Predicting chemical ocular toxicity using a combinatorial QSAR approach

In silico prediction of ocular toxicity of compounds using explainable machine learning and deep learning approaches

The accurate identification of chemicals with ocular toxicity is of paramount importance in health hazard assessment. In contemporary chemical toxicology, there is a growing emphasis on refining, reducing, and replacing animal testing in safety evaluations. Therefore, the development of robust computational tools is crucial for regulatory applications. The performance of predictive models is heavily reliant on the quality and quantity of data. In this investigation, we amalgamated the most extensive dataset (4901 compounds) sourced from governmental GHS-compliant databases and literature to develop binary classification models of chemical ocular toxicity. We employed 12 molecular representations in conjunction with six machine learning algorithms and two deep learning algorithms to create a series of binary classification models. The findings indicated that the deep learning method GCN outperformed the machine learning models in cross-validation, achieving an impressive AUC of 0.915. However, the top-performing machine learning model (RF-Descriptor) demonstrated excellent performance with an AUC of 0.869 on the test set and was therefore selected as the best model. To enhance model interpretability, we conducted the SHAP method and attention weights analysis. The two approaches offered visual depictions of the relevance of key descriptors and substructures in predicting ocular toxicity of chemicals. Thus, we successfully struck a delicate balance between data quality and model interpretability, rendering our model valuable for predicting and comprehending potential ocular-toxic compounds in the early stages of drug discovery.

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.

Data-Driven Quantitative Structure-Activity Relationship Modeling for Human Carcinogenicity by Chronic Oral Exposure

Traditional methodologies for assessing chemical toxicity are expensive and time-consuming. Computational modeling approaches have emerged as low-cost alternatives, especially those used to develop quantitative structure-activity relationship (QSAR) models. However, conventional QSAR models have limited training data, leading to low predictivity for new compounds. We developed a data-driven modeling approach for constructing carcinogenicity-related models and used these models to identify potential new human carcinogens. To this goal, we used a probe carcinogen dataset from the US Environmental Protection Agency's Integrated Risk Information System (IRIS) to identify relevant PubChem bioassays. Responses of 25 PubChem assays were significantly relevant to carcinogenicity. Eight assays inferred carcinogenicity predictivity and were selected for QSAR model training. Using 5 machine learning algorithms and 3 types of chemical fingerprints, 15 QSAR models were developed for each PubChem assay dataset. These models showed acceptable predictivity during 5-fold cross-validation (average CCR = 0.71). Using our QSAR models, we can correctly predict and rank 342 IRIS compounds' carcinogenic potentials (PPV = 0.72). The models predicted potential new carcinogens, which were validated by a literature search. This study portends an automated technique that can be applied to prioritize potential toxicants using validated QSAR models based on extensive training sets from public data resources.

Prediction of serious eye damage or eye irritation potential of compounds via consensus labelling models and active learning models based on uncertainty strategies

Serious eye damage and eye irritation have been authenticated to be significant human health issues in various fields such as ophthalmic pharmaceuticals. Due to the shortcomings of traditional animal testing methods, in silico methods have advanced to study eye toxicity. The models for predicting serious eye damage and eye irritation potential of compounds were developed using 2299 and 5214 compounds, respectively. The 40 global single models and 40 local models were developed by combining 5 molecular description methods and 4 machine learning methods. The 40 active learning models were developed by adopting uncertainty-based active learning strategies and taking local models as initial models. The 110 global consensus models based on 40 global single models were developed using a consensus strategy. Active learning models and global consensus models performed high prediction accuracy. The test accuracy of the best serious eye damage model and eye irritation model reached 0.972 and 0.959, respectively. The applicability domains for all models were calculated to verify the rationality of prediction effect. In addition, 8 structural alerts probably causing serious eye damage or eye irritation were sought out. The prediction models and structural alerts contributed to providing hazard identification and assessing chemical safety.

Predicting Prenatal Developmental Toxicity Based On the Combination of Chemical Structures and Biological Data

For hazard identification, classification, and labeling purposes, animal testing guidelines are required by law to evaluate the developmental toxicity potential of new and existing chemical products. However, guideline developmental toxicity studies are costly, time-consuming, and require many laboratory animals. Computational modeling has emerged as a promising, animal-sparing, and cost-effective method for evaluating the developmental toxicity potential of chemicals, such as endocrine disruptors, without the use of animals. We aimed to develop a predictive and explainable computational model for developmental toxicants. To this end, a comprehensive dataset of 1244 chemicals with developmental toxicity classifications was curated from public repositories and literature sources. Data from 2140 toxicological high-throughput screening assays were extracted from PubChem and the ToxCast program for this dataset and combined with information about 834 chemical fragments to group assays based on their chemical-mechanistic relationships. This effort revealed two assay clusters containing 83 and 76 assays, respectively, with high positive predictive rates for developmental toxicants identified with animal testing guidelines (PPV = 72.4 and 77.3% during cross-validation). These two assay clusters can be used as developmental toxicity models and were applied to predict new chemicals for external validation. This study provides a new strategy for constructing alternative chemical developmental toxicity evaluations that can be replicated for other toxicity modeling studies.

