Next Article in Journal
Residual Dynamics of Chlorantraniliprole and Fludioxonil in Soil and Their Effects on the Microbiome
Previous Article in Journal
Impact of Short-Chain Perfluoropropylene Oxide Acids on Biochemical and Behavioural Parameters in Eisenia fetida (Savigny, 1826)
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
JoX
Volume 15
Issue 1
10.3390/jox15010003
jox-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Simulation of the Long-Term Toxicity Towards Bobwhite Quail (Colinus virginianus) by the Monte Carlo Method

by
Nadia Iovine
,
Alla P. Toropova
*,
Andrey A. Toropov
,
Alessandra Roncaglioni
and
Emilio Benfenati
Department of Environmental, Health Science, Istituto di Ricerche Farmacologiche Mario Negri IRCCS, Via Mario Negri 2, 20156 Milano, Italy
*
Author to whom correspondence should be addressed.
J. Xenobiot. 2025, 15(1), 3; https://doi.org/10.3390/jox15010003
Submission received: 28 October 2024 / Revised: 18 December 2024 / Accepted: 23 December 2024 / Published: 26 December 2024
(This article belongs to the Section Ecotoxicology)
Browse Figure
Versions Notes

Abstract

:
In this study, models for NOEL (No Observed Effect Level) and NOEC (No Observed Effect Concentration) related to long-term/reproduction toxicity of various organic pesticides are built up, evaluated, and compared with similar models proposed in the literature. The data have been obtained from the EFSA OpenFoodTox database, collecting only data for the Bobwhite quail (Colinus virginianus). Models have been developed using the CORAL-2023 program, which can be used to develop quantitative structure–property/activity relationships (QSPRs/QSARs) and the Monte Carlo method for the optimization of the model. The software provided a model which may be considered useful for the practice. The determination coefficient of the best models for the external validation set was 0.665.

1. Introduction

Agricultural and industrial activities rely heavily on the widespread use of chemicals, which has resulted in significant environmental contamination [1,2]. The ongoing rise in chemical pollution underscores the need for comprehensive hazard and risk assessments to prevent adverse effects on various non-target organisms, which are easily exposed to pollutants such as pesticides. Avian species can be exposed to harmful chemicals through several routes: ingestion, inhalation and dermal contact [3]. These exposure pathways can lead to severe toxicological effects, impairing reproduction and survival, and potentially causing population declines that disrupt local ecosystems. Understanding the avian toxicity of various chemicals is critical for protecting bird species and maintaining healthy ecosystems. Ecological risk assessment plays a pivotal role in evaluating the potential side effects of these chemicals on non-target species, thereby helping to preserve ecological balance [4,5].
Birds are essential to both ecosystems and human society due to seed dispersers, and natural pest controllers, which contribute to biodiversity, food production, and agricultural sustainability. They help maintain ecosystem health by regulating insect populations and supporting plant reproduction. Additionally, birds serve as environmental indicators, reflecting changes in habitat quality and potential risks from toxins. Protecting birds and their habitats is crucial for ecological balance and human well-being [6,7].
The reproductive health of birds is especially vulnerable to long-term exposure to chemicals, as they can interfere with hormone systems, impair reproductive success, and cause long-term population declines [8,9,10,11]. In this context, evaluating the long-term toxicity (LTT) and reproductive impacts of pesticides is essential to protect bird species from adverse effects.
To assess the toxicity of chemicals in birds, the Organisation for Economic Co-operation and Development (OECD) has developed a series of standardized test guidelines (TGs), such as the TG 206 (1984) for Reproduction Toxicity, TG 223 (2016) for Acute Oral Toxicity, and TG 205 (1984) for Dietary Toxicity [12,13,14]. The assessment for avian reproduction toxicity (which involves LTT) is usually conducted using either the Northern Bobwhite quail (Colinus virginianus), mallard duck (Anas platyrhynchos) or Japanese quail (Coturnix japonica) as indicator species. Standardized endpoints, such as the No Observed Effect Level (NOEL) and No Observed Effect Concentration (NOEC), play a crucial role in risk assessment, providing key measures of chemical safety for non-target species. NOEL refers to the highest concentration of a substance at which no adverse effects are observed, while NOEC is the concentration at which no significant effects on a population are detected. These endpoints are valuable for assessing the LTT in non-target organisms such as birds and are widely used in ecological risk assessments to inform regulatory decisions [4,15].
The experimental definition of toxicity is expensive and time-consuming. Furthermore, laboratory animals are required, with the related ethical issues [5,16]. For these reasons, the use of in silico methods has been proposed in several studies and can be helpful in assessing a large number of endpoints [17,18,19,20,21,22]. In this paper, we described the data collection and the quantitative structure-activity relationship (QSAR) approach used to develop a model for the LTT of avian. The model is based on the correlation weights of molecular features used to calculate the descriptor within the CORAL software (http://www.insilico.eu/coral/, accessed 24 December 2024). The approach considered here has been applied to a wide variety of problems [23,24].

