Search

Studying the deep learning-based molecular representation has great significance on predicting molecular property, promoted the development of drug screening and new drug discovery, and improving human well-being for avoiding illnesses. It is essential to learn the characterization of drug for various downstream tasks, such as molecular property prediction. In particular, the 3D structure features of molecules play an important role in biochemical function and activity prediction. The 3D characteristics of molecules largely determine the properties of the drug and the binding characteristics of the target. However, most current methods merely rely on 1D or 2D properties while ignoring the 3D topological structure, thereby degrading the performance of molecular inferring. In this paper, we propose 3DMol-Net to enhance the molecular representation, considering both the topology and rotation invariance (RI) of the 3D molecular structure. Specifically, we construct a molecular graph with soft relations related to the spatial arrangement of the 3D coordinates to learn 3D topology of arbitrary graph structure and employ an adaptive graph convolutional network to predict molecular properties and biochemical activities. Comparing with current graph-based methods, 3DMol-Net demonstrates superior performance in terms of both regression and classification tasks. Further verification of RI and visualization also show better robustness and representation capacity of our model.

The graph neural network (GNN) has become a promising method to predict molecular properties with end-to-end supervision, as it can learn molecular features directly from chemical graphs in a black-box manner. However, to achieve high prediction accuracy, it is essential to supervise a huge amount of property data, which is often accompanied by a high property experiment cost. Prior to the deep learning method, descriptor-based quantitative structure–property relationships (QSPR) studies have investigated physical and chemical knowledge to manually design descriptors for effectively predicting properties. In this study, we extend a message-passing neural network (MPNN) to include a novel MPNN architecture called the knowledge-embedded MPNN (KEMPNN) that can be supervised together with nonquantitative knowledge annotations by human experts on a chemical graph that contains information on the important substructure of a molecule and its effect on the target property (e.g., positive or negative effect). We evaluated the performance of the KEMPNN in a small training data setting using a physical chemistry dataset in MoleculeNet (ESOL, FreeSolv, Lipophilicity) and a polymer property (glass-transition temperature) dataset with virtual knowledge annotations. The results demonstrate that the KEMPNN with knowledge supervision can improve the prediction accuracy obtained from the MPNN. The results also demonstrate that the accuracy of the KEMPNN is better than or comparable to those of descriptor-based methods even in the case of small training data.

Aim: In our earlier research, we have synthesized series of substituted 1-(2, 5-dimethyl thiophene-3yl)-(4-substituted phenyl)-2-propene-1-one derivatives and evaluated them for their anti-bacterial and antifungal activity. In recent years, chalcone derivatives are proved for their varied pharmacological effects ranging from antimicrobial activity to anti-cancer effects. In this study, we have hypothesized the efficiency of our earlier synthesized anti-bacterial and antifungal chalcone derivatives for their potential inhibition of epidermal growth factor receptor protein (EGFR), through molecular docking studies. Methodology: Molecular docking simulation studies are performed using the Glide XP module of Schrodinger Suite and ligand binding energies are also calculated. Results: Molecular docking studies of the selected compounds against EGFR revealed docking scores ranging from -6.746 (compound 5) to -5.681 (compound 3) and also provided insight into binding conformations of the ligands in the EGFR protein environment. Additionally, molecular property and Absorption, Distribution, Metabolism, and Excretion (ADME) predictor analysis is also performed for the dataset ligands, which further provided the probable explanation for the binding potentials. Conclusion: Among all the tested dataset ligands, compound 5 has shown the highest dock score (-6.746) with better ADME profiles. Binding energies in the protein-ligand interactions explain how fit the ligand binds with the target protein. Molecular docking studies of these anti-bacterial, antifungal chalcone derivatives provided deeper insights in understanding the probable conformations of these tested ligands in the EGFR protein environment.

Chemical industry is always seeking opportunities to efficiently and economically convert raw materials to commodity chemicals and higher value-added chemical-based products. The life cycles of chemical products involve the procedures of conceptual product designs, experimental investigations, sustainable manufactures through appropriate chemical processes and waste disposals. During these periods, one of the most important keys is the molecular property prediction models associating molecular structures with product properties. In this paper, a framework combining quantum mechanics and quantitative structure-property relationship is established for fast molecular property predictions, such as activity coefficient, and so forth. The workflow of framework consists of three steps. In the first step, a database is created for collections of basic molecular information; in the second step, quantum mechanics-based calculations are performed to predict quantum mechanics-based/derived molecular properties (pseudo experimental data), which are stored in a database and further provided for the developments of quantitative structure-property relationship methods for fast predictions of properties in the third step. The whole framework has been carried out within a molecular property prediction toolbox. Two case studies highlighting different aspects of the toolbox involving the predictions of heats of reaction and solid-liquid phase equilibriums are presented.

