Search

The human soluble epoxide hydrolase (sEH) is a bifunctional enzyme that modulates the levels of regulatory epoxy lipids. The hydrolase activity is carried out by a catalytic triad located at the center of a wide L-shaped binding site, which contains two hydrophobic subpockets at both sides. On the basis of these structural features, it can be assumed that desolvation is a major factor in determining the maximal achievable affinity that can be attained for this pocket. Accordingly, hydrophobic descriptors may be better suited to the search of novel hits targeting this enzyme. This study examines the suitability of quantum mechanically derived hydrophobic descriptors in the discovery of novel sEH inhibitors. To this end, three-dimensional quantitative structure-activity relationship (3D-QSAR) pharmacophores were generated by combining electrostatic and steric or alternatively hydrophobic and hydrogen-bond parameters in conjunction with a tailored list of 76 known sEH inhibitors. The pharmacophore models were then validated by using two external sets chosen (i) to rank the potency of four distinct series of compounds and (ii) to discriminate actives from decoys, using in both cases datasets taken from the literature. Finally, a prospective study was performed including a virtual screening of two chemical libraries to identify new potential hits, which were subsequently experimentally tested for their inhibitory activity on human, rat, and mouse sEH. The use of hydrophobic-based descriptors led to the identification of six compounds as inhibitors of the human enzyme with IC50 < 20 nM, including two with IC50 values of 0.4 and 0.7 nM. The results support the use of hydrophobic descriptors as a valuable tool in the search of novel scaffolds that encode a proper hydrophilic/hydrophobic distribution complementary to the target's binding site.

Whereas 400 million distinct compounds are now purchasable within the span of a few weeks, the biological activities of most are unknown. To facilitate access to new chemistry for biology, we have combined the Similarity Ensemble Approach (SEA) with the maximum Tanimoto similarity to the nearest bioactive to predict activity for every commercially available molecule in ZINC. This method, which we label SEA+TC, outperforms both SEA and a naïve-Bayesian classifier via predictive performance on a 5-fold cross-validation of ChEMBL's bioactivity data set (version 21). Using this method, predictions for over 40% of compounds (>160 million) have either high significance (pSEA ≥ 40), high similarity (ECFP4MaxTc ≥ 0.4), or both, for one or more of 1382 targets well described by ligands in the literature. Using a further 1347 less-well-described targets, we predict activities for an additional 11 million compounds. To gauge whether these predictions are sensible, we investigate 75 predictions for 50 drugs lacking a binding affinity annotation in ChEMBL. The 535 million predictions for over 171 million compounds at 2629 targets are linked to purchasing information and evidence to support each prediction and are freely available via https://zinc15.docking.org and https://files.docking.org .

TBK1, or TANK-binding kinase 1, is an enzyme that functions as a serine/threonine protein kinase. It plays a crucial role in various cellular processes, including the innate immune response to viruses, cell proliferation, apoptosis, autophagy, and antitumor immunity. Dysregulation of TBK1 activity can lead to autoimmune diseases, neurodegenerative disorders, and cancer. Due to its central role in these critical pathways, TBK1 is a significant focus of research for therapeutic drug development. In this paper, we explore data from the CAS Content Collection regarding TBK1 and its implication in a large assortment of diseases and disorders. With the demand for developing efficient TBK1 inhibitors being outlined, we focus on utilizing a machine learning approach for developing predictive models for TBK1 inhibition, derived from the fragment-functional analysis descriptors. Using the extensive CAS Content Collection, we assembled a training set of TBK1 inhibitors with experimentally measured IC50 values. We explored several machine learning techniques combined with various molecular descriptors to derive and select the best TBK1 inhibitor QSAR models. Certain significant structural alerts that potentially contribute to inhibition of TBK1 are outlined and discussed. The merit of the article stems from identifying the most adequate TBK1 QSAR models and subsequent successful development of advanced positive training data to facilitate and enhance drug discovery for an important therapeutic target such as TBK1 inhibitors, based on an extensive, wide-ranging set of scientific information provided by the CAS Content Collection.

