A Bayesian neural network for toxicity prediction

Quantum-to-Classical Neural Network Transfer Learning Applied to Drug Toxicity Prediction

Toxicity is a roadblock that prevents an inordinate number of drugs from being used in potentially life-saving applications. Deep learning provides a promising solution to finding ideal drug candidates; however, the vastness of chemical space coupled with the underlying O ( n 3 ) matrix multiplication means these efforts quickly become computationally demanding. To remedy this, we present a hybrid quantum-classical neural network for predicting drug toxicity utilizing a quantum circuit design that mimics classical neural behavior by explicitly calculating matrix products with complexity O ( n 2 ) . Leveraging the Hadamard test for efficient inner product estimation rather than the conventionally used swap test, we reduce the number of qubits by half and remove the need for quantum phase estimation. Directly computing matrix products quantum mechanically allows for learnable weights to be transferred from a quantum to a classical device for further training. We apply our framework to the Tox21 data set and show that it achieves commensurate predictive accuracy to the model's fully classical O ( n 3 ) analogue. Additionally, we demonstrate that the model continues to learn, without disruption, once transferred to a fully classical architecture. We believe that combining the quantum advantage of reduced complexity and the classical advantage of noise-free calculation will pave the way for more scalable machine learning models.

Large-scale spatiotemporal deep learning predicting urban residential indoor PM2.5 concentration

Indoor PM2.5 pollution is one of the leading causes of death and disease worldwide. As monitoring indoor PM2.5 concentrations on a large scale is challenging, it is urgent to assess population-level exposure and related health risks to develop an easy-to-use and generalized model to predict indoor PM2.5 concentrations and spatiotemporal variations at the global level. Existing machine learning models of indoor PM2.5 are prone to deliver single-point predictions, and their input strategies are not widely applicable. Here, we developed a Bayesian neural network (BNN) model for predicting the distribution of daily average urban residential PM2.5 concentration based on multiple data sources available from nationwide comprehensive sensor-monitoring records in China. The BNN model showed good performance with a 10-fold cross-validation R2 of 0.70, mean-absolute-error of 9.45 μg/m3, root-mean-square error of 13.3 μg/m3, and 95 % prediction interval coverage of 85 %. To demonstrate the application process, this model was applied to predict indoor PM2.5 concentrations on a large spatiotemporal scale. Our modeled population-weighted annual indoor PM2.5 concentration for China in 2019 was 22.8 μg/m3, far exceeding the WHO standard. The validity of the model at the population level can be further bolstered, making it valuable for assessing and managing indoor air pollution-related health risks.

The occurrence, tissue distribution, and PBT potential of per- and polyfluoroalkyl substances in the freshwater organisms from the Yangtze river via nontarget analysis

Numerous emerging per- and polyfluoroalkyl substances (PFASs) occur in the aquatic environment, posing a threat to aquatic ecosystems and human health. In this study, we conducted a nontarget analysis on 3 surface water samples and 92 tissue samples of 16 fish collected from the Yangtze River to investigate the patterns, tissue distribution, and environmental impacts of emerging PFASs. A total of 43 PFASs from 11 classes were identified, including 17 legacy PFASs and 26 emerging PFASs. Among the 43 PFASs, seven PFASs were reported in biota for the first time while five PFASs were reported in the environment for the first time. Chlorine substituted perfluoroalyl ether sulfonic acids were the major emerging PFASs detected in organisms. Our results showed that most emerging PFASs tended to accumulate in the liver whereas perfluorinated sulfonamides tended to accumulate in the blood, and all of the emerging PFASs accumulated less in the muscle. Methods for evaluating the persistence, bioaccumulation, and toxicity (PBT) of PFASs were developed by combining the in-silico methods and experimental methods. Long-chain PFASs were found to have extremely high PBT scores compared to short-chain PFASs. Additionally, most emerging PFASs exhibited comparable PBT characteristics with legacy PFASs, especially Cl-substituted PFASs.