Motivation Molecular property prediction is a hot topic in recent years. Existing graph-based models ignore the hierarchical structures of molecules. According to the knowledge of chemistry and pharmacy, the functional groups of molecules are closely related to its physio-chemical properties and binding affinities. So, it should be helpful to represent molecular graphs by fragments that contain functional groups for molecular property prediction. Results In this article, to boost the performance of molecule property prediction, we first propose a definition of molecule graph fragments that may be or contain functional groups, which are relevant to molecular properties, then develop a fragment-oriented multi-scale graph attention network for molecular property prediction, which is called FraGAT. Experiments on several widely used benchmarks are conducted to evaluate FraGAT. Experimental results show that FraGAT achieves state-of-the-art predictive performance in most cases. Furthermore, our case studies show that when the fragments used to represent the molecule graphs contain functional groups, the model can make better predictions. This conforms to our expectation and demonstrates the interpretability of the proposed model. Availability and implementation The code and data underlying this work are available in GitHub, at https://github.com/ZiqiaoZhang/FraGAT. Supplementary information Supplementary data are available at Bioinformatics online.

Flavonoids are a class of polyphenols that possess diverse properties. The structure-activity relationship of certain flavonoids and resveratrol with ribonuclease A (RNase A) has been investigated. The selected flavonoids have a similar skeleton and the positional preferences of the phenolic moieties toward inhibition of the catalytic activity of RNase A have been studied. The results obtained for RNase A inhibition by flavonoids suggest that the planarity of the molecules is necessary for effective inhibitory potency. Agarose gel electrophoresis and precipitation assay experiments along with kinetic studies reveal Ki values for the various flavonoids in the micromolar range. Minor secondary structural changes of RNase A were observed after interaction with the flavonoids. An insight into the specific amino acid involvement in the binding of the substrate using docking studies is also presented. The dipole moment of the flavonoids that depends on the orientation of the hydroxyl groups in the molecule bears direct correlation with the inhibitory potency against RNase A. The direct association of this molecular property with enzyme inhibition can be exploited for the design and development of inhibitors of proteins.

Chemists are trained to recognize aromaticity semi-intuitively, using pictures of resonance structures and Frost-Musulin diagrams, or simple electron-counting rules such as Huckel's 4n + 2/4n rule. To quantify aromaticity one can use various aromaticity indices, each of which is a number reflecting some experimentally measured or calculated molecular property, or some feature of the molecular wavefunction, which often has no visual interpretation or may not have direct chemical relevance. We show that computed isotropic magnetic shielding isosurfaces and contour plots provide a feature-rich picture of aromaticity and chemical bonding which is both quantitative and easy-to-visualize and interpret. These isosurfaces and contour plots make good chemical sense as at atomic positions they are pinned to the nuclear shieldings which are experimentally measurable through chemical shifts. As examples we discuss the archetypal aromatic and antiaromatic molecules of benzene and square cyclobutadiene, followed by modern visual interpretations of Clar's aromatic sextet theory, the aromaticity of corannulene and heteroaromaticity.

This paper presents a new method of generating property-energy consistent (PEC) basis sets that can be applied to any arbitrary molecular property. The PEC method generates a basis set that is optimized for the molecular property under interest, providing the least possible total molecular energy. The main algorithm of the PEC approach involves Monte Carlo simulations to generate random exponents in the predetermined range. In this work, the PEC method is introduced in the example of generation of new pecJ-n (n = 1, 2) basis sets suited for high-quality correlated calculations of indirect nuclear spin–spin coupling constants involving the most popular NMR-active nuclei: 1H, 13C, 15N, and 19F.

Machine learning algorithms have been found to be effective in predicting the properties of molecules and materials. Recently, a new strategy, Δ-machine learning, which uses low-level calculations as a baseline to predict properties of high-level methods, has been proposed to further reduce computational costs. It has been successfully applied to predictions of potential energy surfaces, bandgaps and chemical shieldings. Here we introduce a new descriptor, in which we used harmonic vibrational frequencies as the descriptor in predictions of molecular properties, namely the frequency descriptor (FD). In detail, we used harmonic vibrational frequencies of several semi-empirical methods (the PM6, PM7 and GFN2-xTB methods) as the descriptor in Δ-machine learning. The energies, enthalpies and HOMO–LUMO gaps of 6095 C7H10O2 isomers at high-level calculations were used as target properties to test the descriptor. We found that the FD generated by the GFN2-xTB method has excellent performance among several semiempirical methods. The chemical accuracy can be achieved with a small training set size according to the combination of single-point calculations at density functional theory levels. In addition, we further included infrared intensities to the FD, namely the FD-II by which the chemical accuracy of energies can be achieved with a small training set size (3%) that represents the smallest sample size in the current dataset (C7H10O2 isomers). We expect that the FD and FD-II can also be used to accelerate other property predictions.

