Best Practices for QSAR Model Reporting: Physical and Chemical Properties, Ecotoxicity, Environmental Fate, Human Health, and Toxicokinetics Endpoints

ARKA: a framework of dimensionality reduction for machine-learning classification modeling, risk assessment, and data gap-filling of sparse environmental toxicity data

Due to the lack of experimental toxicity data for environmental chemicals, there arises a need to fill data gaps by in silico approaches. One of the most commonly used in silico approaches for toxicity assessment of small datasets is the Quantitative Structure-Activity Relationship (QSAR), which generates predictive models for the efficient prediction of query compounds. However, the reliability of the predictions from QSARs derived from small datasets is often questionable from a statistical point of view. This is due to the presence of a larger number of descriptors as compared to the number of training compounds, which reduces the degree of freedom of the developed model. To reduce the overall prediction error for a particular QSAR model, we have proposed here the computation of the novel Arithmetic Residuals in K-groups Analysis (ARKA) descriptors. We have reduced the number of modeling descriptors in a supervised manner by partitioning them into K classes (K = 2 here) depending on the higher mean normalized values of the descriptors to a particular response class, thus preventing the loss of chemical information. A scatter plot of the data points using the values of two ARKA descriptors (ARKA_2 vs. ARKA_1) can potentially identify activity cliffs, less confident data points, and less modelable data points. We have used here five representative environmentally relevant endpoints (skin sensitization, earthworm toxicity, milk/plasma partitioning, algal toxicity, and rodent carcinogenicity of hazardous chemicals) with graded responses to which the ARKA framework was applied for classification modeling. On comparing the performance of the models generated using conventional QSAR descriptors and the ARKA descriptors, the prediction quality of the models derived from ARKA descriptors was found, based on multiple graded-data validation metrics-derived decision criteria, much better than the models derived from QSAR descriptors signifying the potential of ARKA descriptors in ecotoxicological classification modeling of small data sets. Additionally, this holds true for the Read-Across approach as well, since the Read-Across predictions using ARKA descriptors supersede the predictions generated from QSAR descriptors. For the ease of users, a Java-based expert system has been developed that computes the ARKA descriptors from the input of QSAR descriptors.

Machine Learning-Based Prediction of Reduction Potentials for PtIV Complexes

Some of the well-known drawbacks of clinically approved PtII complexes can be overcome using six-coordinate PtIV complexes as inert prodrugs, which release the corresponding four-coordinate active PtII species upon reduction by cellular reducing agents. Therefore, the key factor of PtIV prodrug mechanism of action is their tendency to be reduced which, when the involved mechanism is of outer-sphere type, is measured by the value of the reduction potential. Machine learning (ML) models can be used to effectively capture intricate relationships within PtIV complex data, leading to highly accurate predictions of reduction potentials and other properties, and offering significant insights into their electrochemical behavior and potential applications. In this study, a machine learning-based approach for predicting the reduction potentials of PtIV complexes based on relevant molecular descriptors is presented. Leveraging a data set of experimentally determined reduction potentials and a diverse range of molecular descriptors, the proposed model demonstrates remarkable predictive accuracy (MSE = 0.016 V2, RMSE = 0.13 V, R2 = 0.92). Ab initio calculations and a set of different machine learning algorithms and feature engineering techniques have been employed to systematically explore the relationship between molecular structure and similarity and reduction potential. Specifically, it has been investigated whether the reduction potential of these compounds can be described by combining ML models across different combinations of constitutional, topological, and electronic molecular descriptors. Our results not only provide insights into the crucial factors influencing reduction potentials but also offer a rapid and effective tool for the rational design of PtIV complexes with tailored electrochemical properties for pharmaceutical applications. This approach has the potential to significantly expedite the development and screening of novel PtIV prodrug candidates. The analysis of principal components and key features extracted from the model highlights the significance of structural descriptors of the 2D Atom Pairs type and the lowest unoccupied molecular orbital energy. Specifically, with just 20 appropriately selected descriptors, a notable separation of complexes based on their reduction potential value is achieved.

Pesticide effect on earthworm lethality via interpretable machine learning

Earthworms are among the most important animals (invertebrates) for soil health. Many chemical substances released into nature for agricultural development, such as pesticides, may have unwanted effects on those organisms. However, it is essential to understand the extent of the impact of chemicals on soil health first and then make the proper decisions for regulatory or commercial purposes. We hypothesize that there is an expressible quantitative structure-activity relationship (QSAR) between the structure of pesticide compounds and the acute toxicity effect of earthworm species Eisenia fetida. The description of this relationship allows for a better assessment of the impact of chemicals on the said earthworm. To describe this relationship, a dataset of chemicals was collected from open-access sources to develop a mathematical model. A novel approach, combining genetic algorithm and Bayesian optimization, was used to select structural features into the model and to optimize model parameters. The final QSAR classification model was created with the Random Forest algorithm and exhibited good prediction Accuracy of 0.78 on training set and 0.80 on test set. The model representation follows FAIR principles and is available on QsarDB.org.

Interpretable machine learning for the identification of estrogen receptor agonists, antagonists, and binders

An abnormal hormonal activity or exposure to endocrine-disrupting chemicals (EDCs) can cause endocrine system malfunction. Among the many interactions EDCs can affect is the disruption of estrogen signalling, which can lead to adverse health effects such as cancer, osteoporosis, neurodegenerative diseases, cardiovascular disease, insulin resistance, and obesity. Knowing which chemical can act as an EDC is a significant advantage and a practical necessity. New Approach Methodologies (NAM) computational models offer a quick and cost-effective solution for preliminary hazard assessment of chemicals without animal testing. Therefore, a machine learning approach was used to investigate the relationships between estrogen receptor (ER) activity and chemical structure to identify chemicals that can interact with ER. For this purpose, the consolidated in vitro assay data from ToxCast/Tox21 projects was used for developing Random Forest classification models for ER binding, agonists, and antagonists. The overall classification prediction accuracy reaches up to 82%, depending on whether the model predicted agonists, antagonists, or compounds that bind to the active site. Given the imbalance in endocrine disruption data, the derived models are good candidates for deprioritising chemicals and reducing animal testing. The interpretation of theoretical molecular descriptors of the models was consistent with the molecular interactions known in the ligand binding pocket. The estimated class probabilities enabled the analysis of the applicability domain of the developed models and the assessment of the predictions' reliability, followed by the guidelines for interpreting prediction results. The models are openly accessible and useable at QsarDB.org (http://dx.doi.org/10.15152/QDB.259) according to the FAIR (Findable, Accessible, Interoperable, Reusable) principles.