Artificial neural networks in contemporary toxicology research

Artificial neural networks (ANNs) have a huge potential in toxicology research. They may be used to predict toxicity of various chemical compounds or classify the compounds based on their toxic effects. Today, numerous ANN models have been developed, some of which may be used to detect and possibly explain complex chemico-biological interactions. Fully connected multilayer perceptrons may in some circumstances have high classification accuracy and discriminatory power in separating damaged from intact cells after exposure to a toxic substance. Regularized and not fully connected convolutional neural networks can detect and identify discrete changes in patterns of two-dimensional data associated with toxicity. Bayesian neural networks with weight marginalization sometimes may have better prediction performance when compared to traditional approaches. With the further development of artificial intelligence, it is expected that ANNs will in the future become important parts of various accurate and affordable biosensors for detection of various toxic substances and evaluation of their biochemical properties. In this concise review article, we discuss the recent research focused on the scientific value of ANNs in evaluation and prediction of toxicity of chemical compounds.

Artificial intelligence in pharmacology research and practice

In recent years, the use of artificial intelligence (AI) in health care has risen steadily, including a wide range of applications in the field of pharmacology. AI is now used throughout the entire continuum of pharmacology research and clinical practice and from early drug discovery to real-world datamining. The types of AI models used range from unsupervised clustering of drugs or patients aimed at identifying potential drug compounds or suitable patient populations, to supervised machine learning approaches to improve therapeutic drug monitoring. Additionally, natural language processing is increasingly used to mine electronic health records to obtain real-world data. In this mini-review, we discuss the basics of AI followed by an outline of its application in pharmacology research and clinical practice.

A universal similarity based approach for predictive uncertainty quantification in materials science

Immense effort has been exerted in the materials informatics community towards enhancing the accuracy of machine learning (ML) models; however, the uncertainty quantification (UQ) of state-of-the-art algorithms also demands further development. Most prominent UQ methods are model-specific or are related to the ensembles of models; therefore, there is a need to develop a universal technique that can be readily applied to a single model from a diverse set of ML algorithms. In this study, we suggest a new UQ measure known as the Δ-metric to address this issue. The presented quantitative criterion was inspired by the k-nearest neighbor approach adopted for applicability domain estimation in chemoinformatics. It surpasses several UQ methods in accurately ranking the predictive errors and could be considered a low-cost option for a more advanced deep ensemble strategy. We also evaluated the performance of the presented UQ measure on various classes of materials, ML algorithms, and types of input features, thus demonstrating its universality.

The use of Bayesian methodology in the development and validation of a tiered assessment approach towards prediction of rat acute oral toxicity

There exists consensus that the traditional means by which safety of chemicals is assessed-namely through reliance upon apical outcomes obtained following in vivo testing-is increasingly unfit for purpose. Whilst efforts in development of suitable alternatives continue, few have achieved levels of robustness required for regulatory acceptance. An array of "new approach methodologies" (NAM) for determining toxic effect, spanning in vitro and in silico spheres, have by now emerged. It has been suggested, intuitively, that combining data obtained from across these sources might serve to enhance overall confidence in derived judgment. This concept may be formalised in the "tiered assessment" approach, whereby evidence gathered through a sequential NAM testing strategy is exploited so to infer the properties of a compound of interest. Our intention has been to provide an illustration of how such a scheme might be developed and applied within a practical setting-adopting for this purpose the endpoint of rat acute oral lethality. Bayesian statistical inference is drawn upon to enable quantification of degree of confidence that a substance might ultimately belong to one of five LD50-associated toxicity categories. Informing this is evidence acquired both from existing in silico and in vitro resources, alongside a purposely-constructed random forest model and structural alert set. Results indicate that the combination of in silico methodologies provides moderately conservative estimations of hazard, conducive for application in safety assessment, and for which levels of certainty are defined. Accordingly, scope for potential extension of approach to further toxicological endpoints is demonstrated.