Previously, we and other laboratories have reported an unusual and strong Raman optical activity (ROA) induced in solvents by chiral dyes. Various theories of the phenomenon appeared, but they were not capable of explaining fully the observed ROA band signs and intensities. In this work, an analysis based both on the light scattering theory and dedicated experiments provides a more complete understanding. For example, double‐cell magnetic circular dichroism and magnetic ROA experiments with copper‐porphyrin complex show that the induced chirality is observed without any contact of the solvents with the complex. The results thus indicate that a combination of electronic circular dichroism (ECD) with the polarized Raman scattering is responsible for the effect. The degree of circularity of solvent vibrational bands is a principal molecular property participating in the event. The insight and the possibility to predict the chirality transfer promise future applications in spectroscopy, chemical analysis and polarized imaging., In a series of experiments and by a theoretical analysis we decipher Raman optical activity induced in achiral solvents as an interference of two phenomena, circular dichroism and polarized Raman scattering. Applications of this strong effect can be potentially found in analytical chemistry, spectroscopy, and imaging of biologically relevant systems.

Graph neural networks (GNNs) have attracted significant attention from the chemical science community because molecules can be represented as a featured graph. In particular, graph convolutional network (GCN) and its variants have been widely used and have shown a state-of-the-art performance in analyzing molecules, such as molecular label classification, drug discovery, and molecular property prediction. However, in molecular analysis, existing GCNs have two fundamental limitations: (1) information of the molecular scale is distorted and (2) global structures in a molecule are ignored. These limitations can seriously degrade the performance in the machine learning-based molecular analysis because the scale and global structure information of a molecule occasionally have a significant effect on the molecular properties. To overcome the limitations of existing GCNs, we comprehensively analyzed the structure of GCNs and developed a costless solution for the limitations of GCNs. To demonstrate the effectiveness of our solution, extensive experiments were conducted on various benchmark datasets.

Molecular property prediction is important to drug design. With the development of artificial intelligence, deep learning methods are effective for extracting molecular features. In this paper, we propose a multichannel substructure-graph gated recurrent unit (GRU) architecture, which is a novel GRU-based neural network with attention mechanisms applied to molecular substructures to learn and predict properties. In the architecture, molecular features are extracted at the node level and molecule level for capturing fine-grained and coarse-grained information. In addition, three bidirectional GRUs are adopted to extract the features on three channels to generate the molecular representations. Different attention weights are assigned to the entities in the molecule to evaluate their contributions. Experiments are implemented to compare our model with benchmark models in molecular property prediction for both regression and classification tasks, and the results show that our model has strong robustness and generalizability.

Two-way liquid organic hydrogen carriers (LOHC) – organic molecules that store hydrogen as reversible chemical bonds – is an emerging concept for on-demand storage and transportation of hydrogen (and thereby energy). Given the large chemical universe, a plethora of potential LOHCs exist, however, the optimal candidate depends on satisfying a variety of constraints on physicochemical and thermochemical properties. Computational high-throughput screening of a subspace of this universe can, in principle, reveal several LOHC candidates which can then be experimentally verified; however, to achieve this, the hydrogen rich and lean forms of the LOHC pair have to be simultaneously identified based on a plausible connecting chemical pathway. Here, using a combination of data-driven molecular property models and a cheminformatics-based reaction network generation tool, viz. RING, we develop a novel computational workflow to identify promising LOHC pairs (i.e. hydrogen-rich and hydrogen-lean forms) and the dehydrogenation pathways connecting them. Starting from over 1 million small (containing less than 14 heavy atoms) molecules in the PubChem database as seed, we applied this framework as proof of concept to identify several LOHC pairs that have promising properties in terms of melting point, boiling point, dehydrogenation enthalpy, hydrogen storage capacity, and synthetic accessibility; we further analyze the thermochemistry of dehydrogenation pathways of the top five candidates. We finally show that this screening provides a rich dataset that can be harnessed via supervised learning algorithms to infer descriptive features that determine if a molecule is a good LOHC candidate. We posit that the proposed workflow can be used to scalably analyze a much larger molecular space and multiple classes of dehydrogenation chemistries to discover novel LOHC pairs.