2. Materials and Methods

2.1. Data

Here, we considered the NOEL and NOEC for Bobwhite quail. The molecular structure is represented by the simplified molecular input line entry system (SMILES) [25]. This format is one of the most widely used, because it is compact, and uses symbols related to the atoms present in the molecule, indicating the bonds and ramifications. An example is given in Table 1. SMILES generated by VEGAHUB software were considered here (https://www.vegahub.eu/download, accessed 24 December 2024).
Experimental values were collected from the EFSA OpenFoodTox database version 5 (released in October 2023). Data for NOEC and NOEL (mg/kg bw/day) related to LTT or reproduction toxicity in Bobwhite quail were collected. The data collected were analysed to identify possible duplicates. If a compound had several data points, the most toxic one, in our case the one with the lowest value, was kept. Pruning spurious information led to a dataset of 139 experimental values for pesticides. These were converted to the logarithmic scale.
The set of all compounds was split into: (i) the active training set (≈25%), (ii) the passive training set (≈25%), (iii) the calibration set (≈25%), and (iv) the validation set (≈25%). Each set has a defined task. The active training set is used to build up the preliminary model: molecular features extracted from the SMILES of the substances in the active training set are involved in the Monte Carlo optimization using the CORAL software (http://www.insilico.eu/coral/, accessed 24 December 2024), which provides correlation weights for the above features. The CORAL software simply uses the SMILES of the molecules to build up the model, and the descriptors are parts of the SMILES. Conversely, most of the other in silico models have to generate molecular descriptors to be used as input for the model. Thus, CORAL software is easier, and as a result, it directly identifies the atoms and bonds in the molecule associated with the effect. The correlation weights defined by CORAL software give the maximal target function (TF), calculated using descriptors (the sum of the correlation weights) and endpoint on the active training set. The passive training set has to check whether the model for the active training set is satisfactory for SMILES that were not involved in the active training set. Since there are several parameters to be defined, the calibration set is used to detect the start of overtraining (overfitting). At the beginning of the optimization, the correlation coefficients between experimental values of the endpoint and the descriptor simultaneously increase for all sets. When the correlation coefficient for the calibration set reaches the maximum (this is the start of the overtraining), further attempts at optimization lead to a decrease in the correlation coefficient for the calibration set. Optimization should be stopped when overtraining starts. At this point, the model development is complete, and the model selected using the results of the calibration set should be considered the final one; the statistics on the calibration set should represent the values of the model selected. Once the Monte Carlo optimization procedure is complete, the validation set is used to assess the predictive potential of the model on substances which have not been used in the model development. Five random splits obtained as above are considered here.

2.2. Optimal Descriptors

Modelling based on optimal descriptors involves several levels. The first one is aimed to identify the list of structural parameters of the molecules of the training set. The second level addresses the threshold to establish the list of attributes not used by the model and the list of active SMILES attributes. The third level is used to identify the list of parameters that are apparent promoters of increase or apparent promoters of decrease of the endpoint.
The selected list of molecular features extracted from the database involves the so-called SMILES atoms, which are one symbol or a group of symbols that cannot be considered separately [26]. Fragments of local symmetry (FLS) are the second category of molecular features available from SMILES notations [26]. These are compositions in the forms of XYX, XYYX, and XYZYX, where X, Y, and Z are arbitrary SMILES atoms. Thus, they take into consideration a local, focused moiety of the molecule and not the global symmetry. FLSs use SMILES-atoms containing only one character.
The optimal descriptors (DCW) considered here are calculated as follows:
D C W ( T , N ) = ∑ C W ( S k ) + ∑ C W ( S S k ) + C W ( X Y X ) + C W ( X Y Y X ) + C W ( X Y Z Y X )
CW(x) are correlation weights for molecular features extracted from SMILES. Sk is a SMILES atom; SSk is a pair of SMILES atoms that are neighbors in the SMILES line. Table 1 sets out the definition of fragments of local symmetry in SMILES. For the FLS of the format XYX in our case there are four sequences, and two sequences appear for the format XYZYX, while the sequence XYYX is not represented in the SMILES. The T is the threshold related to the frequency of a SMILES attribute in the active training set. Here T = 1 is used. In other words, if a SMILES attribute occurs even once in the active training set, it is considered an active one, and conversely, if it is absent in the active training set, then it is not used. However, from a statistical point of view, it is preferable to use attributes that appear in the active training set as many times as possible. The N is the number of epochs of the Monte Carlo optimization. One epoch indicates the modification of correlation weights of all active SMILES attributes. Here N = 15 is used. Thus, below in this work, the DCW is defined as DCW (1, 15).