Probabilistic modelling of developmental neurotoxicity based on a simplified adverse outcome pathway network

In a century where toxicology and chemical risk assessment are embracing alternative methods to animal testing, there is an opportunity to understand the causal factors of neurodevelopmental disorders such as learning and memory disabilities in children, as a foundation to predict adverse effects. New testing paradigms, along with the advances in probabilistic modelling, can help with the formulation of mechanistically-driven hypotheses on how exposure to environmental chemicals could potentially lead to developmental neurotoxicity (DNT). This investigation aimed to develop a Bayesian hierarchical model of a simplified AOP network for DNT. The model predicted the probability that a compound induces each of three selected common key events (CKEs) of the simplified AOP network and the adverse outcome (AO) of DNT, taking into account correlations and causal relations informed by the key event relationships (KERs). A dataset of 88 compounds representing pharmaceuticals, industrial chemicals and pesticides was compiled including physicochemical properties as well as in silico and in vitro information. The Bayesian model was able to predict DNT potential with an accuracy of 76%, classifying the compounds into low, medium or high probability classes. The modelling workflow achieved three further goals: it dealt with missing values; accommodated unbalanced and correlated data; and followed the structure of a directed acyclic graph (DAG) to simulate the simplified AOP network. Overall, the model demonstrated the utility of Bayesian hierarchical modelling for the development of quantitative AOP (qAOP) models and for informing the use of new approach methodologies (NAMs) in chemical risk assessment.

In Silico Models for Hepatotoxicity

In this chapter, we review the state of the art of predicting human hepatotoxicity using in silico techniques. There has been significant progress in this area over the past 20 years but there are still some challenges ahead. Principally, these challenges are our partial understanding of a very complex biochemical system and our ability to emulate that in a predictive capacity. Here, we provide an overview of the published modeling approaches in this area to date and discuss their design, strengths and weaknesses. It is interesting to note the diversity in modeling approaches, whether they be statistical algorithms or evidenced-based approaches including structural alerts and pharmacophore models. Irrespective of modeling approach, it appears a common theme of access to appropriate, relevant, and high-quality data is a limitation to all and is likely to continue to be the focus of future research.

DeepCarc: Deep Learning-Powered Carcinogenicity Prediction Using Model-Level Representation

Carcinogenicity testing plays an essential role in identifying carcinogens in environmental chemistry and drug development. However, it is a time-consuming and label-intensive process to evaluate the carcinogenic potency with conventional 2-years rodent animal studies. Thus, there is an urgent need for alternative approaches to providing reliable and robust assessments on carcinogenicity. In this study, we proposed a DeepCarc model to predict carcinogenicity for small molecules using deep learning-based model-level representations. The DeepCarc Model was developed using a data set of 692 compounds and evaluated on a test set containing 171 compounds in the National Center for Toxicological Research liver cancer database (NCTRlcdb). As a result, the proposed DeepCarc model yielded a Matthews correlation coefficient (MCC) of 0.432 for the test set, outperforming four advanced deep learning (DL) powered quantitative structure-activity relationship (QSAR) models with an average improvement rate of 37%. Furthermore, the DeepCarc model was also employed to screen the carcinogenicity potential of the compounds from both DrugBank and Tox21. Altogether, the proposed DeepCarc model could serve as an early detection tool (https://github.com/TingLi2016/DeepCarc) for carcinogenicity assessment.

Artificial Intelligence in Drug Safety and Metabolism

The use of artificial intelligence methods in drug safety began in the early 2000s with applications such as predicting bacterial mutagenicity and hERG inhibition. The field has been endlessly expanding ever since and the models have become more complex. These approaches are now integrated into molecule risk assessment processes along with in vitro and in vivo methods. Today, artificial intelligence can be used in every phase of drug discovery and development, from profiling chemical libraries in early discovery, to predicting off-target effects in the mid-discovery phase, to assessing potential mutagenic impurities in development and degradants as part of life cycle management. This chapter provides an overview of artificial intelligence in drug safety and describes its application throughout the entire discovery and development process.