In silico prediction of the full United Nations Globally Harmonized System eye irritation categories of liquid chemicals by IATA-like bottom-up approach of random forest method

As an alternative to in vivo Draize rabbit eye irritation test, this study aimed to construct an in silico model to predict the complete United Nations (UN) Globally Harmonized System (GHS) for classification and labeling of chemicals for eye irritation category [eye damage (Category 1), irritating to eye (Category 2) and nonirritating (No category)] of liquid chemicals with Integrated approaches to testing and assessment (IATA)-like two-stage random forest approach. Liquid chemicals (n = 219) with 34 physicochemical descriptors and quality in vivo data were collected with no missing values. Seven machine learning algorithms (Naive Bayes, Logistic Regression, First Large Margin, Neural Net, Random Forest (RF), Gradient Boosted Tree, and Support Vector Machine) were examined for the ternary categorization of eye irritation potential at a single run through 10-fold cross-validation. RF, which performed best, was further improved by applying the 'Bottom-up approach' concept of IATA, namely, separating No category first, and discriminating Category 1 from 2, thereafter. The best performing training dataset achieved an overall accuracy of 73% and the correct prediction for Category 1, 2, and No category was 80%, 50%, and 77%, respectively for the test dataset. This prediction model was further validated with an external dataset of 28 chemicals, for which an overall accuracy of 71% was achieved.

Novel computational models offer alternatives to animal testing for assessing eye irritation and corrosion potential of chemicals

Eye irritation and corrosion are fundamental considerations in developing chemicals to be used in or near the eye, from cleaning products to ophthalmic solutions. Unfortunately, animal testing is currently the standard method to identify compounds that cause eye irritation or corrosion. Yet, there is growing pressure on the part of regulatory agencies both in the USA and abroad to develop New Approach Methodologies (NAMs) that help reduce the need for animal testing and address unmet need to modernize safety evaluation of chemical hazards. In furthering the development and applications of computational NAMs in chemical safety assessment, in this study we have collected the largest expertly curated dataset of compounds tested for eye irritation and corrosion, and employed this data to build and validate binary and multi-classification Quantitative Structure-Activity Relationships (QSAR) models that can reliably assess eye irritation/corrosion potential of novel untested compounds. QSAR models were generated with Random Forest (RF) and Multi-Descriptor Read Across (MuDRA) machine learning (ML) methods, and validated using a 5-fold external cross-validation protocol. These models demonstrated high balanced accuracy (CCR of 0.68–0.88), sensitivity (SE of 0.61–0.84), positive predictive value (PPV of 0.65–0.90), specificity (SP of 0.56–0.91), and negative predictive value (NPV of 0.68–0.85). Overall, MuDRA models outperformed RF models and were applied to predict compounds’ irritation/corrosion potential from the Inactive Ingredient Database, which contains components present in FDA-approved drug products, and from the Cosmetic Ingredient Database, the European Commission source of information on cosmetic substances. All models built and validated in this study are publicly available at the STopTox web portal (https://stoptox.mml.unc.edu/). These models can be employed as reliable tools for identifying potential eye irritant/corrosive compounds

Revealing Adverse Outcome Pathways from Public High-Throughput Screening Data to Evaluate New Toxicants by a Knowledge-Based Deep Neural Network Approach

Traditional experimental testing to identify endocrine disruptors that enhance estrogenic signaling relies on expensive and labor-intensive experiments. We sought to design a knowledge-based deep neural network (k-DNN) approach to reveal and organize public high-throughput screening data for compounds with nuclear estrogen receptor α and β (ERα and ERβ) binding potentials. The target activity was rodent uterotrophic bioactivity driven by ERα/ERβ activations. After training, the resultant network successfully inferred critical relationships among ERα/ERβ target bioassays, shown as weights of 6521 edges between 1071 neurons. The resultant network uses an adverse outcome pathway (AOP) framework to mimic the signaling pathway initiated by ERα and identify compounds that mimic endogenous estrogens (i.e., estrogen mimetics). The k-DNN can predict estrogen mimetics by activating neurons representing several events in the ERα/ERβ signaling pathway. Therefore, this virtual pathway model, starting from a compound's chemistry initiating ERα activation and ending with rodent uterotrophic bioactivity, can efficiently and accurately prioritize new estrogen mimetics (AUC = 0.864-0.927). This k-DNN method is a potential universal computational toxicology strategy to utilize public high-throughput screening data to characterize hazards and prioritize potentially toxic compounds.

Mechanism-Driven Read-Across of Chemical Hepatotoxicants Based on Chemical Structures and Biological Data

Hepatotoxicity is a leading cause of attrition in the drug development process. Traditional preclinical and clinical studies to evaluate hepatotoxicity liabilities are expensive and time consuming. With the advent of critical advancements in high-throughput screening, there has been a rapid accumulation of in vitro toxicity data available to inform the risk assessment of new pharmaceuticals and chemicals. To this end, we curated and merged all available in vivo hepatotoxicity data obtained from the literature and public resources, which yielded a comprehensive database of 4089 compounds that includes hepatotoxicity classifications. After dividing the original database of chemicals into modeling and test sets, PubChem assay data were automatically extracted using an in-house data mining tool and clustered based on relationships between structural fragments and cellular responses in in vitro assays. The resultant PubChem assay clusters were further investigated. During the cross-validation procedure, the biological data obtained from several assay clusters exhibited high predictivity of hepatotoxicity and these assays were selected to evaluate the test set compounds. The read-across results indicated that if a new compound contained specific identified chemical fragments (ie, Molecular Initiating Event) and showed active responses in the relevant selected PubChem assays, there was potential for the chemical to be hepatotoxic in vivo. Furthermore, several mechanisms that might contribute to toxicity were derived from the modeling results including alterations in nuclear receptor signaling and inhibition of DNA repair. This modeling strategy can be further applied to the investigation of other complex chemical toxicity phenomena (eg, developmental and reproductive toxicities) as well as drug efficacy.