2.3. Optimization of Correlation Weights

Monte Carlo optimization is the process of the maximization for a target function, which is in fact a mathematical function of many parameters including the step of dividing the available data into four sets: active training, passive training, calibration, and validation sets. The task of the active training set is to identify suitable correlation weights for a model/correlation. The task of the passive training set is to verify whether the correlation weights are available for similar substances. The calibration set is aimed at double-checking the results of the active and passive training sets. Finally, the validation set aims to assess the model’s predictive potential using substances not used to build up the model. The optimization of correlation weights with the target function is carried out by gradual replacement of their initial values using the Monte Carlo algorithm.
The flow chart of the transformation of the correlation weight value is shown in Figure 1.
The optimization process applied here involves special components termed index of ideality of correlation (IIC) and correlation intensity index (CII) [26]. The IIC and CII should increase the “system’s attention” to that part of the training set more useful to identify general lessons, not only related to the training set but relevant to the calibration set. The influence of these factors can be regulated by using the weighting coefficients for the IIC and CII. The selection of these coefficients is carried out empirically, based on the results of preliminary observations of the stochastic optimization system, with different weights for the IIC and CII.
Thus, the target function applied here is calculated as follows:
T F = R A + R P − | R A − R P | × 0.1 + 0.3 × C I I + 0.5 × I I C
RA and RP are the determination coefficient values on the active and passive training sets, respectively.
Having the numerical data on the correlation weights, one can calculate the LTT towards quail with Equation (3):
L T T = C 0 + C 1 × D C W ( 1 , 15 )  
C0 and C1 are regression coefficients. The DCW(1,15) is the optimal descriptor calculated with Equation (1).
Table 2 contains an example of the calculation of the DCW(1,15) for the calculation weights of SMILES attributes obtained for split #1.

3. Results

Attempts to build a model without FLS correlation weights gave the results reported in Table 3. Unfortunately, these correlations are rather weak; for instance, the R2 of the validation set rarely reaches 0.5. However, the use of correlation weights of the FLS, as will be shown further on, improved the predictive potential of the model.
The statistical characteristics of models obtained with the correlation weights of FLS are better than those without these correlation weights. We replicated the modelling using five splits to give a more robust assessment of the results. The five models for the endpoint calculated with FLS are the following:
split-1: LTT = −1.949(± 0.019) + 0.1307(± 0.0023) * DCW(1,15)
split-2: LTT = −1.867(± 0.026) + 0.1894(± 0.0040) * DCW(1,15)
split-3: LTT = −1.587(± 0.021) + 0.0536(± 0.0059) * DCW(1,15)
split-4: LTT = −1.599(± 0.013) + 0.1637(± 0.0028) * DCW(1,15)
split-5: LTT = −2.013(± 0.020) + 0.1237(± 0.0050) * DCW(1,15)
Table 4 lists the statistical characteristics of the models calculated with Equations (4)–(8). R2 on the validation set are always higher than 0.5 and reaches 0.66 in the third split. Another useful statistical parameter, the mean absolute error (MAE), is in the range of 0.23–0.41. These statistical values are good enough to accept the approach used here, also considering the small number of substances used to build up the model. Furthermore, the endpoint is quite a complex one, since it is related to a number of mechanisms associated with the LTT. This endpoint is probably more difficult to be modelled compared with acute toxicity since it involves a larger set of toxicological processes related to prolonged exposure to the toxic substance. Finally, another element is that the model relates to pesticides, which contain several categories of substances with complex chemical structures.
In addition to the statistical characteristics of the models, a highly desirable addition is the mechanistic interpretation that allows to identify molecular features whose presence in the molecule can contribute to the increase in the magnitude of the considered endpoint, as well as structural features that can contribute to the decrease in the magnitude of the endpoint in question. Table 5 contains data on five Monte Carlo optimization runs of correlation weights of different molecular features extracted from SMILES (split #1, simulation using FLS). It is clear that if a correlation weight is consistently positive for all Monte Carlo optimization runs, then the corresponding molecular structure element should be considered as a promoter of increasing the endpoint under consideration. Conversely, if a correlation weight is consistently negative in a series of optimization runs, then the corresponding structure element should be considered as a promoter of decreasing the endpoint under consideration. One can see, that in addition to “traditional SMILES attributes” which are promoters of increase or decrease, the FLS are present in the lists too.
The applicability domain for these models was determined through the so-called statistical defects of SMILES. According to the above statistical defects, 11, 21, 18, 6, and 12 outliers are observed for splits #1–#5, correspondingly.
Table 6 compares the statistical quality of the models suggested here with others from the literature. Unfortunately, there are only a few works devoted to modeling toxicity to quails, so the collection in Table 6 is scarce. We underline that the endpoint that we address here is related to long-term exposure, and thus this endpoint is different from the endpoints addressed for avian toxicity in the other studies in the literature, related to acute toxicity effect. As we commented, the long-term exposure effects are more complex and difficult to evaluate, but they are of higher ecotoxicological relevance since this endpoint is more representative of the real situation. Since the calibration set represents the proper final set, the statistics on that set should be considered too. However, to be closer to the usual way of representing the results of the training set, we also provide the overall statistics of the calibration set, with the active and passive training sets.