Predicting Antituberculosis Drug-Induced Liver Injury Using an Interpretable Machine Learning Method: Model Development and Validation Study

Background

Tuberculosis (TB) is a pandemic, being one of the top 10 causes of death and the main cause of death from a single source of infection. Drug-induced liver injury (DILI) is the most common and serious side effect during the treatment of TB.

Objective

We aim to predict the status of liver injury in patients with TB at the clinical treatment stage.

Methods

We designed an interpretable prediction model based on the XGBoost algorithm and identified the most robust and meaningful predictors of the risk of TB-DILI on the basis of clinical data extracted from the Hospital Information System of Shenzhen Nanshan Center for Chronic Disease Control from 2014 to 2019.

Results

In total, 757 patients were included, and 287 (38%) had developed TB-DILI. Based on values of relative importance and area under the receiver operating characteristic curve, machine learning tools selected patients’ most recent alanine transaminase levels, average rate of change of patients’ last 2 measures of alanine transaminase levels, cumulative dose of pyrazinamide, and cumulative dose of ethambutol as the best predictors for assessing the risk of TB-DILI. In the validation data set, the model had a precision of 90%, recall of 74%, classification accuracy of 76%, and balanced error rate of 77% in predicting cases of TB-DILI. The area under the receiver operating characteristic curve score upon 10-fold cross-validation was 0.912 (95% CI 0.890-0.935). In addition, the model provided warnings of high risk for patients in advance of DILI onset for a median of 15 (IQR 7.3-27.5) days.

Conclusions

Our model shows high accuracy and interpretability in predicting cases of TB-DILI, which can provide useful information to clinicians to adjust the medication regimen and avoid more serious liver injury in patients.

Deep Learning-Based Conformal Prediction of Toxicity

Predictive modeling for toxicity can help reduce risks in a range of applications and potentially serve as the basis for regulatory decisions. However, the utility of these predictions can be limited if the associated uncertainty is not adequately quantified. With recent studies showing great promise for deep learning-based models also for toxicity predictions, we investigate the combination of deep learning-based predictors with the conformal prediction framework to generate highly predictive models with well-defined uncertainties. We use a range of deep feedforward neural networks and graph neural networks in a conformal prediction setting and evaluate their performance on data from the Tox21 challenge. We also compare the results from the conformal predictors to those of the underlying machine learning models. The results indicate that highly predictive models can be obtained that result in very efficient conformal predictors even at high confidence levels. Taken together, our results highlight the utility of conformal predictors as a convenient way to deliver toxicity predictions with confidence, adding both statistical guarantees on the model performance as well as better predictions of the minority class compared to the underlying models.

Improving drug safety predictions by reducing poor analytical practices

Predicting the safety of a drug from preclinical data is a major challenge in drug discovery, and progressing an unsafe compound into the clinic puts patients at risk and wastes resources. In drug safety pharmacology and related fields, methods and analytical decisions known to provide poor predictions are common and include creating arbitrary thresholds, binning continuous values, giving all assays equal weight, and multiple reuse of information. In addition, the metrics used to evaluate models often omit important criteria and models’ performance on new data are often not assessed rigorously. Prediction models with these problems are unlikely to perform well, and published models suffer from many of these issues. We describe these problems in detail, demonstrate their negative consequences, and propose simple solutions that are standard in other disciplines where predictive modelling is used.

Uncertainty quantification in drug design

Machine learning and artificial intelligence are increasingly being applied to the drug-design process as a result of the development of novel algorithms, growing access, the falling cost of computation and the development of novel technologies for generating chemically and biologically relevant data. There has been recent progress in fields such as molecular de novo generation, synthetic route prediction and, to some extent, property predictions. Despite this, most research in these fields has focused on improving the accuracy of the technologies, rather than on quantifying the uncertainty in the predictions. Uncertainty quantification will become a key component in autonomous decision making and will be crucial for integrating machine learning and chemistry automation to create an autonomous design-make-test-analyse cycle. This review covers the empirical, frequentist and Bayesian approaches to uncertainty quantification, and outlines how they can be used for drug design. We also outline the impact of uncertainty quantification on decision making.