Integrated virtual screening and MD simulation approaches toward discovering potential inhibitors for targeting BRPF1 bromodomain in hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is one of the most aggressive and life-threatening cancers. Although multiple treatment options are available, the prognosis of HCC patients is poor due to metastasis and drug resistance. Hence, discovering novel targets is essential for better therapeutic development for HCC. In this study, we used the cancer genome atlas (TCGA) dataset to analyze the expression of bromodomain-containing proteins in HCC, as bromodomains are emerging attractive therapeutic targets. Our analysis identified BRPF1 as the most highly upregulated gene in HCC among the 43 bromodomain-containing genes. Upregulation of BRPF1 was significantly associated with poorer patient survival. Therefore, targeting BRPF1 may be an approach for HCC treatment. Previously, several potential inhibitors of BRPF1 bromodomain have been discovered. However, due to the limited clinical success of the current inhibitors, we aim to search for new inhibitors with high affinity and specificity for the BRPF1 bromodomain. In this study, we utilized high-throughput virtual screening methods to screen synthetic and natural compound databases against the BRPF1 bromodomain. In addition, we used machine learning-based QSAR modeling to predict the IC50 values of the selected BRPF1 bromodomain inhibitors. Extensive MD simulations were used to calculate the binding free energies of BRPF1 bromodomain and inhibitor complexes. Using this approach, we identified four lead scaffolds with a similar or better binding affinity towards the BRPF1 bromodomain than the previously reported inhibitors. Overall, this study discovered some promising compounds that have the potential to act as potent BRPF1 bromodomain inhibitors.

From Black Boxes to Actionable Insights: A Perspective on Explainable Artificial Intelligence for Scientific Discovery

The application of Explainable Artificial Intelligence (XAI) in the field of chemistry has garnered growing interest for its potential to justify the prediction of black-box machine learning models and provide actionable insights. We first survey a range of XAI techniques adapted for chemical applications and categorize them based on the technical details of each methodology. We then present a few case studies to illustrate the practical utility of XAI, such as identifying carcinogenic molecules and guiding molecular optimizations, in order to provide chemists with concrete examples of ways to take full advantage of XAI-augmented machine learning for chemistry. Despite the initial success of XAI in chemistry, we still face the challenges of developing more reliable explanations, assuring robustness against adversarial actions, and customizing the explanation for different applications and needs of the diverse scientific community. Finally, we discuss the emerging role of large language models like GPT in generating natural language explanations and discusses the specific challenges associated with them. We advocate that addressing the aforementioned challenges and actively embracing new techniques may contribute to establishing machine learning as an indispensable technique for chemistry in this digital era.

Advancing Computational Toxicology by Interpretable Machine Learning

Chemical toxicity evaluations for drugs, consumer products, and environmental chemicals have a critical impact on human health. Traditional animal models to evaluate chemical toxicity are expensive, time-consuming, and often fail to detect toxicants in humans. Computational toxicology is a promising alternative approach that utilizes machine learning (ML) and deep learning (DL) techniques to predict the toxicity potentials of chemicals. Although the applications of ML- and DL-based computational models in chemical toxicity predictions are attractive, many toxicity models are "black boxes" in nature and difficult to interpret by toxicologists, which hampers the chemical risk assessments using these models. The recent progress of interpretable ML (IML) in the computer science field meets this urgent need to unveil the underlying toxicity mechanisms and elucidate the domain knowledge of toxicity models. In this review, we focused on the applications of IML in computational toxicology, including toxicity feature data, model interpretation methods, use of knowledge base frameworks in IML development, and recent applications. The challenges and future directions of IML modeling in toxicology are also discussed. We hope this review can encourage efforts in developing interpretable models with new IML algorithms that can assist new chemical assessments by illustrating toxicity mechanisms in humans.

In silico simulations and molecular descriptors to predict in vitro transactivation potencies of Baikal seal estrogen receptors by environmental contaminants

Baikal seals (Pusa sibirica) are vulnerable to high levels of organic pollutants. Here, we evaluated the transactivation potencies of bisphenols (BPs) and hydroxylated polychlorinated biphenyls (OH-PCBs) via the Baikal seal estrogen receptor α and β (bsERα and bsERβ) using in vitro and in silico approaches. In vitro reporter gene assays showed that most BPs and OH-PCBs exhibited estrogenic activity with bsER sub-type-specific potency. Among the BPs tested, bisphenol AF showed the lowest EC50 for both bsERs. 4'-OH-CB50 and 4'-OH-CB30 showed the lowest EC50 among OH-PCBs tested for bsERα and bsERβ, respectively. 4-((4-Isopropoxyphenyl)-sulfonyl)phenol, 4'-OH-CB72, and 4'-OH-CB121 showed weak bsERα-specific transactivation. Only 4-OH-CB107 did not affect both bsERs. In silico docking simulations revealed the binding affinities of these chemicals to bsERs and partially explained the in vitro results. Using the in silico simulations and molecular descriptors as explanatory variables and the in vitro results as objective variables, the quantitative structure-activity relationship (QSAR) models constructed for classification and regression accurately separated bsER-active compounds from non-active compounds and predicted the in vitro bsERα- and bsERβ-transactivation potencies, respectively. The QSAR models also suggested that chemical polarity, van der Waals surface area, bridging atom structure, position of the phenolic-OH group, and ligand interactions with key residues of the ligand binding pocket are critical variables to account for the bsER transactivation potency of the test compounds. We also succeeded in constructing computational models for predicting in vitro transactivation potencies of mouse ERs in the same manner, demonstrating the applicability of our approach independent of species-specific responses.