4. Discussion

The approach to constructing QSPR/QSAR models considered here is based on the use of so-called optimal descriptors. The idea of optimal descriptors was initially related to the use of molecular graphs. More precisely, it was supposed to use, instead of numerical values of graph invariants, some coefficients that, when transformed into their sum, would give the maximum correlation coefficient with the endpoint under study. In fact, this project can be considered as an extension of the flexible descriptors proposed in [26].
However, if the mentioned work focused on modifying the diagonal of the adjacency matrix of a molecular graph (in other words chemical elements), the proposed extension [28] already concerned the off-diagonal elements of the adjacency matrix, which from a chemical point of view is a prototype of covalent bonds.
One of the important, although not the main, circumstances for developing models in general, and for QSPR/QSAR models in particular, is the convenience of constructing the models of interest. Despite the rapid growth of the possibilities of attracting memory resources and the speed of modern computers, the complexity of implementing modeling in conditions of representing molecules through graphs in the corresponding databases intended to serve as input for constructing models remains inconvenient. When developing models, it is much more convenient to use databases where molecules are presented in a more compact form, containing as much relevant information as possible, which can be sorted if desired and, if necessary, shortened.
There are several options for representing the molecular structure that to some extent satisfy the above-mentioned attractive and useful possibilities when applied to QSPR/QSAR analysis. Common molecular representations suitable for the above application are SMILES [26] and InChI [29]. The International Chemical Identifier notation system (InChI) is an alternative to the SMILES molecular representation system. The information content of INCHI is much greater than that of SMILES. However, extracting this information without special software is difficult for the common user.
In relation to InChI, it can be said that although the length of the string InChI is much greater than SMILES, nevertheless it is a compact representation capable of serving as a basis for developing and using databases capable of being input for QSPR/QSAR analysis. However, the investigation of information presented by InChI is much less convenient than SMILES, from the point of view of the human user. Like a SMILES notation, an InChI string is derived from a molecular structure representation. However, InChI is intended for computer use. It is typically derived from structure representations by software, whereas SMILES supports molecular communication between humans and computers [30].
Indeed, the CORAL software was used in the SMILES format quite often for a wide variety of physicochemical properties and biological activities such as toxicity of inorganic compounds [31], cellular activity induced by nanomaterials [32], chromatography retention indices of volatile organic compounds [33], binding affinity of endocrine disruptor [34], search for radiopharmaceutical agents [35], novel inhibitors against pancreatic cancer [36], influenza inhibitors [24], predicting the permeability of drugs [37], search for anti-prostate cancer agents [38], psychoactive drugs [39], simulation of drug-induced nephrotoxicity [40] whereas InChI did not find wide application for the development of optimal descriptors [29].
From a philosophical point of view, the concept of a “dialectical pair” is often used, implying a pair of concepts capable of directing thought in some rational constructive direction, in particular for the formation of useful hypotheses. Examples of dialectical pairs are essence and phenomenon; quality and quantity; space and time; cause and effect; necessity and chance; reality and possibility; matter and consciousness; and object and subject. Within the same perspective, optimal descriptors lead to the formulation of this dialectical pair: randomness and prediction.
It is obvious that the stochastic processes are related to randomness. However, due to the objective functions used in the Monte Carlo optimization processes, randomness becomes capable of creating a prediction based on the information provided by SMILES. In our experience, it is important to use the same algorithm to write the SMILES and the same should be used to generate the SMILES in the external verification.
As noted above, FLS proved to increase the predictivity of the models. We can only formulate a hypothesis regarding the reason for this. A possible explanation is that these fragments identify the fact that in a certain part of the molecule, there is a duality of circumstances, and the same phenomenon, e.g., the initial step generating the toxic effect, can occur in two locations, and this represents an increase in the probability that the event occurs. The identification of this case, offered by the FLS, improves the description of the molecular features on the basis of the adverse effect, probably.
The computer experiments conducted have shown that the FLS can help improve the predictive potential of models. Following the principle of “QSPR/QSAR are random event” [41], any conclusions related to QSPR/QSAR analysis must be confirmed based on the observation of not one, but several distributions of training and validation samples. Regarding the thesis that FLS can be useful as a tool for improving the predictive potential, it can be stated that this is the case for several (five) splits into a training set and a validation set.
While recognizing the shortcomings of the current version of fragments of local symmetry, it is also necessary to note that, in principle, more rigorous versions of the system of FLS can be formulated and developed. In particular, it is possible to specify FLS not only by symbols but also by chemical elements. On the other side, it is possible to remove from consideration such versions of local symmetry fragments that include obviously dubious symbols, such as brackets, numbers, and others (Table S1).

