Uncertainty reduction in environmental data with conflicting information

In Silico Prediction of Skin Sensitization for Compounds via Flexible Evidence Combination Based on Machine Learning and Dempster-Shafer Theory

Skin sensitization is increasingly becoming a significant concern in the development of drugs and cosmetics due to consumer safety and occupational health problems. In silico methods have emerged as alternatives to traditional in vivo animal testing due to ethical and economic considerations. In this study, machine learning methods were used to build quantitative structure-activity relationship (QSAR) models on five skin sensitization data sets (GPMT, LLNA, DPRA, KeratinoSens, and h-CLAT), achieving effective predictive accuracies (correct classification rates of 0.688-0.764 on test sets). To address the complex mechanisms of human skin sensitization, the Dempster-Shafer theory was applied to merge multiple QSAR models, resulting in an evidence-based integrated decision model. Various evidence combinations and combination rules were explored, with the self-defined Q3 rule showing superior balance. The combination of evidence such as GPMT and KeratinoSens and h-CLAT achieved a correct classification rate (CCR) of 0.880 and coverage of 0.893 while maintaining the competitiveness of other combinations. Additionally, the Shapley additive explanations (SHAP) method was used to interpret important features and substructures related to skin sensitization. A comparative analysis of an external human test set demonstrated the superior performance of the proposed method. Finally, to enhance accessibility, the workflow was implemented into a user-friendly software named HSkinSensDS.

Prioritization of in silico models and molecular descriptors for the assessment of ready biodegradability

Ready biodegradability is a key property for evaluating the long-term effects of chemicals on the environment and human health. As such, it is used as a screening test for the assessment of persistent, bioaccumulative and toxic substances. Regulators encourage the use of non-testing methods, such as in silico models, to save money and time. A dataset of 757 chemicals was collected to assess the performance of four freely available in silico models that predict ready biodegradability. They were applied to develop a new consensus method that prioritizes the use of each individual model according to its performance on chemical subsets driven by the presence or absence of different molecular descriptors. This consensus method was capable of almost eliminating unpredictable chemicals, while the performance of combined models was substantially improved with respect to that of the individual models.

In silico analysis of nanomaterials hazard and risk

Because a variety of human-related activities, engineer-ed nanoparticles (ENMs) may be released to various environmental media and may cross environmental boundaries, and thus will be found in most media. Therefore, the potential environmental impacts of ENMs must be assessed from a multimedia perspective and with an integrated risk management approach that considers rapid developments and increasing use of new nanomaterials. Accordingly, this Account presents a rational process for the integration of in silico ENM toxicity and fate and transport analyses for environmental impact assessment. This approach requires knowledge of ENM toxicity and environmental exposure concentrations. Considering the large number of current different types of ENMs and that those numbers are likely to increase, there is an urgent need to accelerate the evaluation of their toxicity and the assessment of their potential distribution in the environment. Developments in high throughput screening (HTS) are now enabling the rapid generation of large data sets for ENM toxicity assessment. However, these analyses require the establishment of reliable toxicity metrics, especially when HTS includes data from multiple assays, cell lines, or organisms. Establishing toxicity metrics with HTS data requires advanced data processing techniques in order to clearly identify significant biological effects associated with exposure to ENMs. HTS data can form the basis for developing and validating in silico toxicity models (e.g., quantitative structure-activity relationships) and for generating data-driven hypotheses to aid in establishing and/or validating possible toxicity mechanisms. To correlate the toxicity of ENMs with their physicochemical properties, researchers will need to develop quantitative structure-activity relationships for nanomaterials (i.e., nano-SARs). However, as nano-SARs are applied in regulatory applications, researchers must consider their applicability and the acceptance level of false positive relative to false negative predictions and the reliability of toxicity data. To establish the environmental impact of ENMs identified as toxic, researchers will need to estimate the potential level of environmental exposure concentration of ENMs in the various media such as air, water, soil, and vegetation. When environmental monitoring data are not available, models of ENMs fate and transport (at various levels of complexity) serve as alternative approaches for estimating exposure concentrations. Risk management decisions regarding the manufacturing, use, and environmental regulations of ENMs would clearly benefit from both the assessment of potential ENMs exposure concentrations and suitable toxicity metrics. The decision process should consider the totality of available information: quantitative and qualitative data and the analysis of nanomaterials toxicity, and fate and transport behavior in the environment. Effective decision-making to address the potential impacts of nanomaterials will require considerations of the relevant environmental, ecological, technological, economic, and sociopolitical factors affecting the complete lifecycle of nanomaterials, while accounting for data and modeling uncertainties. Accordingly, researchers will need to establish standardized data management and analysis tools through nanoinformatics as a basis for the development of rational decision tools.