Identification and Prioritization of Environmental Organic Pollutants: From an Analytical and Toxicological Perspective

Exposure to environmental organic pollutants has triggered significant ecological impacts and adverse health outcomes, which have been received substantial and increasing attention. The contribution of unidentified chemical components is considered as the most significant knowledge gap in understanding the combined effects of pollutant mixtures. To address this issue, remarkable analytical breakthroughs have recently been made. In this review, the basic principles on recognition of environmental organic pollutants are overviewed. Complementary analytical methodologies (i.e., quantitative structure-activity relationship prediction, mass spectrometric nontarget screening, and effect-directed analysis) and experimental platforms are briefly described. The stages of technique development and/or essential parts of the analytical workflow for each of the methodologies are then reviewed. Finally, plausible technique paths and applications of the future nontarget screening methods, interdisciplinary techniques for achieving toxicant identification, and burgeoning strategies on risk assessment of chemical cocktails are discussed.

Machine learning approaches to study the structure-activity relationships of LpxC inhibitors

Antimicrobial resistance (AMR) has emerged as one of the global threats to human health in the 21st century. Drug discovery of inhibitors against novel targets rather than conventional bacterial targets has been considered an inevitable strategy for the growing threat of AMR infections. In this study, we applied quantitative structure-activity relationship (QSAR) modeling to the LpxC inhibitors to predict the inhibitory activity. In addition, we performed various cheminformatics analysis consisting of the exploration of the chemical space, identification of chemotypes, performing structure-activity landscape and activity cliffs as well as construction of the Structure-Activity Similarity (SAS) map. We built a total of 24 QSAR classification models using PubChem and MACCS fingerprint with 12 various machine learning algorithms. The best model with PubChem fingerprint is the Extremely Gradient Boost model (accuracy on the training set: 0.937; accuracy on the 10-fold cross-validation set: 0.795; accuracy on the test set: 0.799). Furthermore, it was found that the best model using the MACCS fingerprint was the Random Forest model (accuracy on the training set: 0.955; accuracy on the 10-fold cross-validation set: 0.803; accuracy on the test set: 0.785). In addition, we have identified eight consensus activity cliff generators that are highly informative for further SAR investigations. It is hoped that findings presented herein can provide guidance for further lead optimization of LpxC inhibitors.

Navigating bioactivity space in anti-tubercular drug discovery through the deployment of advanced machine learning models and cheminformatics tools: a molecular modeling based retrospective study

Mycobacterium tuberculosis is the bacterial strain that causes tuberculosis (TB). However, multidrug-resistant and extensively drug-resistant tuberculosis are significant obstacles to effective treatment. As a result, novel therapies against various strains of M. tuberculosis have been developed. Drug development is a lengthy procedure that includes identifying target protein and isolation, preclinical testing of the drug, and various phases of a clinical trial, etc., can take decades for a molecule to reach the market. Computational approaches such as QSAR, molecular docking techniques, and pharmacophore modeling have aided drug development. In this review article, we have discussed the various techniques in tuberculosis drug discovery by briefly introducing them and their importance. Also, the different databases, methods, approaches, and software used in conducting QSAR, pharmacophore modeling, and molecular docking have been discussed. The other targets targeted by these techniques in tuberculosis drug discovery have also been discussed, with important molecules discovered using these computational approaches. This review article also presents the list of drugs in a clinical trial for tuberculosis found drugs. Finally, we concluded with the challenges and future perspectives of these techniques in drug discovery.

ADis-QSAR: a machine learning model based on biological activity differences of compounds

Drug candidates identified by the pharmaceutical industry typically have unique structural characteristics to ensure they interact strongly and specifically with their biological targets. Identifying these characteristics is a key challenge for developing new drugs, and quantitative structure-activity relationship (QSAR) analysis has generally been used to perform this task. QSAR models with good predictive power improve the cost and time efficiencies invested in compound development. Generating these good models depends on how well differences between "active" and "inactive" compound groups can be conveyed to the model to be learned. Efforts to solve this difference issue have been made, including generating a "molecular descriptor" that compressively expresses the structural characteristics of compounds. From the same perspective, we succeeded in developing the Activity Differences-Quantitative Structure-Activity Relationship (ADis-QSAR) model by generating molecular descriptors that more explicitly convey features of the group through a pair system that performs direct connections between active and inactive groups. We used popular machine learning algorithms, such as Support Vector Machine, Random Forest, XGBoost and Multi-Layer Perceptron for model learning and evaluated the model using scores such as accuracy, area under curve, precision and specificity. The results showed that the Support Vector Machine performed better than the others. Notably, the ADis-QSAR model showed significant improvements in meaningful scores such as precision and specificity compared to the baseline model, even in datasets with dissimilar chemical spaces. This model reduces the risk of selecting false positive compounds, improving the efficiency of drug development.

Application of QSAR Approach to Assess the Effects of Organic Pollutants on Bacterial Virulence Factors

The release of a wide variety of persistent chemical contaminants into wastewater has become a growing concern due to their potential health and environmental risks. While the toxic effects of these pollutants on aquatic organisms have been extensively studied, their impact on microbial pathogens and their virulence mechanisms remains largely unexplored. This research paper focuses on the identification and prioritization of chemical pollutants that increase bacterial pathogenicity, which is a public health concern. In order to predict how chemical compounds, such as pesticides and pharmaceuticals, would affect the virulence mechanisms of three bacterial strains (Escherichia coli K12, Pseudomonas aeruginosa H103, and Salmonella enterica serovar. Typhimurium), this study has developed quantitative structure-activity relationship (QSAR) models. The use of analysis of variance (ANOVA) functions assists in developing QSAR models based on the chemical structure of the compounds, to predict their effect on the growth and swarming behavior of the bacterial strains. The results showed an uncertainty in the created model, and that increases in virulence factors, including growth and motility of bacteria, after exposure to the studied compounds are possible to be predicted. These results could be more accurate if the interactions between groups of functions are included. For that, to make an accurate and universal model, it is essential to incorporate a larger number of compounds of similar and different structures.