QSRR is a valuable technique for the retention time predictions of small molecules. This aims to bridge the gap between molecular structure and chromatographic behavior, offering invaluable insights for analytical chemistry. Given the challenge of simultaneous target prediction with variable experimental conditions and the scarcity of comprehensive data sets for such predictive modelings in chromatography, this study introduces a transfer learning-based multitarget QSRR approach to enhance retention time prediction. Through a comparative study of four models, both with and without the transfer learning approach, the performance of both single and multitarget QSRR was evaluated based on Mean Squared Error (MSE) and R 2 metrics. Individual models were also tested for their performance against benchmark studies in this field. The findings suggest that transfer learning based multitarget models exhibit potential for enhanced accuracy in predicting retention times of small molecules, presenting a promising avenue for QSRR modeling. These models will be highly beneficial for optimizing experimental conditions in method development by better retention time predictions in Reversed-Phase Liquid Chromatography (RPLC). The reliable and effective predictive capabilities of these models make them valuable tools for pharmaceutical research and development endeavors.

Nucleophilic index ( N Nu ) as a significant parameter plays a crucial role in screening of amine catalysts. Indeed, the quantity and variety of amines are extensive. However, only limited amines exhibit an N values by using Machine Learning (ML) and high-throughput Density Functional Theory (DFT) calculations. Our approach commenced by training ML models and the exploration of Molecular Fingerprint methods as well as the development of quantitative structure-activity relationship (QSAR) models for the well-known amines based on Nu value exceeding 4.0 eV, rendering them potential nucleophiles in chemical reactions. To address this issue, we proposed a computational method to quickly identify amines with high N Nu values by using Machine Learning (ML) and high-throughput Density Functional Theory (DFT) calculations. Our approach commenced by training ML models and the exploration of Molecular Fingerprint methods as well as the development of quantitative structure-activity relationship (QSAR) models for the well-known amines based on N Nu values derived from DFT calculations. Utilizing explainable Shapley Additive Explanation plots, we were able to determine the five critical substructures that significantly impact the N Nu values of amine. The aforementioned conclusion can be applied to produce and cultivate 4920 novel hypothetical amines with high N Nu values. The QSAR models were employed to predict the N Nu values of 259 well-known and 4920 hypothetical amines, resulting in the identification of five novel hypothetical amines with exceptional N Nu values (>4.55 eV). The enhanced N Nu values of these novel amines were validated by DFT calculations. One novel hypothetical amine, H1, exhibits an unprecedentedly high N Nu value of 5.36 eV, surpassing the maximum value (5.35 eV) observed in well-established amines. Our research strategy efficiently accelerates the discovery of the high nucleophilicity of amines using ML predictions, as well as the DFT calculations.

Computational approaches are widely applied in drug discovery to explore properties related to bioactivity, physiochemistry, and toxicology. Over at least the last 20 years, the exploitation of machine learning on molecular data sets has been used to understand the structure-activity relationships that exist between biomolecules and druggable targets. More recently, these methods have also seen application for phenotypic screening data for neglected diseases such as tuberculosis and malaria. Herein, we apply machine learning to build quantum Quantitative Structure Activity Relationship models from antimalarial data sets. There is a continual need for new antimalarials to address drug resistance, and the readily available in vitro data sets could be utilized with newer machine learning approaches as these develop. Furthermore, quantum machine learning is a relatively new method that uses a quantum computer to perform the calculations. First, we present a classical-quantum hybrid computational approach by building a Latent Bernoulli Autoencoder machine learning model for compressing bit-vector descriptors to a size that can be adapted to quantum computers for classification tasks with limited loss of embedded information. Second, we apply our method for feature map compression to quantum classification algorithms, including a completely novel machine learning algorithm with no analogy in classical computers: the Quantum Fourier Transform Classifier. We apply both these approaches to build quantum machine learning models for small-molecule antimalarials with quantum simulation software and then benchmark these quantum models against classical machine learning approaches. While there are many challenges currently facing the development of reliable quantum computers, our results demonstrate that there is potential for the use of this technology in the field of drug discovery.

In drug candidate design, clearance is one of the most crucial pharmacokinetic parameters to consider. Recent advancements in machine learning techniques coupled with the growing accumulation of drug data have paved the way for the construction of computational models to predict drug clearance. However, concerns persist regarding the reliability of data collected from public sources, and a majority of current in silico quantitative structure-property relationship models tend to neglect the influence of molecular chirality. In this study, we meticulously examined human liver microsome (HLM) data from public databases and constructed two distinct data sets with varying HLM data quantity and quality. Two baseline models (RF and DNN) and three chirality-focused GNNs (DMPNN, TetraDMPNN, and ChIRo) were proposed, and their performance on HLM data was evaluated and compared with each other. The TetraDMPNN model, which leverages chirality from 2D structure, exhibited the best performance with a test R 2 of 0.639 and a test root-mean-squared error of 0.429. The applicability domain of the model was also defined by using a molecular similarity-based method. Our research indicates that graph neural networks capable of capturing molecular chirality have significant potential for practical application and can deliver superior performance.