Predictive modeling of estrogen receptor agonism, antagonism, and binding activities using machine- and deep-learning approaches

As defined by the World Health Organization, an endocrine disruptor is an exogenous substance or mixture that alters function(s) of the endocrine system and consequently causes adverse health effects in an intact organism, its progeny, or (sub)populations. Traditional experimental testing regimens to identify toxicants that induce endocrine disruption can be expensive and time-consuming. Computational modeling has emerged as a promising and cost-effective alternative method for screening and prioritizing potentially endocrine-active compounds. The efficient identification of suitable chemical descriptors and machine-learning algorithms, including deep learning, is a considerable challenge for computational toxicology studies. Here, we sought to apply classic machine-learning algorithms and deep-learning approaches to a panel of over 7500 compounds tested against 18 Toxicity Forecaster assays related to nuclear estrogen receptor (ERα and ERβ) activity. Three binary fingerprints (Extended Connectivity FingerPrints, Functional Connectivity FingerPrints, and Molecular ACCess System) were used as chemical descriptors in this study. Each descriptor was combined with four machine-learning and two deep- learning (normal and multitask neural networks) approaches to construct models for all 18 ER assays. The resulting model performance was evaluated using the area under the receiver- operating curve (AUC) values obtained from a fivefold cross-validation procedure. The results showed that individual models have AUC values that range from 0.56 to 0.86. External validation was conducted using two additional sets of compounds (n = 592 and n = 966) with established interactions with nuclear ER demonstrated through experimentation. An agonist, antagonist, or binding score was determined for each compound by averaging its predicted probabilities in relevant assay models as an external validation, yielding AUC values ranging from 0.63 to 0.91. The results suggest that multitask neural networks offer advantages when modeling mechanistically related endpoints. Consensus predictions based on the average values of individual models remain the best modeling strategy for computational toxicity evaluations.

Development of In Vitro Corneal Models: Opportunity for Pharmacological Testing

The human eye is a specialized organ with a complex anatomy and physiology, because it is characterized by different cell types with specific physiological functions. Given the complexity of the eye, ocular tissues are finely organized and orchestrated. In the last few years, many in vitro models have been developed in order to meet the 3Rs principle (Replacement, Reduction and Refinement) for eye toxicity testing. This procedure is highly necessary to ensure that the risks associated with ophthalmic products meet appropriate safety criteria. In vitro preclinical testing is now a well-established practice of significant importance for evaluating the efficacy and safety of cosmetic, pharmaceutical, and nutraceutical products. Along with in vitro testing, also computational procedures, herein described, for evaluating the pharmacological profile of potential ocular drug candidates including their toxicity, are in rapid expansion. In this review, the ocular cell types and functionality are described, providing an overview about the scientific challenge for the development of three-dimensional (3D) in vitro models.

Predictive Modeling of Angiotensin I-Converting Enzyme Inhibitory Peptides Using Various Machine Learning Approaches

Food-derived angiotensin I-converting enzyme (ACE) inhibitory peptides could potentially be used as safe supportive therapeutic products for high blood pressure. Theoretical approaches are promising methods with the advantage through exploring the relationships between peptide structures and their bioactivities. In this study, peptides with ACE inhibitory activity were collected and curated. Quantitative structure-activity relationship (QSAR) models were developed by using the combination of various machine learning approaches and chemical descriptors. The resultant models have revealed several structure features accounting for the ACE inhibitions. 14 new dipeptides predicted to lower blood pressure by inhibiting ACE were selected. Molecular docking indicated that these dipeptides formed hydrogen bonds with ACE. Five of these dipeptides were synthesized for experimental testing. The QSAR models developed were proofed to design and propose novel ACE inhibitory peptides. Machine learning algorithms and properly selected chemical descriptors can be promising modeling approaches for rational design of natural functional food components.

Analysis of model PM2.5-induced inflammation and cytotoxicity by the combination of a virtual carbon nanoparticle library and computational modeling

Health risks induced by PM2.5 have become one of the major concerns among living populations, especially in regions facing serious pollution such as China and India. Furthermore, the composition of PM2.5 is complex and it also varies with time and locations. To facilitate our understanding of PM2.5-induced toxicity, a predictive modeling framework was developed in the present study. The core of this study was 1) to construct a virtual carbon nanoparticle library based on the experimental data to simulate the PM2.5 structures; 2) to quantify the nanoparticle structures by novel nanodescriptors; and 3) to perform computational modeling for critical toxicity endpoints. The virtual carbon nanoparticle library was developed to represent the nanostructures of 20 carbon nanoparticles, which were synthesized to simulate PM2.5 structures and tested for potential health risks. Based on the calculated nanodescriptors from virtual carbon nanoparticles, quantitative nanostructure-activity relationship (QNAR) models were developed to predict cytotoxicity and four different inflammatory responses induced by model PM2.5. The high predictability (R² > 0.65 for leave-one-out validations) of the resulted consensus models indicated that this approach could be a universal tool to predict and analyze the potential toxicity of model PM2.5, ultimately understanding and evaluating the ambient PM2.5-induced toxicity.