The evaluation of absorption, distribution, metabolism, exclusion, and toxicity (ADMET) properties plays a key role in a variety of domains including industrial chemicals, agrochemicals, cosmetics, environmental science, food chemistry, and particularly drug development. Since molecules are often intrinsically described as molecular graphs, graph neural networks have recently been studied to improve the prediction of ADMET properties. Among many graph neural networks published in recent years, Graph Isomorphism Network (GIN) is a relatively recent and very promising one. In this paper, we propose an enhanced GIN, called MolGIN, via exploiting the bond features and differences influence of the atom neighbors to end-to-end predict ADMET properties. Based on GIN, MolGIN concatenates the bond feature together with node feature in the feature aggregator and applies a gate unit to adjust the atomic neighborhood weights to map the differences in the interaction strength between the central atom and its neighbors, such that more meaningful structural patterns of molecules can be explored toward better molecular modeling. Extensive experiments were conducted on seven public datasets to evaluate MolGIN against four baseline models with benchmark metrics. Experimental results of MolGIN were also compared with state-of-the-art results published in the last three years on each dataset. Experimental results in terms of RMSE and AUC show that MolGIN significantly boosts the prediction performance of GIN and markedly outperforms the baseline models, and achieves comparable or superior performance to state-of-the-art results.

Crossing cellular membranes is a versatile molecular property that allows for a wide variety of peptides with cell penetrating capabilities. This broadness complicates identification of candidates suited best for a specific application. To facilitate the screening of this enormous molecular space in a supervised manner we here present a method to "breed" the desired molecules by applying the rules of Darwinian evolution. With this mate-and-check protocol, which combines an in silico evolution step with an in vitro performance test, cell penetrating peptides that are optimized for a specific task can be achieved in a few rounds of breeding. The procedure is simple and straightforward on the synthetic site but requires robust, highly reproducible and close-to-reality biological assays to yield realistic functional output. With this technology even top-performing peptides can be further improved and functionally adjusted.

Predicting molecular properties remains a challenging task with numerous potential applications, notably in drug discovery. Recently, the development of deep learning, combined with rising amounts of data, has provided powerful tools to build predictive models. Since molecules can be encoded as graphs, Graph Neural Networks (GNNs) have emerged as a popular choice of architecture to tackle this task. Training GNNs to predict molecular properties however faces the challenge of collecting annotated data which is a costly and time consuming process. On the other hand, it is easy to access large databases of molecules without annotations. In this setting, self-supervised learning can efficiently leverage large amounts of non-annotated data to compensate for the lack of annotated ones. In this work, we introduce a self-supervised framework for GNNs tailored specifically for molecular property prediction. Our framework uses multiple pretext tasks focusing on different scales of molecules (atoms, fragments and entire molecules). We evaluate our method on a representative set of GNN architectures and datasets and also consider the impact of the choice of input features. Our results show that our framework can successfully improve performance compared to training from scratch, especially in low data regimes. The improvement varies depending on the dataset, model architecture and, importantly, on the choice of input feature representation.

Accounting for solvent effects in theoretical predictions of spectroscopic properties may be of significant importance since a solvent—and on a more general basis any environment—may influence key spectroscopic parameters in nontrivial ways—especially in relation to calculation of nonlinear optical properties. At the simplest level, solvent effects may be included into quantum chemistry calculations by use of a dielectric continuum approach; however, such a description may fail in addressing correctly environment anisotropies as well as specific interactions such as hydrogen bonding. On the other hand, discrete embedding methods allow for a more correct description of an environment since such methods keep the atomistic nature of the environment intact. In this paper, two-photon circular dichroism (TPCD) spectra will be calculated for two different biaryl derivatives introducing solvent effects by two different discrete embedding schemes, i.e., polarizable embedding based on induced dipoles (PE) or the fluctuating charge (FQ) model. While we find inclusion of solvent effects on this molecular property to be important, we conclude at the same time that the influence of the solvent on the TPCD rotatory strength is accounted for in an overall rather equivalent manner by either embedding methods.

The accurate prediction of molecular properties, such as lipophilicity and aqueous solubility, are of great importance and pose challenges in several stages of the drug discovery pipeline. Machine learning methods, such as graph-based neural networks (GNNs), have shown exceptionally good performance in predicting these properties. In this work, we introduce a novel GNN architecture, called directed edge graph isomorphism network (D-GIN). It is composed of two distinct sub-architectures (D-MPNN, GIN) and achieves an improvement in accuracy over its sub-architectures employing various learning, and featurization strategies. We argue that combining models with different key aspects help make graph neural networks deeper and simultaneously increase their predictive power. Furthermore, we address current limitations in assessment of deep-learning models, namely, comparison of single training run performance metrics, and offer a more robust solution.