Artificial intelligence (AI)-aided drug design has demonstrated unprecedented effects on modern drug discovery, but there is still an urgent need for user-friendly interfaces that bridge the gap between these sophisticated tools and scientists, particularly those who are less computer savvy. Herein, we present DrugFlow, an AI-driven one-stop platform that offers a clean, convenient, and cloud-based interface to streamline early drug discovery workflows. By seamlessly integrating a range of innovative AI algorithms, covering molecular docking, quantitative structure-activity relationship modeling, molecular generation, ADMET (absorption, distribution, metabolism, excretion and toxicity) prediction, and virtual screening, DrugFlow can offer effective AI solutions for almost all crucial stages in early drug discovery, including hit identification and hit/lead optimization. We hope that the platform can provide sufficiently valuable guidance to aid real-word drug design and discovery. The platform is available at https://drugflow.com.

Machine-learning (ML) and deep-learning (DL) approaches to predict the molecular properties of small molecules are increasingly deployed within the design-make-test-analyze (DMTA) drug design cycle to predict molecular properties of interest. Despite this uptake, there are only a few automated packages to aid their development and deployment that also support uncertainty estimation, model explainability, and other key aspects of model usage. This represents a key unmet need within the field, and the large number of molecular representations and algorithms (and associated parameters) means it is nontrivial to robustly optimize, evaluate, reproduce, and deploy models. Here, we present QSARtuna, a molecule property prediction modeling pipeline, written in Python and utilizing the Optuna, Scikit-learn, RDKit, and ChemProp packages, which enables the efficient and automated comparison between molecular representations and machine learning models. The platform was developed by considering the increasingly important aspect of model uncertainty quantification and explainability by design. We provide details for our framework and provide illustrative examples to demonstrate the capability of the software when applied to simple molecular property, reaction/reactivity prediction, and DNA encoded library enrichment classification. We hope that the release of QSARtuna will further spur innovation in automatic ML modeling and provide a platform for education of best practices in molecular property modeling. The code for the QSARtuna framework is made freely available via GitHub.

The acid dissociation constant is an important molecular property, and it can be successfully predicted by Quantitative Structure-Property Relationship (QSPR) models, even for in silico designed molecules. We analyzed how the methodology of in silico 3D structure preparation influences the quality of QSPR models. Specifically, we evaluated and compared QSPR models based on six different 3D structure sources (DTP NCI, Pubchem, Balloon, Frog2, OpenBabel, and RDKit) combined with four different types of optimization. These analyses were performed for three classes of molecules (phenols, carboxylic acids, anilines), and the QSPR model descriptors were quantum mechanical (QM) and empirical partial atomic charges. Specifically, we developed 516 QSPR models and afterward systematically analyzed the influence of the 3D structure source and other factors on their quality. Our results confirmed that QSPR models based on partial atomic charges are able to predict pKa with high accuracy. We also confirmed that ab initio and semiempirical QM charges provide very accurate QSPR models and using empirical charges based on electronegativity equalization is also acceptable, as well as advantageous, because their calculation is very fast. On the other hand, Gasteiger-Marsili empirical charges are not applicable for pKa prediction. We later found that QSPR models for some classes of molecules (carboxylic acids) are less accurate. In this context, we compared the influence of different 3D structure sources. We found that an appropriate selection of 3D structure source and optimization method is essential for the successful QSPR modeling of pKa. Specifically, the 3D structures from the DTP NCI and Pubchem databases performed the best, as they provided very accurate QSPR models for all the tested molecular classes and charge calculation approaches, and they do not require optimization. Also, Frog2 performed very well. Other 3D structure sources can also be used but are not so robust, and an unfortunate combination of molecular class and charge calculation approach can produce weak QSPR models. Additionally, these 3D structures generally need optimization in order to produce good quality QSPR models.