Toxicity Prediction Method Based on Multi-Channel Convolutional Neural Network

Molecular toxicity prediction is one of the key studies in drug design. In this paper, a deep learning network based on a two-dimension grid of molecules is proposed to predict toxicity. At first, the van der Waals force and hydrogen bond were calculated according to different descriptors of molecules, and multi-channel grids were generated, which could discover more detail and helpful molecular information for toxicity prediction. The generated grids were fed into a convolutional neural network to obtain the result. A Tox21 dataset was used for the evaluation. This dataset contains more than 12,000 molecules. It can be seen from the experiment that the proposed method performs better compared to other traditional deep learning and machine learning methods.

Big Data and Artificial Intelligence Modeling for Drug Discovery

Due to the massive data sets available for drug candidates, modern drug discovery has advanced to the big data era. Central to this shift is the development of artificial intelligence approaches to implementing innovative modeling based on the dynamic, heterogeneous, and large nature of drug data sets. As a result, recently developed artificial intelligence approaches such as deep learning and relevant modeling studies provide new solutions to efficacy and safety evaluations of drug candidates based on big data modeling and analysis. The resulting models provided deep insights into the continuum from chemical structure to in vitro, in vivo, and clinical outcomes. The relevant novel data mining, curation, and management techniques provided critical support to recent modeling studies. In summary, the new advancement of artificial intelligence in the big data era has paved the road to future rational drug development and optimization, which will have a significant impact on drug discovery procedures and, eventually, public health.

Using a hybrid read-across method to evaluate chemical toxicity based on chemical structure and biological data

Read-across has become a primary approach to fill data gaps for chemical safety assessments. Chemical similarity based on structure, reactivity, and physic-chemical property information is a traditional approach applied for read-across toxicity studies. However, toxicity mechanisms are usually complicated in a biological system, so only using chemical similarity to perform the read-across for new compounds was not satisfactory for most toxicity endpoints, especially when the chemically similar compounds show dissimilar toxicities. This study aims to develop an enhanced read-across method for chemical toxicity predictions. To this end, we used two large toxicity datasets for read-across purposes. One consists of 3979 compounds with Ames mutagenicity data, and the other contains 7332 compounds with rat acute oral toxicity data. First, biological data for all compounds in these two datasets were obtained by querying thousands of PubChem bioassays. The PubChem bioassays with at least five compounds from either of these two datasets showing active responses were selected to generate comprehensive bioprofiles. The read-across studies were performed by using chemical similarity search only and also by using a hybrid similarity search based on both chemical descriptors and bioprofiles. Compared to traditional read-across based on chemical similarity, the hybrid read-across approach showed improved accuracy of predictions for both Ames mutagenicity and acute oral toxicity. Furthermore, we could illustrate potential toxicity mechanisms by analyzing the bioprofiles used for this hybrid read-across study. The results of this study indicate that the new hybrid read-across approach could be an applicable computational tool for chemical toxicity predictions. In this way, the bottleneck of traditional read-across studies can be overcome by introducing public biological data into the traditional process. The incorporation of bioprofiles generated from the additional biological data for compounds can partially solve the "activity cliff" issue and reveal their potential toxicity mechanisms. This study leads to a promising direction to utilize data-driven approaches for computational toxicology studies in the big data era.

Evaluation and Identification of the Neuroprotective Compounds of Xiaoxuming Decoction by Machine Learning: A Novel Mode to Explore the Combination Rules in Traditional Chinese Medicine Prescription

Xiaoxuming decoction (XXMD), a classic traditional Chinese medicine (TCM) prescription, has been used as a therapeutic in the treatment of stroke in clinical practice for over 1200 years. However, the pharmacological mechanisms of XXMD have not yet been elucidated. The purpose of this study was to develop neuroprotective models for identifying neuroprotective compounds in XXMD against hypoxia-induced and H2O2-induced brain cell damage. In this study, a phenotype-based classification method was designed by machine learning to identify neuroprotective compounds and to clarify the compatibility of XXMD components. Four different single classifiers (AB, kNN, CT, and RF) and molecular fingerprint descriptors were used to construct stacked naïve Bayesian models. Among them, the RF algorithm had a better performance with an average MCC value of 0.725±0.014 and 0.774±0.042 from 5-fold cross-validation and test set, respectively. The probability values calculated by four models were then integrated into a stacked Bayesian model. In total, two optimal models, s-NB-1-LPFP6 and s-NB-2-LPFP6, were obtained. The two validated optimal models revealed Matthews correlation coefficients (MCC) of 0.968 and 0.993 for 5-fold cross-validation and of 0.874 and 0.959 for the test set, respectively. Furthermore, the two models were used for virtual screening experiments to identify neuroprotective compounds in XXMD. Ten representative compounds with potential therapeutic effects against the two phenotypes were selected for further cell-based assays. Among the selected compounds, two compounds significantly inhibited H2O2-induced and Na2S2O4-induced neurotoxicity simultaneously. Together, our findings suggested that machine learning algorithms such as combination Bayesian models were feasible to predict neuroprotective compounds and to preliminarily demonstrate the pharmacological mechanisms of TCM.