5. Conclusions

In this paper, we introduced QSAR models for long-term toxicity towards birds. The data has been obtained from the EFSA database OpenFoodTox, which represents an authoritative source. The CORAL software with Monte Carlo optimization has been used. In this way, optimal descriptors provided models characterized by satisfactory predictive potential for long-term toxicity towards quail. This is the first model, at the best of our knowledge, for this important kind of endpoint, since it is related to effects which may occur in real conditions, considering prolonged exposure. Other models, previously published, also by our laboratory, addressed acute toxicity towards birds, which is a particular case of effect, easier to be modelled, but less relevant. The predictive potential was estimated based on the results of constructing five models using significantly different partitions into training and validation sets. Structured training sets were used. They consisted of three groups of compounds, the so-called active training, passive training, and calibration sets. The observed average value of the determination coefficient on the validation set for five computer experiments of constructing toxicity models is 0.57±0.05. Semi-quantitative molecular features conveyed through the fragments of local symmetry tested here may be useful for developing models of the endpoint considered. The contribution of the index of ideality of correlation in the stochastic process of the Monte Carlo method optimization is quite valuable.
These models will be implemented in the VEGAHUB website (www.vegahub.eu), for free use.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/jox15010003/s1, Table S1 in Supplementary materials section contains the technical details of the considered models (Split #1).

Author Contributions

N.I.: conceptualization, resources; writing-original draft preparation; visualization; A.P.T.: conceptualization, methodology, resources; writing-original draft preparation; visualization; formal analysis; A.A.T.: conceptualization, methodology, resources; writing-original draft preparation; visualization; formal analysis; writing—review and editing; A.R.: writing—review and editing; funding acquisition; E.B.: project administration; writing-review and editing. All authors have read and agreed to the published version of the manuscript.