Study of Lipophilicity and ADME Properties of 1,9-Diazaphenothiazines with Anticancer Action

Lipophilicity is one of the key properties of a potential drug that determines the solubility, the ability to penetrate through cell barriers, and transport to the molecular target. It affects pharmacokinetic processes such as adsorption, distribution, metabolism, excretion (ADME). The 10-substituted 1,9-diazaphenothiazines show promising if not impressive in vitro anticancer potential, which is associated with the activation of the mitochondrial apoptosis pathway connected with to induction BAX, forming a channel in MOMP and releasing cytochrome c for the activation of caspases 9 and 3. In this publication, the lipophilicity of previously obtained 1,9-diazaphenothiazines was determined theoretically using various computer programs and experimentally using reverse-phase thin-layer chromatography (RP-TLC) and a standard curve. The study presents other physicochemical, pharmacokinetic, and toxicological properties affecting the bioavailability of the test compounds. ADME analysis was determined in silico using the SwissADME server. Molecular targets studies were identified in silico using the SwissTargetPrediction server. Lipinski's rule of five, Ghose's, and Veber's rules were checked for the tested compounds, confirming their bioavailability.

Making in silico predictive models for toxicology FAIR

In silico predictive models for toxicology include quantitative structure-activity relationship (QSAR) and physiologically based kinetic (PBK) approaches to predict physico-chemical and ADME properties, toxicological effects and internal exposure. Such models are used to fill data gaps as part of chemical risk assessment. There is a growing need to ensure in silico predictive models for toxicology are available for use and reproducible. This paper describes how the FAIR (Findable, Accessible, Interoperable, Reusable) principles, developed for data sharing, have been applied to in silico predictive models. In particular, this investigation has focussed on how the FAIR principles could be applied to improved regulatory acceptance of predictions from such models. Eighteen principles have been developed that cover all aspects of FAIR. It is intended that FAIRification of in silico predictive models for toxicology will increase their use and acceptance.

Exploring the Chemical Space of CYP17A1 Inhibitors Using Cheminformatics and Machine Learning

Cytochrome P450 17A1 (CYP17A1) is one of the key enzymes in steroidogenesis that produces dehydroepiandrosterone (DHEA) from cholesterol. Abnormal DHEA production may lead to the progression of severe diseases, such as prostatic and breast cancers. Thus, CYP17A1 is a druggable target for anti-cancer molecule development. In this study, cheminformatic analyses and quantitative structure-activity relationship (QSAR) modeling were applied on a set of 962 CYP17A1 inhibitors (i.e., consisting of 279 steroidal and 683 nonsteroidal inhibitors) compiled from the ChEMBL database. For steroidal inhibitors, a QSAR classification model built using the PubChem fingerprint along with the extra trees algorithm achieved the best performance, reflected by the accuracy values of 0.933, 0.818, and 0.833 for the training, cross-validation, and test sets, respectively. For nonsteroidal inhibitors, a systematic cheminformatic analysis was applied for exploring the chemical space, Murcko scaffolds, and structure-activity relationships (SARs) for visualizing distributions, patterns, and representative scaffolds for drug discoveries. Furthermore, seven total QSAR classification models were established based on the nonsteroidal scaffolds, and two activity cliff (AC) generators were identified. The best performing model out of these seven was model VIII, which is built upon the PubChem fingerprint along with the random forest algorithm. It achieved a robust accuracy across the training set, the cross-validation set, and the test set, i.e., 0.96, 0.92, and 0.913, respectively. It is anticipated that the results presented herein would be instrumental for further CYP17A1 inhibitor drug discovery efforts.

Understanding and overcoming the technical challenges in using in silico predictions in regulatory decisions of complex toxicological endpoints - A pesticide perspective for regulatory toxicologists with a focus on machine learning models

There are many challenges that must be overcome before in silico toxicity predictions are ripe for regulatory decision-making. Today, mandates in the United States of America and the European Union to avoid animal usage in toxicity testing is driving the need to consider alternative technologies, including Quantitative Structure Activity Relationship (QSAR) models, and read across approaches. However, when adopting new methods, it is critical that both new approach developers as well as regulatory users understand the strengths and challenges with these new approaches. In this paper, we identify potential sources of bias in machine learning methods specific to toxicity predictions, that may impact the overall performance of in silico models. We also discuss ways to mitigate these biases. Based on our experiences, the most prevalent sources of bias include class imbalance (differing numbers of "toxic" vs "nontoxic" compounds), limited numbers of chemicals within a particular chemistry, and biases within the studies that make up the database used for model building, as well as model evaluation biases. While this is already complex for repeated dose toxicity, in reproduction and developmental toxicity a further level of complexity is introduced by the need to evaluate effects on individual animal and litter basis (e.g., a hierarchal structure). We also discuss key considerations developers and regulators need to make when they use machine learning models to predict chemical safety. Our objective is for our paper to serve as a desk reference for model developers and regulators as they evaluate machine learning models and as they make decisions using these models.

Enabling data-limited chemical bioactivity predictions through deep neural network transfer learning

The main limitation in developing deep neural network (DNN) models to predict bioactivity properties of chemicals is the lack of sufficient assay data to train the network's classification layers. Focusing on feedforward DNNs that use atom- and bond-based structural fingerprints as input, we examined whether layers of a fully trained DNN based on large amounts of data to predict one property could be used to develop DNNs to predict other related or unrelated properties based on limited amounts of data. Hence, we assessed if and under what conditions the dense layers of a pre-trained DNN could be transferred and used for the development of another DNN associated with limited training data. We carried out a quantitative study employing more than 400 pairs of assay datasets, where we used fully trained layers from a large dataset to augment the training of a small dataset. We found that the higher the correlation r between two assay datasets, the more efficient the transfer learning is in reducing prediction errors associated with the smaller dataset DNN predictions. The reduction in mean squared prediction errors ranged from 10 to 20% for every 0.1 increase in r² between the datasets, with the bulk of the error reductions associated with transfers of the first dense layer. Transfer of other dense layers did not result in additional benefits, suggesting that deeper, dense layers conveyed more specialized and assay-specific information. Importantly, depending on the dataset correlation, training sample size could be reduced by up to tenfold without any loss of prediction accuracy.