Nonanimal Models for Acute Toxicity Evaluations: Applying Data-Driven Profiling and Read-Across

BACKGROUND

Low-cost, high-throughput in vitro bioassays have potential as alternatives to animal models for toxicity testing. However, incorporating in vitro bioassays into chemical toxicity evaluations such as read-across requires significant data curation and analysis based on knowledge of relevant toxicity mechanisms, lowering the enthusiasm of using the massive amount of unstructured public data.

OBJECTIVE

We aimed to develop a computational method to automatically extract useful bioassay data from a public repository (i.e., PubChem) and assess its ability to predict animal toxicity using a novel bioprofile-based read-across approach.

METHODS

A training database containing 7,385 compounds with diverse rat acute oral toxicity data was searched against PubChem to establish in vitro bioprofiles. Using a novel subspace clustering algorithm, bioassay groups that may inform on relevant toxicity mechanisms underlying acute oral toxicity were identified. These bioassays groups were used to predict animal acute oral toxicity using read-across through a cross-validation process. Finally, an external test set of over 600 new compounds was used to validate the resulting model predictivity.

RESULTS

Several bioassay clusters showed high predictivity for acute oral toxicity (positive prediction rates range from 62-100%) through cross-validation. After incorporating individual clusters into an ensemble model, chemical toxicants in the external test set were evaluated for putative acute toxicity (positive prediction rate equal to 76%). Additionally, chemical fragment -in vitro-in vivo relationships were identified to illustrate new animal toxicity mechanisms.

CONCLUSIONS

The in vitro bioassay data-driven profiling strategy developed in this study meets the urgent needs of computational toxicology in the current big data era and can be extended to develop predictive models for other complex toxicity end points. https://doi.org/10.1289/EHP3614.

Multiple Machine Learning Comparisons of HIV Cell-based and Reverse Transcriptase Data Sets

The human immunodeficiency virus (HIV) causes over a million deaths every year and has a huge economic impact in many countries. The first class of drugs approved were nucleoside reverse transcriptase inhibitors. A newer generation of reverse transcriptase inhibitors have become susceptible to drug resistant strains of HIV, and hence, alternatives are urgently needed. We have recently pioneered the use of Bayesian machine learning to generate models with public data to identify new compounds for testing against different disease targets. The current study has used the NIAID ChemDB HIV, Opportunistic Infection and Tuberculosis Therapeutics Database for machine learning studies. We curated and cleaned data from HIV-1 wild-type cell-based and reverse transcriptase (RT) DNA polymerase inhibition assays. Compounds from this database with ≤1 μM HIV-1 RT DNA polymerase activity inhibition and cell-based HIV-1 inhibition are correlated (Pearson r = 0.44, n = 1137, p < 0.0001). Models were trained using multiple machine learning approaches (Bernoulli Naive Bayes, AdaBoost Decision Tree, Random Forest, support vector classification, k-Nearest Neighbors, and deep neural networks as well as consensus approaches) and then their predictive abilities were compared. Our comparison of different machine learning methods demonstrated that support vector classification, deep learning, and a consensus were generally comparable and not significantly different from each other using 5-fold cross validation and using 24 training and test set combinations. This study demonstrates findings in line with our previous studies for various targets that training and testing with multiple data sets does not demonstrate a significant difference between support vector machine and deep neural networks.

Predicting Nano-Bio Interactions by Integrating Nanoparticle Libraries and Quantitative Nanostructure Activity Relationship Modeling

The discovery of biocompatible or bioactive nanoparticles for medicinal applications is an expensive and time-consuming process that may be significantly facilitated by incorporating more rational approaches combining both experimental and computational methods. However, it is currently hindered by two limitations: (1) the lack of high-quality comprehensive data for computational modeling and (2) the lack of an effective modeling method for the complex nanomaterial structures. In this study, we tackled both issues by first synthesizing a large library of nanoparticles and obtained comprehensive data on their characterizations and bioactivities. Meanwhile, we virtually simulated each individual nanoparticle in this library by calculating their nanostructural characteristics and built models that correlate their nanostructure diversity to the corresponding biological activities. The resulting models were then used to predict and design nanoparticles with desired bioactivities. The experimental testing results of the designed nanoparticles were consistent with the model predictions. These findings demonstrate that rational design approaches combining high-quality nanoparticle libraries, big experimental data sets, and intelligent computational models can significantly reduce the efforts and costs of nanomaterial discovery.

From machine learning to deep learning: progress in machine intelligence for rational drug discovery

Machine intelligence, which is normally presented as artificial intelligence, refers to the intelligence exhibited by computers. In the history of rational drug discovery, various machine intelligence approaches have been applied to guide traditional experiments, which are expensive and time-consuming. Over the past several decades, machine-learning tools, such as quantitative structure-activity relationship (QSAR) modeling, were developed that can identify potential biological active molecules from millions of candidate compounds quickly and cheaply. However, when drug discovery moved into the era of 'big' data, machine learning approaches evolved into deep learning approaches, which are a more powerful and efficient way to deal with the massive amounts of data generated from modern drug discovery approaches. Here, we summarize the history of machine learning and provide insight into recently developed deep learning approaches and their applications in rational drug discovery. We suggest that this evolution of machine intelligence now provides a guide for early-stage drug design and discovery in the current big data era.