Funding

We acknowledge EFSA for the financial contribution within the project sOFT-ERA, OC/EFSA/IDATA/2022/02.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data is contained within the article or Supplementary Material.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Corrêa-Junior, D.; Parente, C.E.T.; Frases, S. Hazards associated with the combined application of fungicides and poultry litter in agricultural areas. J. Xenobiot. 2024, 14, 110–134. [Google Scholar] [CrossRef] [PubMed]
  2. Moreau, J.; Rabdeau, J.; Badenhausser, I.; Giraudeau, M.; Sepp, T.; Crépin, M.; Gaffard, A.; Bretagnolle, V.; Monceau, K. Pesticide impacts on avian species with special reference to farmland birds: A review. Environ. Monit. Assess. 2022, 194, 790. [Google Scholar] [CrossRef] [PubMed]
  3. Katagi, T.; Fujisawa, T. Acute toxicity and metabolism of pesticides in birds. J. Pestic. Sci. 2021, 46, 305–321. [Google Scholar] [CrossRef]
  4. Aagaard, A.; Berny, P.; Chaton, P.F.; Antia, A.L.; McVey, E.; Arena, M.; Fait, G.; Ippolito, A.; Linguadoca, A.; Sharp, R.; et al. Guidance on the risk assessment for birds and mammals. EFSA J. 2023, 21, 300. [Google Scholar] [CrossRef]
  5. Commission Regulation (EC) No 1907/2006 of the European Parliament and of the Council of 18 December 2006 concerning the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH), establishing a European Chemicals Agency, amending Directive 1999/45/EC and repealing Council Regulation (EEC) No 793/93 and Commission Regulation (EC) No 1488/94 as well as Council Directive 76/769/EEC and Commission Directives 91/155/EEC, 93/67/EEC, 93/105/EC and 2000/21/EC. Available online: http://data.europa.eu/eli/reg/2006/1907/2024-10-10 (accessed on 20 December 2024).
  6. Mariyappan, M.; Rajendran, M.; Velu, S.; Johnson, A.D.; Dinesh, G.K.; Solaimuthu, K.; Kaliyappan, M.; Sankar, M. Ecological Role and Ecosystem Services of Birds: A Review. Int. J. Environ. Clim. Chang. 2023, 13, 76–87. [Google Scholar] [CrossRef]
  7. Furness, R.W.; Greenwood, J.J.D. (Eds.) Birds as Monitors of Environmental Change; Springer: Dordrecht, The Netherlands, 1993. [Google Scholar] [CrossRef]
  8. Mohanty, B. Pesticides exposure and compromised fitness in wild birds: Focusing on the reproductive endocrine disruption. Pestic. Biochem. Physiol. 2024, 199, 105800. [Google Scholar] [CrossRef] [PubMed]
  9. Fry, D.M. Reproductive effects in birds exposed to pesticides and industrial chemicals. Environ. Health Perspect. 1995, 103 (Suppl. 7), 165–171. [Google Scholar] [CrossRef] [PubMed]
  10. Grace, J.; Duran, E.; Ann Ottinger, M.; Maness, T. Sublethal effects of early-life exposure to common and emerging contaminants in birds. Curr. Res. Toxicol. 2024, 7, 100190. [Google Scholar] [CrossRef]
  11. Ottinger, M.; Abdelnabi, M.; Henry, P.; Mcgary, S.; Thompson, N.; Wu, J. Neuroendocrine and Behavioral Implications of Endocrine Disrupting Chemicals in Quail. Horm. Behav. 2001, 40, 234–247. [Google Scholar] [CrossRef] [PubMed]
  12. OECD. Test No. 205: Avian Dietary Toxicity Test, OECD Guidelines for the Testing of Chemicals; Section 2; OECD Publishing: Paris, France, 1984. [Google Scholar] [CrossRef]
  13. OECD. Test No. 206: Avian Reproduction Test, OECD Guidelines for the Testing of Chemicals; Section 2; OECD Publishing: Paris, France, 1984. [Google Scholar] [CrossRef]
  14. OECD. Test No. 223: Avian Acute Oral Toxicity Test, OECD Guidelines for the Testing of Chemicals; Section 2; OECD Publishing: Paris, France, 1984. [Google Scholar] [CrossRef]
  15. Mineau, P. A review and analysis of study endpoints relevant to the assessment of “long term” pesticide toxicity in avian and mammalian wildlife. Ecotoxicology 2005, 14, 775–799. [Google Scholar] [CrossRef]
  16. Miccoli, A.; Marx-Stoelting, P.; Braeuning, A. The use of NAMs andomics data in risk assessment. EFSA J. 2022, 20, e200908. [Google Scholar] [CrossRef]
  17. Banjare, P.; Singh, J.; Roy, P.P. Predictive classification-based QSTR models for toxicity study of diverse pesticides on multiple avian species. Environ. Sci. Pollut. Res. 2021, 28, 17992–18003. [Google Scholar] [CrossRef] [PubMed]
  18. Zhang, C.; Cheng, F.; Sun, L.; Zhuang, S.; Li, W.; Liu, G.; Lee, P.W.; Tang, Y. In silico prediction of chemical toxicity on avian species using chemical category approaches. Chemosphere 2015, 122, 280–287. [Google Scholar] [CrossRef]
  19. Hengstler, J.G.; Foth, H.; Kahl, R.; Kramer, P.J.; Lilienblum, W.; Schulz, T.; Schweinfurth, H. The REACH concept and its impact on toxicological sciences. Toxicology 2006, 220, 232–239. [Google Scholar] [CrossRef] [PubMed]