Towards combating antibiotic resistance by exploring the quantitative structure-activity relationship of NDM-1 inhibitors

The emergence of New Delhi metallo-beta-lactamase-1 (NDM-1) has conferred enteric bacteria resistance to almost all beta-lactam antibiotics. Its capability of horizontal transfer through plasmids, amongst humans, animal reservoirs and the environment, has added up to the totality of antimicrobial resistance control, animal husbandry and food safety. Thus far, there have been no effective drugs for neutralizing NDM-1. This study explores the structure-activity relationship of NDM-1 inhibitors. IC₅₀ values of NDM-1 inhibitors were compiled from both the ChEMBL database and literature. After curation, a final set of 686 inhibitors were used for machine learning model building using the random forest algorithm against 12 sets of molecular fingerprints. Benchmark results indicated that the KlekotaRothCount fingerprint provided the best overall performance with an accuracy of 0.978 and 0.778 for the training and testing set, respectively. Model interpretation revealed that nitrogen-containing features (KRFPC 4080, KRFPC 3882, KRFPC 677, KRFPC 3608, KRFPC 3750, KRFPC 4287 and KRFPC 3943), sulfur-containing substructures (KRFPC 2855 and KRFPC 4843), aromatic features (KRFPC 1566, KRFPC 1564, KRFPC 1642, KRFPC 3608, KRFPC 4287 and KRFPC 3943), carbonyl features (KRFPC 1193 and KRFPC 3025), aliphatic features (KRFPC 2975, KRFPC 297, KRFPC 3224 and KRFPC 669) are features contributing to NDM-1 inhibitory activity. It is anticipated that findings from this study would help facilitate the drug discovery of NDM-1 inhibitors by providing guidelines for further lead optimization.

Chemical named entity recognition in the texts of scientific publications using the naïve Bayes classifier approach

Motivation

Application of chemical named entity recognition (CNER) algorithms allows retrieval of information from texts about chemical compound identifiers and creates associations with physical–chemical properties and biological activities. Scientific texts represent low-formalized sources of information. Most methods aimed at CNER are based on machine learning approaches, including conditional random fields and deep neural networks. In general, most machine learning approaches require either vector or sparse word representation of texts. Chemical named entities (CNEs) constitute only a small fraction of the whole text, and the datasets used for training are highly imbalanced.

Methods and results

We propose a new method for extracting CNEs from texts based on the naïve Bayes classifier combined with specially developed filters. In contrast to the earlier developed CNER methods, our approach uses the representation of the data as a set of fragments of text (FoTs) with the subsequent preparati`on of a set of multi-n-grams (sequences from one to n symbols) for each FoT. Our approach may provide the recognition of novel CNEs. For CHEMDNER corpus, the values of the sensitivity (recall) was 0.95, precision was 0.74, specificity was 0.88, and balanced accuracy was 0.92 based on five-fold cross validation. We applied the developed algorithm to the extracted CNEs of potential Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) main protease (Mpro) inhibitors. A set of CNEs corresponding to the chemical substances evaluated in the biochemical assays used for the discovery of Mpro inhibitors was retrieved. Manual analysis of the appropriate texts showed that CNEs of potential SARS-CoV-2 Mpro inhibitors were successfully identified by our method.

Conclusion

The obtained results show that the proposed method can be used for filtering out words that are not related to CNEs; therefore, it can be successfully applied to the extraction of CNEs for the purposes of cheminformatics and medicinal chemistry.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13321-022-00633-4.

Machine Learning Quantitative Structure-Property Relationships as a Function of Ionic Liquid Cations for the Gas-Ionic Liquid Partition Coefficient of Hydrocarbons

Ionic liquids (ILs) are known for their unique characteristics as solvents and electrolytes. Therefore, new ILs are being developed and adapted as innovative chemical environments for different applications in which their properties need to be understood on a molecular level. Computational data-driven methods provide means for understanding of properties at molecular level, and quantitative structure-property relationships (QSPRs) provide the framework for this. This framework is commonly used to study the properties of molecules in ILs as an environment. The opposite situation where the property is considered as a function of the ionic liquid does not exist. The aim of the present study was to supplement this perspective with new knowledge and to develop QSPRs that would allow the understanding of molecular interactions in ionic liquids based on the structure of the cationic moiety. A wide range of applications in electrochemistry, separation and extraction chemistry depends on the partitioning of solutes between the ionic liquid and the surrounding environment that is characterized by the gas-ionic liquid partition coefficient. To model this property as a function of the structure of a cationic counterpart, a series of ionic liquids was selected with a common bis-(trifluoromethylsulfonyl)-imide anion, [Tf2N]-, for benzene, hexane and cyclohexane. MLR, SVR and GPR machine learning approaches were used to derive data-driven models and their performance was compared. The cross-validation coefficients of determination in the range 0.71-0.93 along with other performance statistics indicated a strong accuracy of models for all data series and machine learning methods. The analysis and interpretation of descriptors revealed that generally higher lipophilicity and dispersion interaction capability, and lower polarity in the cations induces a higher partition coefficient for benzene, hexane, cyclohexane and hydrocarbons in general. The applicability domain analysis of models concluded that there were no highly influential outliers and the models are applicable to a wide selection of cation families with variable size, polarity and aliphatic or aromatic nature.

Linking electron ionization mass spectra of organic chemicals to toxicity endpoints through machine learning and experimentation

Quantitative structure-activity relationship (QSAR) modeling has been widely used to predict the potential harm of chemicals, in which the prediction heavily relies on the accurate annotation of chemical structures. However, it is difficult to determine the accurate structure of an unknown compound in many cases, such as in complex water environments. Here, we solved the above problem by linking electron ionization mass spectra (EI-MS) of organic chemicals to toxicity endpoints through various machine learning methods. The proposed method was verified by predicting 50% growth inhibition of Tetrahymena pyriformis (T. pyriformis) and liver toxicity. The optimal model performance obtained an R² > 0.7 or balanced accuracy > 0.72 for both the training set and test set. External experimentation further verified the application potential of our proposed method in the toxicity prediction of unknown chemicals. Feature importance analysis allowed us to identify critical spectral features that were responsible for chemical-induced toxicity. Our approach has the potential for toxicity prediction in such fields that it is difficult to determine accurate chemical structures.