Experimental Errors in QSAR Modeling Sets: What We Can Do and What We Cannot Do

Numerous chemical data sets have become available for quantitative structure-activity relationship (QSAR) modeling studies. However, the quality of different data sources may be different based on the nature of experimental protocols. Therefore, potential experimental errors in the modeling sets may lead to the development of poor QSAR models and further affect the predictions of new compounds. In this study, we explored the relationship between the ratio of questionable data in the modeling sets, which was obtained by simulating experimental errors, and the QSAR modeling performance. To this end, we used eight data sets (four continuous endpoints and four categorical endpoints) that have been extensively curated both in-house and by our collaborators to create over 1800 various QSAR models. Each data set was duplicated to create several new modeling sets with different ratios of simulated experimental errors (i.e., randomizing the activities of part of the compounds) in the modeling process. A fivefold cross-validation process was used to evaluate the modeling performance, which deteriorates when the ratio of experimental errors increases. All of the resulting models were also used to predict external sets of new compounds, which were excluded at the beginning of the modeling process. The modeling results showed that the compounds with relatively large prediction errors in cross-validation processes are likely to be those with simulated experimental errors. However, after removing a certain number of compounds with large prediction errors in the cross-validation process, the external predictions of new compounds did not show improvement. Our conclusion is that the QSAR predictions, especially consensus predictions, can identify compounds with potential experimental errors. But removing those compounds by the cross-validation procedure is not a reasonable means to improve model predictivity due to overfitting.

Predictive Modeling of Estrogen Receptor Binding Agents Using Advanced Cheminformatics Tools and Massive Public Data

Estrogen receptors (ERα) are a critical target for drug design as well as a potential source of toxicity when activated unintentionally. Thus, evaluating potential ERα binding agents is critical in both drug discovery and chemical toxicity areas. Using computational tools, e.g., Quantitative Structure-Activity Relationship (QSAR) models, can predict potential ERα binding agents before chemical synthesis. The purpose of this project was to develop enhanced predictive models of ERα binding agents by utilizing advanced cheminformatics tools that can integrate publicly available bioassay data. The initial ERα binding agent data set, consisting of 446 binders and 8307 non-binders, was obtained from the Tox21 Challenge project organized by the NIH Chemical Genomics Center (NCGC). After removing the duplicates and inorganic compounds, this data set was used to create a training set (259 binders and 259 non-binders). This training set was used to develop QSAR models using chemical descriptors. The resulting models were then used to predict the binding activity of 264 external compounds, which were available to us after the models were developed. The cross-validation results of training set [Correct Classification Rate (CCR) = 0.72] were much higher than the external predictivity of the unknown compounds (CCR = 0.59). To improve the conventional QSAR models, all compounds in the training set were used to search PubChem and generate a profile of their biological responses across thousands of bioassays. The most important bioassays were prioritized to generate a similarity index that was used to calculate the biosimilarity score between each two compounds. The nearest neighbors for each compound within the set were then identified and its ERα binding potential was predicted by its nearest neighbors in the training set. The hybrid model performance (CCR = 0.94 for cross validation; CCR = 0.68 for external prediction) showed significant improvement over the original QSAR models, particularly for the activity cliffs that induce prediction errors. The results of this study indicate that the response profile of chemicals from public data provides useful information for modeling and evaluation purposes. The public big data resources should be considered along with chemical structure information when predicting new compounds, such as unknown ERα binding agents.

Supporting read-across using biological data

Read-across, i.e. filling toxicological data gaps by relating to similar chemicals, for which test data are available, is usually done based on chemical similarity. Besides structure and physico-chemical properties, however, biological similarity based on biological data adds extra strength to this process. In the context of developing Good Read-Across Practice guidance, a number of case studies were evaluated to demonstrate the use of biological data to enrich read-across. In the simplest case, chemically similar substances also show similar test results in relevant in vitro assays. This is a well-established method for the read-across of e.g. genotoxicity assays. Larger datasets of biological and toxicological properties of hundreds and thousands of substances become increasingly available enabling big data approaches in read-across studies. Several case studies using various big data sources are described in this paper. An example is given for the US EPA's ToxCast dataset allowing read-across for high quality uterotrophic assays for estrogenic endocrine disruption. Similarly, an example for REACH registration data enhancing read-across for acute toxicity studies is given. A different approach is taken using omics data to establish biological similarity: Examples are given for stem cell models in vitro and short-term repeated dose studies in rats in vivo to support read-across and category formation. These preliminary biological data-driven read-across studies highlight the road to the new generation of read-across approaches that can be applied in chemical safety assessment.

ADMET evaluation in drug discovery: 15. Accurate prediction of rat oral acute toxicity using relevance vector machine and consensus modeling

Background

Determination of acute toxicity, expressed as median lethal dose (LD₅₀), is one of the most important steps in drug discovery pipeline. Because in vivo assays for oral acute toxicity in mammals are time-consuming and costly, there is thus an urgent need to develop in silico prediction models of oral acute toxicity.