  20. Mazzatorta, P.; Cronin, M.T.D.; Benfenati, E. A QSAR study of avian oral toxicity using support vector machines and genetic algorithms. QSAR Comb. Sci. 2006, 25, 616–628. [Google Scholar] [CrossRef]
  21. Kar, S.; Leszczynski, J. Is intraspecies QSTR model answer to toxicity data gap filling: Ecotoxicity modeling of chemicals to avian species. Sci. Total Environ. 2020, 738, 139858. [Google Scholar] [CrossRef]
  22. Basant, N.; Gupta, S.; Singh, K.P. Predicting Toxicities of Diverse Chemical Pesticides in Multiple Avian Species Using Tree-Based QSAR Approaches for Regulatory Purposes. J. Chem. Inf. Model. 2015, 55, 1337–1348. [Google Scholar] [CrossRef] [PubMed]
  23. Kumar, P.; Kumar, A.; Singh, D. CORAL: Development of a hybrid descriptor based QSTR model to predict the toxicity of dioxins and dioxin-like compounds with correlation intensity index and consensus modelling. Environ. Toxicol. Pharmacol. 2022, 93, 103893. [Google Scholar] [CrossRef]
  24. Azimi, A.; Ahmadi, S.; Kumar, A.; Qomi, M.; Almasirad, A. SMILES-Based QSAR and Molecular Docking Study of Oseltamivir Derivatives as Influenza Inhibitors. Polycycl. Aromat. Compd. 2023, 43, 3257–3277. [Google Scholar] [CrossRef]
  25. Weininger, D. SMILES, a Chemical Language and Information System: 1: Introduction to Methodology and Encoding Rules. J. Chem. Inf. Comput. Sci. 1988, 28, 31–36. [Google Scholar] [CrossRef]
  26. Toropov, A.A.; Toropova, A.P.; Roncaglioni, A.; Benfenati, E. In silico prediction of the mutagenicity of nitroaromatic compounds using correlation weights of fragments of local symmetry. Mutat. Res. Genet. Toxicol. Environ. Mutagen. 2023, 891, 503684. [Google Scholar] [CrossRef]
  27. Toropov, A.A.; Benfenati, E. QSAR models of quail dietary toxicity based on the graph of atomic orbitals. Bioorg. Med. Chem. Lett. 2006, 16, 1941–1943. [Google Scholar] [CrossRef] [PubMed]
  28. Randić, M.; Pompe, M.; Mills, D.; Basak, S.C. Variable connectivity index as a tool for modeling structure-property relationships. Molecules 2004, 9, 1177–1193. [Google Scholar] [CrossRef]
  29. Toropov, A.A.; Toropova, A.P.; Benfenati, E. QSAR-modeling of toxicity of organometallic compounds by means of the balance of correlations for InChI-based optimal descriptors. Mol. Divers. 2010, 14, 183–192. [Google Scholar] [CrossRef] [PubMed]
  30. Drefahl, A. CurlySMILES: A chemical language to customize and annotate encodings of molecular and nanodevice structures. J. Cheminfor. 2011, 3, 1. [Google Scholar] [CrossRef]
  31. Toropova, A.P.; Toropov, A.A.; Benfenati, E.; Gini, G. Co-evolutions of correlations for QSAR of toxicity of organometallic and inorganic substances: An unexpected good prediction based on a model that seems untrustworthy. Chemom. Intell. Lab. Syst. 2011, 105, 215–219. [Google Scholar] [CrossRef]
  32. Yuan, B.; Wang, Y.; Zong, C.; Sang, L.; Chen, S.; Liu, C.; Pan, Y.; Zhang, H. Modeling study for predicting altered cellular activity induced by nanomaterials based on Dlk1-Dio3 gene expression and structural relationships. Chemosphere 2023, 335, 139090. [Google Scholar] [CrossRef] [PubMed]
  33. Ahmadi, S.; Lotfi, S.; Hamzehali, H.; Kumar, P. A simple and reliable QSPR model for prediction of chromatography retention indices of volatile organic compounds in peppers. RSC Adv. 2024, 14, 3186–3201. [Google Scholar] [CrossRef] [PubMed]
  34. Goyal, S.; Rani, P.; Chahar, M.; Hussain, K.; Kumar, P.; Sindhu, J. Quantitative structure activity relationship studies of androgen receptor binding affinity of endocrine disruptor chemicals with index of ideality of correlation, their molecular docking, molecular dynamics and ADME studies. J. Biomol. Struct. Dyn. 2023, 41, 13616–13631. [Google Scholar] [CrossRef] [PubMed]
  35. Bamdi, F.; Shiri, F.; Ahmadi, S.; Salahinejad, M.; Bazzi-Allahri, F. Optimization of Monte Carlo Method-Based QSPR modeling for lipophilicity in radiopharmaceuticals. Chem. Phys. Lett. 2024, 843, 141239. [Google Scholar] [CrossRef]
  36. Ahmadi, S.; Lotfi, S.; Azimi, A.; Kumar, P. Multicellular target QSAR models for predicting of novel inhibitors against pancreatic cancer by Monte Carlo approach. Results Chem. 2024, 10, 101734. [Google Scholar] [CrossRef]
  37. Zivkovic, M.; Zlatanovic, M.; Zlatanovic, N.; Golubović, M.; Veselinović, A.M. A QSAR model for predicting the corneal permeability of drugs—The application of the Monte Carlo optimization method. New J. Chem. 2022, 47, 224–230. [Google Scholar] [CrossRef]