In drug discovery, it is generally accepted that neighboring molecules in a given descriptor's space display similar activities. However, even in regions that provide strong predictability, structurally similar molecules can occasionally display large differences in potency. In QSAR jargon, these discontinuities in the activity landscape are known as "activity cliffs". In this study, we assessed the reliability of ligand docking and virtual ligand screening schemes in predicting activity cliffs. We performed our calculations on a diverse, independently collected database of cliff-forming cocrystals. Starting from ideal situations, which allowed us to establish our baseline, we progressively moved toward simulating more realistic scenarios. Ensemble- and template-docking achieved a significant level of accuracy, suggesting that, despite the well-known limitations of empirical scoring schemes, activity cliffs can be accurately predicted by advanced structure-based methods.

The intricate nature of the blood-brain barrier (BBB) poses a significant challenge in predicting drug permeability, which is crucial for assessing central nervous system (CNS) drug efficacy and safety. This research utilizes an innovative approach, the classification read-across structure-activity relationship (c-RASAR) framework, that leverages machine learning (ML) to enhance the accuracy of BBB permeability predictions. The c-RASAR framework seamlessly integrates principles from both read-across and QSAR methodologies, underscoring the need to consider similarity-related aspects during the development of the c-RASAR model. It is crucial to note that the primary goal of this research is not to introduce yet another model for predicting BBB permeability but rather to showcase the refinement in predicting the BBB permeability of organic compounds through the introduction of a c-RASAR approach. This groundbreaking methodology aims to elevate the accuracy of assessing neuropharmacological implications and streamline the process of drug development. In this study, an ML-based c-RASAR linear discriminant analysis (LDA) model was developed using a dataset of 7807 compounds, encompassing both BBB-permeable and -nonpermeable substances sourced from the B3DB database (freely accessible from https://github.com/theochem/B3DB), for predicting BBB permeability in lead discovery for CNS drugs. The model's predictive capability was then validated using three external sets: one containing 276,518 natural products (NPs) from the LOTUS database (accessible from https://lotus.naturalproducts.net/download) for data gap filling, another comprising 13,002 drug-like/drug compounds from the DrugBank database (available from https://go.drugbank.com/), and a third set of 56 FDA-approved drugs to assess the model's reliability. Further diversifying the predictive arsenal, various other ML-based c-RASAR models were also developed for comparison purposes. The proposed c-RASAR framework emerged as a powerful tool for predicting BBB permeability. This research not only advances the understanding of molecular determinants influencing CNS drug permeability but also provides a versatile computational platform for the rapid assessment of diverse compounds, facilitating informed decision-making in drug development and design.

Generating the three-dimensional (3D) structure of small molecules is crucial in both structure- and ligand-based drug design. Structure-based drug design needs bioactive conformations of compounds for lead identification and optimization. Ligand-based drug design techniques, such as 3D shape similarity search, 3D pharmacophore model, 3D-QSAR, etc., all require high-quality small-molecule ligand conformations to obtain reliable results. Although predicting a small molecular bioactive conformer requires information from the receptor, a crystal structure of the molecule is a proper approximation to its bioactive conformer in a specific receptor because the binding pose of a small molecule in its receptor's binding pockets should be energetically close to the crystal structures. This study presents a de novo small molecular structure predictor (dMXP) with graph attention networks based on crystal data derived from the Cambridge Structural Database (CSD) combined with molecular electrostatic information calculated by density-functional theory (DFT). Two featuring strategies (topological and atomic partial change features) were employed to explore the relation between these features and the 3D crystal structure of a small molecule. These features were then assembled to construct the holistic 3D crystal structure of a molecule. Molecular graphs were encoded using a graph attention mechanism to deal with the issues of the inconsistencies of local substructures contributing to the entire molecular structure. The root-mean-square deviation (RMSDs) of approximately 80% dMXP predicted structures and the native binding poses within receptors are less than 2.0 Å.