Results

In this study, based on a comprehensive data set containing 7314 diverse chemicals with rat oral LD₅₀ values, relevance vector machine (RVM) technique was employed to build the regression models for the prediction of oral acute toxicity in rate, which were compared with those built using other six machine learning approaches, including k-nearest-neighbor regression, random forest (RF), support vector machine, local approximate Gaussian process, multilayer perceptron ensemble, and eXtreme gradient boosting. A subset of the original molecular descriptors and structural fingerprints (PubChem or SubFP) was chosen by the Chi squared statistics. The prediction capabilities of individual QSAR models, measured by q_ext² for the test set containing 2376 molecules, ranged from 0.572 to 0.659.

Conclusion

Considering the overall prediction accuracy for the test set, RVM with Laplacian kernel and RF were recommended to build in silico models with better predictivity for rat oral acute toxicity. By combining the predictions from individual models, four consensus models were developed, yielding better prediction capabilities for the test set (q_ext² = 0.669–0.689). Finally, some essential descriptors and substructures relevant to oral acute toxicity were identified and analyzed, and they may be served as property or substructure alerts to avoid toxicity. We believe that the best consensus model with high prediction accuracy can be used as a reliable virtual screening tool to filter out compounds with high rat oral acute toxicity. Graphical abstract

Developing Enhanced Blood-Brain Barrier Permeability Models: Integrating External Bio-Assay Data in QSAR Modeling

PURPOSE

Experimental Blood-Brain Barrier (BBB) permeability models for drug molecules are expensive and time-consuming. As alternative methods, several traditional Quantitative Structure-Activity Relationship (QSAR) models have been developed previously. In this study, we aimed to improve the predictivity of traditional QSAR BBB permeability models by employing relevant public bio-assay data in the modeling process.

METHODS

We compiled a BBB permeability database consisting of 439 unique compounds from various resources. The database was split into a modeling set of 341 compounds and a validation set of 98 compounds. Consensus QSAR modeling workflow was employed on the modeling set to develop various QSAR models. A five-fold cross-validation approach was used to validate the developed models, and the resulting models were used to predict the external validation set compounds. Furthermore, we used previously published membrane transporter models to generate relevant transporter profiles for target compounds. The transporter profiles were used as additional biological descriptors to develop hybrid QSAR BBB models.

RESULTS

The consensus QSAR models have R(2) = 0.638 for five-fold cross-validation and R(2) = 0.504 for external validation. The consensus model developed by pooling chemical and transporter descriptors showed better predictivity (R(2) = 0.646 for five-fold cross-validation and R(2) = 0.526 for external validation). Moreover, several external bio-assays that correlate with BBB permeability were identified using our automatic profiling tool.

CONCLUSIONS

The BBB permeability models developed in this study can be useful for early evaluation of new compounds (e.g., new drug candidates). The combination of chemical and biological descriptors shows a promising direction to improve the current traditional QSAR models.

The identification of complex interactions in epidemiology and toxicology: a simulation study of boosted regression trees

BACKGROUND

There is a need to evaluate complex interaction effects on human health, such as those induced by mixtures of environmental contaminants. The usual approach is to formulate an additive statistical model and check for departures using product terms between the variables of interest. In this paper, we present an approach to search for interaction effects among several variables using boosted regression trees.

METHODS

We simulate a continuous outcome from real data on 27 environmental contaminants, some of which are correlated, and test the method's ability to uncover the simulated interactions. The simulated outcome contains one four-way interaction, one non-linear effect and one interaction between a continuous variable and a binary variable. Four scenarios reflecting different strengths of association are simulated. We illustrate the method using real data.

RESULTS

The method succeeded in identifying the true interactions in all scenarios except where the association was weakest. Some spurious interactions were also found, however. The method was also capable to identify interactions in the real data set.

CONCLUSIONS

We conclude that boosted regression trees can be used to uncover complex interaction effects in epidemiological studies.

Design, synthesis and experimental validation of novel potential chemopreventive agents using random forest and support vector machine binary classifiers

Compared to the current knowledge on cancer chemotherapeutic agents, only limited information is available on the ability of organic compounds, such as drugs and/or natural products, to prevent or delay the onset of cancer. In order to evaluate chemical chemopreventive potentials and design novel chemopreventive agents with low to no toxicity, we developed predictive computational models for chemopreventive agents in this study. First, we curated a database containing over 400 organic compounds with known chemoprevention activities. Based on this database, various random forest and support vector machine binary classifiers were developed. All of the resulting models were validated by cross validation procedures. Then, the validated models were applied to virtually screen a chemical library containing around 23,000 natural products and derivatives. We selected a list of 148 novel chemopreventive compounds based on the consensus prediction of all validated models. We further analyzed the predicted active compounds by their ease of organic synthesis. Finally, 18 compounds were synthesized and experimentally validated for their chemopreventive activity. The experimental validation results paralleled the cross validation results, demonstrating the utility of the developed models. The predictive models developed in this study can be applied to virtually screen other chemical libraries to identify novel lead compounds for the chemoprevention of cancers.