Extended continuous similarity indices: theory and application for QSAR descriptor selection

Extended (or n-ary) similarity indices have been recently proposed to extend the comparative analysis of binary strings. Going beyond the traditional notion of pairwise comparisons, these novel indices allow comparing any number of objects at the same time. This results in a remarkable efficiency gain with respect to other approaches, since now we can compare N molecules in O(N) instead of the common quadratic O(N²) timescale. This favorable scaling has motivated the application of these indices to diversity selection, clustering, phylogenetic analysis, chemical space visualization, and post-processing of molecular dynamics simulations. However, the current formulation of the n-ary indices is limited to vectors with binary or categorical inputs. Here, we present the further generalization of this formalism so it can be applied to numerical data, i.e. to vectors with continuous components. We discuss several ways to achieve this extension and present their analytical properties. As a practical example, we apply this formalism to the problem of feature selection in QSAR and prove that the extended continuous similarity indices provide a convenient way to discern between several sets of descriptors.

Machine learning models on chemical inhibitors of mitochondrial electron transport chain

Chemicals can induce adverse effects in humans by inhibiting mitochondrial electron transport chain (ETC) such as disrupting mitochondrial membrane potential, enhancing oxidative stress and causing some diseases. Thus, identifying ETC inhibitors (ETCi) is important to chemical risk assessment and protecting the public health. However, it is not feasible to identify all ETCi with experimental methods. Quantitative structure-activity relationship (QSAR) modeling is a promising method to rapidly and effectively identify ETCi. In this study, QSAR models for predicting ETCi were developed using machine learning methods. A clustering-based under-sampling (CBUS) method was developed to handle the imbalance issue in training sets. Structure-activity landscapes were generated and analyzed for training sets generated by the CBUS method. The consensus QSAR models constructed with CBUS achieved satisfactory performances (balanced accuracy = 0.852) in 100 iterations of five-fold cross validations, indicating the models can effectively classify ETCi. The classification model was further employed to screen chemicals in the Inventory of Existing Chemical Substances of China and 13 chemicals were identified as ETCi. Fifteen structural alerts for ETCi were identified in this study. These results demonstrated that the model and structural alerts are useful to screen ETCi.

A Classification Model to Identify Direct-Acting Mutagenic Polycyclic Aromatic Hydrocarbon Transformation Products

Polycyclic aromatic hydrocarbons (PAHs) are a complex group of environmental contaminants, many having long environmental half-lives. As these compounds degrade, the changes in their structure can result in a substantial increase in mutagenicity compared to the parent compound. Over time, each individual PAH can potentially degrade into several thousand unique transformation products, creating a complex, constantly evolving set of intermediates. Microbial degradation is the primary mechanism of their transformation and ultimate removal from the environment, and this process can result in mutagenic activation similar to the metabolic activation that can occur in multicellular organisms. The diversity of the potential intermediate structures in PAH-contaminated environments renders hazard assessment difficult for both remediation professionals and regulators. A mixture of structural and energetic descriptors has proven effective in existing studies for classifying which PAH transformation products will be mutagenic. However, most existing studies of environmental PAH mutagens primarily focus on nitrogenated derivatives, which are prevalent in the atmosphere and not as relevant in soil. Additionally, PAH products commonly found in the environment can range from as large as five rings to as small as a single ring, requiring a broadly inclusive methodology to comprehensively evaluate mutagenic potential. We developed a combination of supervised and unsupervised machine learning methods to predict environmentally induced PAH mutagenicity with improved performance over currently available tools. K-means clustering with principal component analysis allows us to identify molecular clusters that we hypothesize to have similar mechanisms of action. Recursive feature elimination identifies the most influential descriptors. The cluster-specific regression outperforms available classifiers in predicting direct-acting mutagens resulting from the microbial biodegradation of PAHs and provides direction for future studies evaluating the environmental hazards resulting from PAH biodegradation.

dGPredictor: Automated fragmentation method for metabolic reaction free energy prediction and de novo pathway design

Group contribution (GC) methods are conventionally used in thermodynamics analysis of metabolic pathways to estimate the standard Gibbs energy change (Δ_rG′^o) of enzymatic reactions from limited experimental measurements. However, these methods are limited by their dependence on manually curated groups and inability to capture stereochemical information, leading to low reaction coverage. Herein, we introduce an automated molecular fingerprint-based thermodynamic analysis tool called dGPredictor that enables the consideration of stereochemistry within metabolite structures and thus increases reaction coverage. dGPredictor has comparable prediction accuracy compared to existing GC methods and can capture Gibbs energy changes for isomerase and transferase reactions, which exhibit no overall group changes. We also demonstrate dGPredictor’s ability to predict the Gibbs energy change for novel reactions and seamless integration within de novo metabolic pathway design tools such as novoStoic for safeguarding against the inclusion of reaction steps with infeasible directionalities. To facilitate easy access to dGPredictor, we developed a graphical user interface to predict the standard Gibbs energy change for reactions at various pH and ionic strengths. The tool allows customized user input of known metabolites as KEGG IDs and novel metabolites as InChI strings (https://github.com/maranasgroup/dGPredictor).

The standard Gibbs energy change is commonly used to check for the feasibility of enzyme-catalyzed reactions as thermodynamics plays a crucial role in pathway design for biochemical synthesis. The group contribution methods using expert-defined functional groups have been extensively used for estimating standard Gibbs energy change. Here, we introduce a molecular fingerprint-based thermodynamic tool, dGPredictor, that enables distinguishing between (stereo)isomers in metabolic reactions leading to improved reaction coverage and comparable prediction accuracy as GC methods. dGPredictor can also be used alongside de novo pathway design tools to ensure the correct directionality of chosen reaction steps. We applied and tested dGPredictor on reactions from the KEGG database and applied it to screen an isobutanol synthesis pathway design. An open-source, user-friendly web interface is provided to facilitate easy access for standard Gibbs energy change of reactions at different pH values. (https://github.com/maranasgroup/dGPredictor).