Artificial intelligence is expected to help identify excellent candidates in drug discovery. However, we face a lack of data, as it is time-consuming and expensive to acquire raw data perfectly for many compounds. Hence, we tried to develop a novel quantitative structure-activity relationship (QSAR) method to predict a parameter more precisely from an incomplete data set via optimizing data handling by making use of predicted explanatory variables. As a case study we focused on the tissue-to-plasma partition coefficient (Kp), which is an important parameter for understanding drug distribution in tissues and building the physiologically based pharmacokinetic model and is a representative of small and sparse data sets. In this study, we predicted the Kp values of 119 compounds in nine tissues (adipose, brain, gut, heart, kidney, liver, lung, muscle, and skin), although some of these were not available. To fill the missing values in Kp for each tissue, first we predicted those Kp values by the nonmissing data set using a random forest (RF) model with in vitro parameters (log P, fu, Drug Class, and fi) like a classical prediction by a QSAR model. Next, to predict the tissue-specific Kp values in a test data set, we constructed a second RF model with not only in vitro parameters but also the Kp values of other tissues (i.e., other than target tissues) predicted by the first RF model as explanatory variables. Furthermore, we tested all possible combinations of explanatory variables and selected the model with the highest predictability from the test data set as the final model. The evaluation of Kp prediction accuracy based on the root-mean-square error and R 2 value revealed that the proposed models outperformed other machine learning methods such as the conventional RF and message-passing neural networks. Significant improvements were observed in the Kp values of adipose tissue, brain, kidney, liver, and skin. These improvements indicated that the Kp information on other tissues can be used to predict the same for a specific tissue. Additionally, we found a novel relationship between each tissue by evaluating all combinations of explanatory variables. In conclusion, we developed a novel RF model to predict Kp values. We hope that this method will be applied to various problems in the field of experimental biology which often contains missing values in the near future.

Half-life is a significant pharmacokinetic parameter included in the excretion phase of absorption, distribution, metabolism, and excretion. It is one of the key factors for the successful marketing of drug candidates. Therefore, predicting half-life is of great significance in drug design. In this study, we emplo yed eXtreme Gradient Boosting (XGboost), randomForest (RF), gradient boosting machine (GBM), and supporting vector machine (SVM) to build quantitative structure-activity relationship (QSAR) models on 3512 compounds and evaluated model performance by using root-mean-square error (RMSE), R 2 , and mean absolute error (MAE) metrics and interpreted features by SHapley Additive exPlanation (SHAP). Furthermore, we developed consensus models through integrating four individual models and validated their performance using a Y-randomization test and applicability domain analysis. Finally, matched molecular pair analysis was used to extract the transformation rules. Our results revealed that XGboost outperformed other individual models (RMSE = 0.176, R 2 = 0.845, MAE = 0.141). The consensus model integrating all four models continued to enhance prediction performance (RMSE = 0.172, R 2 = 0.856, MAE = 0.138). We evaluated the reliability, robustness, and generalization ability via Y-randomization test and applicability domain analysis. Meanwhile, we utilized SHAP to interpret features and employed matched molecular pair analysis to extract chemical transformation rules that provide suggestions for optimizing drug structure. In conclusion, we believe that the consensus model developed in this study serve as a reliable tool to evaluate half-life in drug discovery, and the chemical transformation rules concluded in this study could provide valuable suggestions in drug discovery.

Self-supervised pretrained models are gaining increasingly more popularity in AI-aided drug discovery, leading to more and more pretrained models with the promise that they can extract better feature representations for molecules. Yet, the quality of learned representations has not been fully explored. In this work, inspired by the two phenomena of Activity Cliffs (ACs) and Scaffold Hopping (SH) in traditional Quantitative Structure-Activity Relationship analysis, we propose a method named Re presentation- P roperty R elationship A nalysis (RePRA) to evaluate the quality of the representations extracted by the pretrained model and visualize the relationship between the representations and properties. The concepts of ACs and SH are generalized from the structure-activity context to the representation-property context, and the underlying principles of RePRA are analyzed theoretically. Two scores are designed to measure the generalized ACs and SH detected by RePRA, and therefore, the quality of representations can be evaluated. In experiments, representations of molecules from 10 target tasks generated by 7 pretrained models are analyzed. The results indicate that the state-of-the-art pretrained models can overcome some shortcomings of canonical Extended-Connectivity FingerPrints, while the correlation between the basis of the representation space and specific molecular substructures are not explicit. Thus, some representations could be even worse than the canonical fingerprints. Our method enables researchers to evaluate the quality of molecular representations generated by their proposed self-supervised pretrained models. And our findings can guide the community to develop better pretraining techniques to regularize the occurrence of ACs and SH.