A meta-analysis of carbon nanotube pulmonary toxicity studies--how physical dimensions and impurities affect the toxicity of carbon nanotubes

This article presents a regression-tree-based meta-analysis of rodent pulmonary toxicity studies of uncoated, nonfunctionalized carbon nanotube (CNT) exposure. The resulting analysis provides quantitative estimates of the contribution of CNT attributes (impurities, physical dimensions, and aggregation) to pulmonary toxicity indicators in bronchoalveolar lavage fluid: neutrophil and macrophage count, and lactate dehydrogenase and total protein concentrations. The method employs classification and regression tree (CART) models, techniques that are relatively insensitive to data defects that impair other types of regression analysis: high dimensionality, nonlinearity, correlated variables, and significant quantities of missing values. Three types of analysis are presented: the RT, the random forest (RF), and a random-forest-based dose-response model. The RT shows the best single model supported by all the data and typically contains a small number of variables. The RF shows how much variance reduction is associated with every variable in the data set. The dose-response model is used to isolate the effects of CNT attributes from the CNT dose, showing the shift in the dose-response caused by the attribute across the measured range of CNT doses. It was found that the CNT attributes that contribute the most to pulmonary toxicity were metallic impurities (cobalt significantly increased observed toxicity, while other impurities had mixed effects), CNT length (negatively correlated with most toxicity indicators), CNT diameter (significantly positively associated with toxicity), and aggregate size (negatively correlated with cell damage indicators and positively correlated with immune response indicators). Increasing CNT N2 -BET-specific surface area decreased toxicity indicators.

Progress in computational toxicology

INTRODUCTION

Computational methods have been widely applied to toxicology across pharmaceutical, consumer product and environmental fields over the past decade. Progress in computational toxicology is now reviewed.

METHODS

A literature review was performed on computational models for hepatotoxicity (e.g. for drug-induced liver injury (DILI)), cardiotoxicity, renal toxicity and genotoxicity. In addition various publications have been highlighted that use machine learning methods. Several computational toxicology model datasets from past publications were used to compare Bayesian and Support Vector Machine (SVM) learning methods.

RESULTS

The increasing amounts of data for defined toxicology endpoints have enabled machine learning models that have been increasingly used for predictions. It is shown that across many different models Bayesian and SVM perform similarly based on cross validation data.

DISCUSSION

Considerable progress has been made in computational toxicology in a decade in both model development and availability of larger scale or 'big data' models. The future efforts in toxicology data generation will likely provide us with hundreds of thousands of compounds that are readily accessible for machine learning models. These models will cover relevant chemistry space for pharmaceutical, consumer product and environmental applications.

Exploring QSTR modeling and toxicophore mapping for identification of important molecular features contributing to the chemical toxicity in Escherichia coli

Biodiversity deprivation can affect functions and services of the ecosystem. Changes in biodiversity alter ecosystem processes and change the resilience of ecosystems to ecological changes. Bacterial communities are the main form of biomass in the ecosystem and one of largest populations on the planet. Bacterial communities provide important services to biodiversity. They break down pollutants, municipal waste and ingested food, and they are the primary route for recycling of organic matter to plants and other autotrophs, conversion of inorganic matter into new biological tissue using sunlight, management of energy crisis through use of biofuel. In the present study, computational chemistry and statistical modeling have been used to develop mathematical equations which can be applied to calculate toxicity of new/unknown chemicals/biofuels/metabolites in Escherichia coli. 2D and 3D descriptors were generated from molecular structure of compounds and mathematical models have been developed using genetic function approximation followed by multiple linear regression (GFA-MLR) method. Model validity was checked through defined internal (R(2)=0.751 and Q(2)=0.711), and external (Rpred(2)=0.773) statistical parameters. Molecular features responsible for toxicity were also assessed through 3D toxicophore study. The toxicophore-based model was validated (R=0.785) using qualitative statistical metrics and randomization test (Fischer validation).

Fusing dual-event data sets for Mycobacterium tuberculosis machine learning models and their evaluation

The search for new tuberculosis treatments continues as we need to find molecules that can act more quickly, be accommodated in multidrug regimens, and overcome ever increasing levels of drug resistance. Multiple large scale phenotypic high-throughput screens against Mycobacterium tuberculosis (Mtb) have generated dose response data, enabling the generation of machine learning models. These models also incorporated cytotoxicity data and were recently validated with a large external data set. A cheminformatics data-fusion approach followed by Bayesian machine learning, Support Vector Machine, or Recursive Partitioning model development (based on publicly available Mtb screening data) was used to compare individual data sets and subsequent combined models. A set of 1924 commercially available molecules with promising antitubercular activity (and lack of relative cytotoxicity to Vero cells) were used to evaluate the predictive nature of the models. We demonstrate that combining three data sets incorporating antitubercular and cytotoxicity data in Vero cells from our previous screens results in external validation receiver operator curve (ROC) of 0.83 (Bayesian or RP Forest). Models that do not have the highest 5-fold cross-validation ROC scores can outperform other models in a test set dependent manner. We demonstrate with predictions for a recently published set of Mtb leads from GlaxoSmithKline that no single machine learning model may be enough to identify compounds of interest. Data set fusion represents a further useful strategy for machine learning construction as illustrated with Mtb. Coverage of chemistry and Mtb target spaces may also be limiting factors for the whole-cell screening data generated to date.