Quantitative consensus of bioaccumulation models for integrated testing strategies

A quantitative consensus model based on bioconcentration factor (BCF) predictions obtained from five quantitative structure-activity relationship models was developed for bioaccumulation assessment as an integrated testing approach for waiving. Three categories were considered: non-bioaccumulative, bioaccumulative and very bioaccumulative. Five in silico BCF models were selected and included into a quantitative consensus model by means of the continuous formulation of Bayes' theorem. The discrete likelihoods commonly used in the qualitative Bayesian model were substituted by probability density functions to reduce the loss of information that occurred when continuous BCF values were distributed across the three bioaccumulation categories. Results showed that the continuous Bayesian model yielded the best classification predictions compared not only to the discrete Bayesian model, but also to the individual BCF models. The proposed quantitative consensus model proved to be a suitable approach for integrated testing strategies for continuous endpoints of environmental interest.

Quantitative read-across for predicting the acute fish toxicity of organic compounds

Read-across enables the interpolation of a property for a target chemical from respective experimental data of sufficiently similar compounds. Employing a set of 692 organic compounds with experimental values for the 96 h fish toxicity toward the fathead minnow in terms of LC(50) (lethal concentration 50%) values, a read-across method has been developed that is based on atom-centered fragments (ACFs) for evaluating chemical similarity. Prediction of log LC(50) proceeds through reading across the toxicity enhancement over predicted narcosis-level toxicity in terms of the respective logarithmic ratio, log T(e), and adding the respective baseline narcosis LC(50) estimated from log K(ow) (octanol/water partition coefficient). Depending on the minimum similarity imposed on a compound to serve as read-across basis for the target chemical, three different standard settings have been introduced, allowing one to perform screening-level estimations as well as predictions with intermediate and good confidence. The respective squared correlation coefficients (r(2)) are 0.73, 0.78, and 0.87, with root-mean square errors (rms) of 0.73, 0.60, and 0.39 log units, respectively. As a general trend, increasing the ACF minimum similarity increases the prediction quality at the cost of decreasing the application range. The method has the potential to assist in the predictive evaluation of fish toxicity for regulatory purposes such as under the REACH legislation.

Data Quality Assessment for In Silico Methods: A Survey of Approaches and Needs

As indicated in Chapter 3, there are a large number of potential sources of data now available for modelling purposes. These range from historical literature references for a few compounds to highly curated databases of hundreds of thousands of compounds, available via the internet. Before including any data in an in silico model, the question of data quality must be addressed. Although it is difficult to define the quality of data in absolute terms, it is possible to assess the suitability of data for a given purpose. There are many reasons for variability within data and the degree of error that is acceptable for one model may not be the same as for another. For example generating a global model intended to pre-screen large numbers of compounds does not require the same degree of accuracy as performing an individual risk assessment for a chemical of interest. In this chapter, sources of data variability and error will be discussed and formal methods to score data quality, such as use of the Klimisch criteria, will be described. Examples of data quality issues will be given for specific endpoints relating to both environmental and human health effects. Mathematical approaches (Dempster-Schafer theory and Bayesian networks) demonstrating how this information relating to confidence in the data can be incorporated into in silico models is also discussed.