Imputation machine learning (ML) surpasses traditional approaches in modeling toxicity data. The method was tested on an open-source data set comprising approximately 2500 ingredients with limited in vitro and in vivo data obtained from the OECD QSAR Toolbox. By leveraging the relationships between different toxicological end points, imputation extracts more valuable information from each data point compared to well-established single end point methods, such as ML-based Quantitative Structure Activity Relationship (QSAR) approaches, providing a final improvement of up to around 0.2 in the coefficient of determination. A significant aspect of this methodology is its resilience to the inclusion of extraneous chemical or experimental data. While additional data typically introduces a considerable level of noise and can hinder performance of single end point QSAR modeling, imputation models remain unaffected. This implies a reduction in the need for laborious manual preprocessing tasks such as feature selection, thereby making data preparation for ML analysis more efficient. This successful test, conducted on open-source data, validates the efficacy of imputation approaches in toxicity data analysis. This work opens the way for applying similar methods to other types of sparse toxicological data matrices, and so we discuss the development of regulatory authority guidelines to accept imputation models, a key aspect for the wider adoption of these methods.

AlvaBuilder is a software tool for de novo molecular design and can be used to generate novel molecules having desirable characteristics. Such characteristics can be defined using a simple step by step graphical interface, and they can be based on molecular descriptors, on predictions of QSAR/QSPR models, and on matching molecular fragments or used to design compounds similar to a given one. The molecules generated are always syntactically valid since they are composed by combining fragments of molecules taken from a training data set chosen by the user. In this paper, we demonstrate how the software can be used to design new compounds for a defined case study. AlvaBuilder is available at https://www.alvascience.com/alvabuilder/.

Optimizing the activities and properties of lead compounds is an essential step in the drug discovery process. Despite recent advances in machine learning-aided drug discovery, most of the existing methods focus on making predictions for the desired objectives directly while ignoring the explanations for predictions. Although several techniques can provide interpretations for machine learning-based methods such as feature attribution, there are still gaps between these interpretations and the principles commonly adopted by medicinal chemists when designing and optimizing molecules. Here, we propose an interpretation framework, named MolSHAP, for quantitative structure-activity relationship analysis by estimating the contributions of R-groups. Instead of attributing the activities to individual input features, MolSHAP regards the R-group fragments as the basic units of interpretation, which is in accordance with the fragment-based modifications in molecule optimization. MolSHAP is a model-agnostic method that can interpret activity regression models with arbitrary input formats and model architectures. Based on the evaluations of numerous representative activity regression models on a specially designed R-group ranking task, MolSHAP achieved significantly better interpretation power compared with other methods. In addition, we developed a compound optimization algorithm based on MolSHAP and illustrated the reliability of the optimized compounds using an independent case study. These results demonstrated that MolSHAP can provide a useful tool for accurately interpreting the quantitative structure-activity relationships and rationally optimizing the compound activities in drug discovery.

Drug development involves the thorough assessment of the candidate's safety and efficacy. In silico toxicology (IST) methods can contribute to the assessment, complementing in vitro and in vivo experimental methods, since they have many advantages in terms of cost and time. Also, they are less demanding concerning the requirements of product and experimental animals. One of these methods, Quantitative Structure-Activity Relationships (QSAR), has been proven successful in predicting simple toxicity end points but has more difficulties in predicting end points involving more complex phenomena. We hypothesize that QSAR models can produce better predictions of these end points by combining multiple QSAR models describing simpler biological phenomena and incorporating pharmacokinetic (PK) information, using quantitative in vitro to in vivo extrapolation (QIVIVE) models. In this study, we applied our methodology to the prediction of cholestasis and compared it with direct QSAR models. Our results show a clear increase in sensitivity. The predictive quality of the models was further assessed to mimic realistic conditions where the query compounds show low similarity with the training series. Again, our methodology shows clear advantages over direct QSAR models in these situations. We conclude that the proposed methodology could improve existing methodologies and could be suitable for being applied to other toxicity end points.