It is shown that the electric dipole- and electric quadrupole–anapole polarizabilities, denoted respectively by fαβ′ and gα,βγ′, and the anapole magnetizability aαβ, are intrinsic properties of the electron cloud of molecules responding to optical fields. fαβ′ is a nonvanishing property of chiral and achiral compounds, whereas aαβ is suitable for enantiomer discrimination of chiral species. They can conveniently be evaluated by numerical integration, employing a formulation complementary to that provided by perturbation theory and relying on the preliminary computation of electronic current density tensors all over the molecular domain. The origin dependence of the dynamic anapolar response is rationalized via related computational techniques employing numerical integration, as well as definitions of molecular property tensors, for example, electric dipole and electric quadrupole polarizabilties and magnetizability. A preliminary application of the theory is reported for the Ra enantiomer of the hydrogen peroxide molecule, evaluating tensor components of electric dipole-anapole polarizability and anapole magnetizability as functions of the dihedral angle ϕ≡∠ H-O-O-H in the range 0∘≤ϕ≤180∘.

Effective molecular representation learning is of great importance to facilitate molecular property prediction, which is a fundamental task for the drug and material industry. Recent advances in graph neural networks (GNNs) have shown great promise in applying GNNs for molecular representation learning. Moreover, a few recent studies have also demonstrated successful applications of self-supervised learning methods to pre-train the GNNs to overcome the problem of insufficient labeled molecules. However, existing GNNs and pre-training strategies usually treat molecules as topological graph data without fully utilizing the molecular geometry information. Whereas, the three-dimensional (3D) spatial structure of a molecule, a.k.a molecular geometry, is one of the most critical factors for determining molecular physical, chemical, and biological properties. To this end, we propose a novel Geometry Enhanced Molecular representation learning method (GEM) for Chemical Representation Learning (ChemRL). At first, we design a geometry-based GNN architecture that simultaneously models atoms, bonds, and bond angles in a molecule. To be specific, we devised double graphs for a molecule: The first one encodes the atom-bond relations; The second one encodes bond-angle relations. Moreover, on top of the devised GNN architecture, we propose several novel geometry-level self-supervised learning strategies to learn spatial knowledge by utilizing the local and global molecular 3D structures. We compare ChemRL-GEM with various state-of-the-art (SOTA) baselines on different molecular benchmarks and exhibit that ChemRL-GEM can significantly outperform all baselines in both regression and classification tasks. For example, the experimental results show an overall improvement of 8.8% on average compared to SOTA baselines on the regression tasks, demonstrating the superiority of the proposed method., Comment: Nature Machine Intelligence, 2022

Graph neural networks (GNNs) have been proposed for a wide range of graph-related learning tasks. In particular, in recent years, an increasing number of GNN systems were applied to predict molecular properties. However, a direct impediment is to select appropriate hyperparameters to achieve satisfactory performance with lower computational cost. Meanwhile, many molecular datasets are far smaller than many other datasets in typical deep learning applications. Most hyperparameter optimization (HPO) methods have not been explored in terms of their efficiencies on such small datasets in the molecular domain. In this paper, we conducted a theoretical analysis of common and specific features for two state-of-the-art and popular algorithms for HPO: TPE and CMA-ES, and we compared them with random search (RS), which is used as a baseline. Experimental studies are carried out on several benchmarks in MoleculeNet, from different perspectives to investigate the impact of RS, TPE, and CMA-ES on HPO of GNNs for molecular property prediction. In our experiments, we concluded that RS, TPE, and CMA-ES have their individual advantages in tackling different specific molecular problems. Finally, we believe our work will motivate further research on GNN as applied to molecular machine learning problems in chemistry and materials sciences.

The key to generating the best deep learning model for predicting molecular property is to test and apply various optimization methods. While individual optimization methods from different past works outside the pharmaceutical domain each succeeded in improving the model performance, better improvement may be achieved when specific combinations of these methods and practices are applied. In this work, three high-performance optimization methods in the literature that have been shown to dramatically improve model performance from other fields are used and discussed, eventually resulting in a general procedure for generating optimized CNN models on different properties of molecules. The three techniques are the dynamic batch size strategy for different enumeration ratios of the SMILES representation of compounds, Bayesian optimization for selecting the hyperparameters of a model and feature learning using chemical features obtained by a feedforward neural network, which are concatenated with the learned molecular feature vector. A total of seven different molecular properties (water solubility, lipophilicity, hydration energy, electronic properties, blood–brain barrier permeability and inhibition) are used. We demonstrate how each of the three techniques can affect the model and how the best model can generally benefit from using Bayesian optimization combined with dynamic batch size tuning.