novel class of vasorelaxant and antioxidant compounds which have potential to be further developed as promising therapeutics.

Synthesis and theoretical study of molecularly imprinted nanospheres for recognition of tocopherols

Molecular imprinting is a technology that facilitates the production of artificial receptors toward compounds of interest. The molecularly imprinted polymers act as artificial antibodies, artificial receptors, or artificial enzymes with the added benefit over their biological counterparts of being highly durable. In this study, we prepared molecularly imprinted polymers for the purpose of binding specifically to tocopherol (vitamin E) and its derivative, tocopherol acetate. Binding of the imprinted polymers to the template was found to be two times greater than that of the control, non-imprinted polymers, when using only 10 mg of polymers. Optimization of the rebinding solvent indicated that ethanol-water at a molar ratio of 6:4 (v/v) was the best solvent system as it enhanced the rebinding performance of the imprinted polymers toward both tocopherol and tocopherol acetate with a binding capacity of approximately 2 mg/g of polymer. Furthermore, imprinted nanospheres against tocopherol was successfully prepared by precipitation polymerization with ethanol-water at a molar ratio of 8:2 (v/v) as the optimal rebinding solvent. Computer simulation was also performed to provide mechanistic insights on the binding mode of template-monomer complexes. Such polymers show high potential for industrial and medical applications, particularly for selective separation of tocopherol and derivatives.

Modeling the LPS neutralization activity of anti-endotoxins

Bacterial lipopolysaccharides (LPS), also known as endotoxins, are major structural components of the outer membrane of Gram-negative bacteria that serve as a barrier and protective shield between them and their surrounding environment. LPS is considered to be a major virulence factor as it strongly stimulates the secretion of pro-inflammatory cytokines which mediate the host immune response and culminating in septic shock. Quantitative structure-activity relationship studies of the LPS neutralization activities of anti-endotoxins were performed using charge and quantum chemical descriptors. Artificial neural network implementing the back-propagation algorithm was selected for the multivariate analysis. The predicted activities from leave-one-out cross-validation were well correlated with the experimental values as observed from the correlation coefficient and root mean square error of 0.930 and 0.162, respectively. Similarly, the external testing set also yielded good predictivity with correlation coefficient and root mean square error of 0.983 and 0.130. The model holds great potential for the rational design of novel and robust compounds with enhanced neutralization activity.

Modeling the excitation wavelengths (lambda(ex)) of boronic acids

The quantitative structure-property relationship (QSPR) method was used to model the fluorescence excitation wavelengths (lambda(ex)) of 42 boronic acid-based fluorescent biosensors (30 in the training set and 12 in the test set). In this QSPR study, unsupervised forward selection (UFS), stepwise multiple linear regression (SMLR), partial least squares regression (PLS) and associative neural networks (ASNN) were employed to simulate linear and nonlinear models. All models were validated by a test set and Tropsha's validation model. The resulting ASNN nonlinear model demonstrates significant improvement on the predictive ability of the neural network compared to the SMLR and PLS linear models. The descriptors used in the models are discussed in detail. These QSPR models are useful tools for the prediction of fluorescence excitation wavelengths of arylboronic acids.

Rational design of analyte channels of the green fluorescent protein for biosensor applications

A novel solvent-exposed analyte channel, generated by F165G substitution, on the surface of green fluorescent protein (designated His(6)GFPuv/F165G) was successfully discovered by the aid of molecular modeling software (PyMOL) in conjunction with site-directed mutagenesis. Regarding the high predictive performance of PyMOL, two pore-containing mutants namely His(6)GFPuv/H148G and His(6)GFPuv/H148G/F165G were also revealed. The pore sizes of F165G, H148G, and the double mutant H148G/F165G were in the order of 4, 4.5 and 5.5 A, respectively. These mutants were subjected to further investigation on the effect of small analytes (e.g. metal ions and hydrogen peroxide) as elucidated by fluorescence quenching experiments. Results revealed that the F165G mutant exhibited the highest metal sensitivity at physiological pH. Meanwhile, the other 2 mutants lacking histidine at position 148 had lower sensitivity against Zn(2+) and Cu(2+) than those of the template protein (His(6)GFPuv). Hence, a significant role of this histidine residue in mediating metal transfer toward the GFP chromophore was proposed and evidently demonstrated by testing in acidic condition. Results revealed that at pH 6.5 the order of metal sensitivity was found to be inverted whereby the H148G/F165G became the most sensitive mutant. The dissociation constants (K(d)) to metal ions were in the order of 4.88 x 10(-6) M, 16.67 x 10(-6) M, 25 x 10(-6) M, and 33.33 x 10(-6) M for His(6)GFPuv/F165G, His(6)GFPuv, His(6)GFPuv/H148G/F165G and His(6)GFPuv/H148G, respectively. Sensitivity against hydrogen peroxide was in the order of H148G/F165G > H148G > F165G indicating the crucial role of pore diameters. However, it should be mentioned that H148G substitution caused a markedly decrease in pH- and thermo-stability. Taken together, our findings rendered the novel pore of GFP as formed by F165G substitution to be a high impact channel without adversely affecting the intrinsic fluorescent properties. This opens up a great potential of using F165G mutant in enhancing the sensitivity of GFP in future development of biosensors.