Web-Based Quantitative Structure-Activity Relationship Resources Facilitate Effective Drug Discovery

Traditional drug discovery effectively contributes to the treatment of many diseases but is limited by high costs and long cycles. Quantitative structure-activity relationship (QSAR) methods were introduced to evaluate the activity of compounds virtually, which saves the significant cost of determining the activities of the compounds experimentally. Over the past two decades, many web tools for QSAR modeling with various features have been developed to facilitate the usage of QSAR methods. These web tools significantly reduce the difficulty of using QSAR and indirectly promote drug discovery. However, there are few comprehensive summaries of these QSAR tools, and researchers may have difficulty determining which tool to use. Hence, we systematically surveyed the mainstream web tools for QSAR modeling. This work may guide researchers in choosing appropriate web tools for developing QSAR models, and may also help develop more bioinformatics tools based on these existing resources. For nonprofessionals, we also hope to make more people aware of QSAR methods and expand their use.

Machine learning models for classification tasks related to drug safety

In this review, we outline the current trends in the field of machine learning-driven classification studies related to ADME (absorption, distribution, metabolism and excretion) and toxicity endpoints from the past six years (2015-2021). The study focuses only on classification models with large datasets (i.e. more than a thousand compounds). A comprehensive literature search and meta-analysis was carried out for nine different targets: hERG-mediated cardiotoxicity, blood-brain barrier penetration, permeability glycoprotein (P-gp) substrate/inhibitor, cytochrome P450 enzyme family, acute oral toxicity, mutagenicity, carcinogenicity, respiratory toxicity and irritation/corrosion. The comparison of the best classification models was targeted to reveal the differences between machine learning algorithms and modeling types, endpoint-specific performances, dataset sizes and the different validation protocols. Based on the evaluation of the data, we can say that tree-based algorithms are (still) dominating the field, with consensus modeling being an increasing trend in drug safety predictions. Although one can already find classification models with great performances to hERG-mediated cardiotoxicity and the isoenzymes of the cytochrome P450 enzyme family, these targets are still central to ADMET-related research efforts.

Determination of "fitness-for-purpose" of quantitative structure-activity relationship (QSAR) models to predict (eco-)toxicological endpoints for regulatory use

In silico models are used to predict toxicity and molecular properties in chemical safety assessment, gaining widespread regulatory use under a number of legislations globally. This study has rationalised previously published criteria to evaluate quantitative structure-activity relationships (QSARs) in terms of their uncertainty, variability and potential areas of bias, into ten assessment components, or higher level groupings. The components have been mapped onto specific regulatory uses (i.e. data gap filling for risk assessment, classification and labelling, and screening and prioritisation) identifying different levels of uncertainty that may be acceptable for each. Twelve published QSARs were evaluated using the components, such that their potential use could be identified. High uncertainty was commonly observed with the presentation of data, mechanistic interpretability, incorporation of toxicokinetics and the relevance of the data for regulatory purposes. The assessment components help to guide strategies that can be implemented to improve acceptability of QSARs through the reduction of uncertainties. It is anticipated that model developers could apply the assessment components from the model design phase (e.g. through problem formulation) through to their documentation and use. The application of the components provides the possibility to assess QSARs in a meaningful manner and demonstrate their fitness-for-purpose against pre-defined criteria.

Binary and multi-class classification for androgen receptor agonists, antagonists and binders

Androgens and androgen receptor regulate a variety of biological effects in the human body. The impaired functioning of androgen receptor may have different adverse health effects from cancer to infertility. Therefore, it is important to determine whether new chemicals have any binding activity and act as androgen agonists or antagonists before commercial use. Due to the large number of chemicals that require experimental testing, the computational methods are a viable alternative. Therefore, the aim of the present study was to develop predictive QSAR models for classifying compounds according to their activity at the androgen receptor. A large data set of chemicals from the CoMPARA project was used for this purpose and random forest classification models have been developed for androgen binding, agonistic, and antagonistic activity. In addition, a unique effort has been made for multi-class approach that discriminates between inactive compounds, agonists and antagonists simultaneously. For the evaluation set, the classification models predicted agonists with 80% of accuracy and for the antagonists' and binders' the respective metrics were 72% and 78%. Combining agonists, antagonists and inactive compounds into a multi-class approach added complexity to the modelling task and resulted to 64% prediction accuracy for the evaluation set. Considering the size of the training data sets and their imbalance, the achieved evaluation accuracy is very good. The final classification models are available for exploring and predicting at QsarDB repository (https://doi.org/10.15152/QDB.236).

Do we need different machine learning algorithms for QSAR modeling? A comprehensive assessment of 16 machine learning algorithms on 14 QSAR data sets

Although a wide variety of machine learning (ML) algorithms have been utilized to learn quantitative structure-activity relationships (QSARs), there is no agreed single best algorithm for QSAR learning. Therefore, a comprehensive understanding of the performance characteristics of popular ML algorithms used in QSAR learning is highly desirable. In this study, five linear algorithms [linear function Gaussian process regression (linear-GPR), linear function support vector machine (linear-SVM), partial least squares regression (PLSR), multiple linear regression (MLR) and principal component regression (PCR)], three analogizers [radial basis function support vector machine (rbf-SVM), K-nearest neighbor (KNN) and radial basis function Gaussian process regression (rbf-GPR)], six symbolists [extreme gradient boosting (XGBoost), Cubist, random forest (RF), multiple adaptive regression splines (MARS), gradient boosting machine (GBM), and classification and regression tree (CART)] and two connectionists [principal component analysis artificial neural network (pca-ANN) and deep neural network (DNN)] were employed to learn the regression-based QSAR models for 14 public data sets comprising nine physicochemical properties and five toxicity endpoints. The results show that rbf-SVM, rbf-GPR, XGBoost and DNN generally illustrate better performances than the other algorithms. The overall performances of different algorithms can be ranked from the best to the worst as follows: rbf-SVM > XGBoost > rbf-GPR > Cubist > GBM > DNN > RF > pca-ANN > MARS > linear-GPR ≈ KNN > linear-SVM ≈ PLSR > CART ≈ PCR ≈ MLR. In terms of prediction accuracy and computational efficiency, SVM and XGBoost are recommended to the regression learning for small data sets, and XGBoost is an excellent choice for large data sets. We then investigated the performances of the ensemble models by integrating the predictions of multiple ML algorithms. The results illustrate that the ensembles of two or three algorithms in different categories can indeed improve the predictions of the best individual ML algorithms.