In molecular optimization, one popular way is R-group decoration on molecular scaffolds, and many efforts have been made to generate R-groups based on deep generative models. However, these methods mostly use information on known binding ligands, without fully utilizing target structure information. In this study, we proposed a new method, DiffDec, to involve 3D pocket constraints by a modified diffusion technique for optimizing molecules through molecular scaffold decoration. For end-to-end generation of R-groups with different sizes, we designed a novel fake atom mechanism. DiffDec was shown to be able to generate structure-aware R-groups with realistic geometric substructures by the analysis of bond angles and dihedral angles and simultaneously generate multiple R-groups for one scaffold on different growth anchors. The growth anchors could be provided by users or automatically determined by our model. DiffDec achieved R-group recovery rates of 69.67% and 45.34% in the single and multiple R-group decoration tasks, respectively, and these values were significantly higher than competing methods (37.33% and 26.85%). According to the molecular docking study, our decorated molecules obtained a better average binding affinity than baseline methods. The docking pose analysis revealed that DiffDec could decorate scaffolds with R-groups that exhibited improved binding affinities and more favorable interactions with the pocket. These results demonstrated the potential and applicability of DiffDec in real-world scaffold decoration for molecular optimization.

Machine learning-based predictive models allow rapid and reliable prediction of material properties and facilitate innovative materials design. Base oils used in the formulation of lubricant products are complex hydrocarbons of varying sizes and structure. This study developed Gaussian process regression-based models to accurately predict the temperature-dependent density and dynamic viscosity of 305 complex hydrocarbons. In our approach, strongly correlated/collinear predictors were trimmed, important predictors were selected by least absolute shrinkage and selection operator (LASSO) regularization and prior domain knowledge, hyperparameters were systematically optimized by Bayesian optimization, and the models were interpreted. The approach provided versatile and quantitative structure-property relationship (QSPR) models with relatively simple predictors for determining the dynamic viscosity and density of complex hydrocarbons at any temperature. In addition, we developed molecular dynamics simulation-based descriptors and evaluated the feasibility and versatility of dynamic descriptors from simulations for predicting the material properties. It was found that the models developed using a comparably smaller pool of dynamic descriptors performed similarly in predicting density and viscosity to models based on many more static descriptors. The best models were shown to predict density and dynamic viscosity with coefficient of determination ( R 2 ) values of 99.6% and 97.7%, respectively, for all data sets, including a test data set of 45 molecules. Finally, partial dependency plots (PDPs), individual conditional expectation (ICE) plots, local interpretable model-agnostic explanation (LIME) values, and trimmed model R 2 values were used to identify the most important static and dynamic predictors of the density and viscosity.

Federated multipartner machine learning has been touted as an appealing and efficient method to increase the effective training data volume and thereby the predictivity of models, particularly when the generation of training data is resource-intensive. In the landmark MELLODDY project, indeed, each of ten pharmaceutical companies realized aggregated improvements on its own classification or regression models through federated learning. To this end, they leveraged a novel implementation extending multitask learning across partners, on a platform audited for privacy and security. The experiments involved an unprecedented cross-pharma data set of 2.6+ billion confidential experimental activity data points, documenting 21+ million physical small molecules and 40+ thousand assays in on-target and secondary pharmacodynamics and pharmacokinetics. Appropriate complementary metrics were developed to evaluate the predictive performance in the federated setting. In addition to predictive performance increases in labeled space, the results point toward an extended applicability domain in federated learning. Increases in collective training data volume, including by means of auxiliary data resulting from single concentration high-throughput and imaging assays, continued to boost predictive performance, albeit with a saturating return. Markedly higher improvements were observed for the pharmacokinetics and safety panel assay-based task subsets.

The biophysical basis of passive membrane permeability is well-understood, but most methods for predicting membrane permeability in the context of drug design are based on statistical relationships that indirectly capture the key physical aspects. Here, we investigate molecular mechanics-based models of passive membrane permeability and evaluate their performance against different types of experimental data, including parallel artificial membrane permeability assays (PAMPA), cell-based assays, in vivo measurements, and other in silico predictions. The experimental data sets we use in these tests are diverse, including peptidomimetics, congeneric series, and diverse FDA approved drugs. The physical models are not specifically trained for any of these data sets; rather, input parameters are based on standard molecular mechanics force fields, such as partial charges, and an implicit solvent model. A systematic approach is taken to analyze the contribution from each component in the physics-based permeability model. A primary factor in determining rates of passive membrane permeation is the conformation-dependent free energy of desolvating the molecule, and this measure alone provides good agreement with experimental permeability measurements in many cases. Other factors that improve agreement with experimental data include deionization and estimates of entropy losses of the ligand and the membrane, which lead to size-dependence of the permeation rate.