Modelling of antiproliferative activity measured in HeLa cervical cancer cells in a series of xanthene derivatives

Cancer remains one of the leading causes of death in humans, and new drug substances are therefore being developed. Thus, the anti-cancer activity of xanthene derivatives has become an important topic in the development of new and potent anti-cancer drug substances. Previously published novel series of xanthen-3-one and xanthen-1,8-dione derivatives have been synthesized in one of our laboratories and showed anti-proliferative activity in HeLa cancer cell lines. This series serves as a good basis to develop quantitative structure-activity relationship (QSAR), to study the relations between anti-proliferative activity and chemical structures. A QSAR model has been derived that relies only on two-dimensional molecular descriptors, providing mechanistic insight into the anti-proliferative activity of xanthene derivatives. The model is validated internally and externally and additionally with the set of inactive compounds of the original data, confirming model applicability for the design and discovery of novel xanthene derivatives. The QSAR model is available at the QsarDB repository (http://dx.doi.10.15152/QDB.237).

A Review of In Silico Tools as Alternatives to Animal Testing: Principles, Resources and Applications

Across the spectrum of industrial sectors, including pharmaceuticals, chemicals, personal care products, food additives and their associated regulatory agencies, there is a need to develop robust and reliable methods to reduce or replace animal testing. It is generally recognised that no single alternative method will be able to provide a one-to-one replacement for assays based on more complex toxicological endpoints. Hence, information from a combination of techniques is required. A greater understanding of the time and concentration-dependent mechanisms, underlying the interactions between chemicals and biological systems, and the sequence of events that can lead to apical effects, will help to move forward the science of reducing and replacing animal experiments. In silico modelling, in vitro assays, high-throughput screening, organ-on-a-chip technology, omics and mathematical biology, can provide complementary information to develop a complete picture of the potential response of an organism to a chemical stressor. Adverse outcome pathways (AOPs) and systems biology frameworks enable relevant information from diverse sources to be logically integrated. While individual researchers do not need to be experts across all disciplines, it is useful to have a fundamental understanding of what other areas of science have to offer, and how knowledge can be integrated with other disciplines. The purpose of this review is to provide those who are unfamiliar with predictive in silico tools, with a fundamental understanding of the underlying theory. Current applications, software, barriers to acceptance, new developments and the use of integrated approaches are all discussed, with additional resources being signposted for each of the topics.

Development of classification models for predicting inhibition of mitochondrial fusion and fission using machine learning methods

Mitochondrial fusion and fission are processes to maintain mitochondrial function when cells respond to environment stresses. Disruption of mitochondrial fusion and fission influences cell health and can cause adverse events such as neurodegenerative disorders. It is critical to identify environmental chemicals that can disrupt mitochondrial fusion and fission. However, experimentally testing all the chemicals is not practical because experimental methods are time-consuming and costly. Quantitative structure-activity relationship (QSAR) modeling is an attractive approach for evaluation of chemicals disrupting potential on mitochondrial fusion and fission. In this study, QSAR models were developed for differentiating chemicals capable of inhibition of mitochondrial fusion and fission using machine learning algorithms (i.e. random forest, logistic regression, Bernoulli naive Bayes, and deep neural network). One hundred iterations of five-fold cross validations and external validations showed that the best model on mitochondrial fusion had area under the receiver operating characteristic curve (AUC) of 82.8% and 78.1%, respectively; and the best model for mitochondrial fission yielded AUC of 84.3% and 97.5%, respectively. Furthermore, 45 and 56 structural alerts were identified for inhibition of mitochondrial fusion and fission, respectively. The results demonstrated that the models and the structural alerts could be useful for screening chemicals that inhibit mitochondrial fusion and fission.

Quantum chemical predictions of water-octanol partition coefficients applied to the SAMPL6 logP blind challenge

Theoretical approaches for predicting physicochemical properties are valuable tools for accelerating the drug discovery process. In this work, quantum chemical methods are used to predict water-octanol partition coefficients as a part of the SAMPL6 blind challenge. The SMD continuum solvent model was employed with MP2 and eight DFT functionals in conjunction with correlation consistent basis sets to determine the water-octanol transfer free energy. Several tactics towards improving the predictions of the partition coefficient were examined, including increasing the quality of basis sets, considering tautomerization, and accounting for inhomogeneities in the water and n-octanol phases. Evaluation of these various schemes highlights the impact of modeling approaches across different methods. With the inclusion of tautomers and adjustments to the permittivity constants, the best predictions were obtained with smaller basis sets and the O3LYP functional, which yielded an RMSE of 0.79 logP units. The results presented correspond to the SAMPL6 logP submission IDs: DYXBT, O7DJK, and AHMTF.

Multi-Level Comparison of Machine Learning Classifiers and Their Performance Metrics

Machine learning classification algorithms are widely used for the prediction and classification of the different properties of molecules such as toxicity or biological activity. the prediction of toxic vs. non-toxic molecules is important due to testing on living animals, which has ethical and cost drawbacks as well. The quality of classification models can be determined with several performance parameters. which often give conflicting results. In this study, we performed a multi-level comparison with the use of different performance metrics and machine learning classification methods. Well-established and standardized protocols for the machine learning tasks were used in each case. The comparison was applied to three datasets (acute and aquatic toxicities) and the robust, yet sensitive, sum of ranking differences (SRD) and analysis of variance (ANOVA) were applied for evaluation. The effect of dataset composition (balanced vs. imbalanced) and 2-class vs. multiclass classification scenarios was also studied. Most of the performance metrics are sensitive to dataset composition, especially in 2-class classification problems. The optimal machine learning algorithm also depends significantly on the composition of the dataset.