Artificial Neural Networks (ANNs) are transforming how we understand chemical mixtures, providing an expressive view of the chemical space and multiscale processes. Their hybridization with physical knowledge can bridge the gap between predictivity and understanding of the underlying processes. This overview explores recent progress in ANNs, particularly their potential in the 'recomposition' of chemical mixtures. Graph-based representations reveal patterns among mixture components, and deep learning models excel in capturing complexity and symmetries when compared to traditional Quantitative Structure-Property Relationship models. Key components, such as Hamiltonian networks and convolution operations, play a central role in representing multiscale mixtures. The integration of ANNs with Chemical Reaction Networks and Physics-Informed Neural Networks for inverse chemical kinetic problems is also examined. The combination of sensors with ANNs shows promise in optical and biomimetic applications. A common ground is identified in the context of statistical physics, where ANN-based methods iteratively adapt their models by blending their initial states with training data. The concept of mixture recomposition unveils a reciprocal inspiration between ANNs and reactive mixtures, highlighting learning behaviors influenced by the training environment.

In this work, we study the use of three configurations of an autoencoder neural network to process organic substances with the aim of generating meaningful molecular descriptors that can be employed to develop property prediction models. A total of 18,322,500 compounds represented as SMILES strings were used to train the model, demonstrating that a latent space of 24 units is able to adequately reconstruct the data. After AE training, an analysis of the latent space properties in terms of compound similarity was carried out, indicating that this space possesses desired properties for the potential development of models for forecasting physical properties of organic compounds. As a final step, a QSPR model was developed to predict the boiling point of chemical substances based on the AE descriptors. 5276 substances were used for the regression task, and the predictive ability was compared with models available in the literature evaluated on the same database. The final AE model has an overall error of 1.40% (1.39% with augmented SMILES) in the prediction of the boiling temperature, while other models have errors between 2.0 and 3.2%. This shows that the SMILES representation is comparable and even outperforms the state-of-the-art representations widely used in the literature.

De novo molecular design plays an important role in drug discovery. Here, a novel generative model, Tree-Invent, was proposed to integrate topological constraints in the generation of a molecular graph. In this model, a molecular graph is represented as a topological tree in which a ring system, a nonring atom, and a chemical bond are regarded as the ring node, single node, and edge, respectively. The molecule generation is driven by three independent submodels for carrying out operations of node addition, ring generation, and node connection. One unique feature of the generative model is that the topological tree structure can be specified as a constraint for structure generation, which provides more precise control of structure generation. Combined with reinforcement learning, the Tree-Invent model could efficiently explore targeted chemical space. Moreover, the Tree-Invent model is flexible enough to be used in versatile molecule design settings such as scaffold decoration, scaffold hopping, and linker generation.

Polo-like kinase 1 (PLK1) and p38γ mitogen-activated protein kinase (p38γ) play important roles in cancer pathogenesis by controlling cell cycle progression and are therefore attractive cancer targets. The design of multitarget inhibitors may offer synergistic inhibition of distinct targets and reduce the risk of drug-drug interactions to improve the balance between therapeutic efficacy and safety. We combined deep-learning-based quantitative structure-activity relationship (QSAR) modeling and hybrid-based consensus scoring to screen for inhibitors with potential activity against the targeted proteins. Using this combination strategy, we identified a potent PLK1 inhibitor (compound 4 ) that inhibited PLK1 activity and liver cancer cell growth in the nanomolar range. Next, we deployed both our QSAR models for PLK1 and p38γ on the Enamine compound library to identify dual-targeting inhibitors against PLK1 and p38γ. Likewise, the identified hits were subsequently subjected to hybrid-based consensus scoring. Using this method, we identified a promising compound (compound 14 ) that could inhibit both PLK1 and p38γ activities. At nanomolar concentrations, compound 14 inhibited the growth of human hepatocellular carcinoma and hepatoblastoma cells in vitro. This study demonstrates the combined screening strategy to identify novel potential inhibitors for existing targets.

Substituents modulate reactions, but their effects are commonly described by using proxies to their functional group properties. Substituent descriptors from the quantum theory of atoms in molecules, which are true functional group properties, are related here to these proxies, which have historically had chemically relevant meaning. Due to the large number of descriptors, multivariate analysis is used to intuit their significance. Multiple linear regression, principal component, and partial least squares regression analyses highlight that these substituent descriptors contain similar information to the proxies while being intrinsic, predictable substituent properties. Sources of error limiting quantitative reproduction of the proxy data include transferability, experimental accuracy, and solvation issues.