  38. Tajiani, F.; Ahmadi, S.; Lotfi, S.; Kumar, P.; Almasirad, A. In-silico activity prediction and docking studies of some flavonol derivatives as anti-prostate cancer agents based on Monte Carlo optimization. BMC Chem. 2023, 17, 87. [Google Scholar] [CrossRef]
  39. Veselinović, J.B.; Đorđević, V.; Bogdanović, M.; Morić, I.; Veselinović, A.M. QSAR modeling of dihydrofolate reductase inhibitors as a therapeutic target for multiresistant bacteria. Struct. Chem. 2018, 29, 541–551. [Google Scholar] [CrossRef]
  40. Toropov, A.A.; Barnes, D.A.; Toropova, A.P.; Roncaglioni, A.; Irvine, A.R.; Masereeuw, R.; Benfenati, E. CORAL models for drug-induced nephrotoxicity. Toxics 2023, 11, 293. [Google Scholar] [CrossRef] [PubMed]
  41. Toropova, A.P.; Toropov, A.A.; Benfenati, E.; Leszczynska, D.; Leszczynski, J. QSAR model as a random event: A case of rat toxicity. Bioorganic Med. Chem. 2015, 23, 1223–1230. [Google Scholar] [CrossRef]
Jox 15 00003 g001
Figure 1. The flow chart of the optimization of the correlation weight with target function (TF). D is some delta, i.e., the coefficient for modification of the correlation weights.
Figure 1. The flow chart of the optimization of the correlation weight with target function (TF). D is some delta, i.e., the coefficient for modification of the correlation weights.
Jox 15 00003 g001
Table 1. An example of formulation of fragments of local symmetry (FLS) for the SMILES of view O=C(O)c1nc(c(c(N)c1Cl)Cl)Cl.
Table 1. An example of formulation of fragments of local symmetry (FLS) for the SMILES of view O=C(O)c1nc(c(c(N)c1Cl)Cl)Cl.
Types of Fragments of Local SymmetryCodes for Calculation of Optimal Descriptor
FLS XYX
c(c;
(c(;
c(c;
(c(		[xyx4]
FLS XYYX
Absent [xyyx0]
FLS XYZYX
c(c(c;
(c(c(		[xyzyx2]
Table 2. An example of calculation of the DCW(1,15) that is equal to the sum of all correlation weights for pesticide represented by SMILES of view O=C(O)c1nc(c(c(N)c1Cl)Cl)Cl (the descriptor value is 16.10).
Table 2. An example of calculation of the DCW(1,15) that is equal to the sum of all correlation weights for pesticide represented by SMILES of view O=C(O)c1nc(c(c(N)c1Cl)Cl)Cl (the descriptor value is 16.10).
SMILES AttributeCorrelation Weight of SMILES Attribute, CW(x)NA *NPNC
(Sk)
O...........−0.4891333133
=...........−0.1378302534
P...........2.5439421
(...........−0.2469343335
O...........−0.4891333133
c...........−0.2147283030
1...........0.5435303232
c...........−0.2147283030
c...........−0.2147283030
c...........−0.2147283030
(...........−0.2469343335
c...........−0.2147283030
(...........−0.2469343335
c...........−0.2147283030
1...........0.5435303232
(...........−0.2469343335
C...........0.0241343335
(...........−0.2469343335
S...........0.7864121111
C...........0.0241343335
(...........−0.2469343335
(...........−0.2469343335
O...........−0.4891333133
C...........0.0241343335
C...........0.0241343335
(...........−0.2469343335
N...........0.7755252329
C...........0.0241343335
(...........−0.2469343335
C...........0.0241343335
(...........−0.2469343335
C...........0.0241343335
SSk
O...=.......0.0016292430
P...=.......8.2913110
P...(.......2.0343421
O...(.......0.2868302326
c...O.......0.7112788
c...1.......0.2000252630
c...1.......0.2000252630
c...c.......0.0593282929
c...c.......0.0593282929
c...(.......0.0243272827
c...(.......0.0243272827
c...(.......0.0243272827
c...(.......0.0243272827
c...1.......0.2000252630
1...(.......1.6836191911
C...(.......0.0663333235
C...(.......0.0663333235
S...(.......1.14591298
S...C.......0.8502558
C...(.......0.0663333235
(...(.......−0.5271221921
O...(.......0.2868302326
O...C.......0.1543212216
C...C.......0.7926191618
C...(.......0.0663333235
N...(.......−0.7782192122
N...C.......0.04559138
C...(.......0.0663333235
C...(.......0.0663333235
C...(.......0.0663333235
C...(.......0.0663333235
FLS
[xyx7]......−0.8997616
[xyyx0].....0.5900292924
[xyzyx1]....0.0854988
* NA, NP, and NC are frequencies of SMILES attribute in the active training, passive training, and calibration sets, respectively.
Table 3. The statistical characteristics of models of long-term toxicity towards Bobwhite quail built up without correlation weights of FLS.
Table 3. The statistical characteristics of models of long-term toxicity towards Bobwhite quail built up without correlation weights of FLS.
SplitSet *nR2CCCIICCIIQ2RMSEMAEF
1A350.48000.64860.65430.74270.42150.5930.48530
P330.72440.37480.85110.84100.69510.9180.85481
C350.54410.66990.73760.76730.45010.3360.26439
V350.3707----0.350.28-
2A340.44980.62050.59620.74260.39250.6560.53826
P350.52950.42750.26490.69780.35050.8800.78337
C350.54160.69500.73570.75180.50070.3460.28639
V340.4496----0.410.34-
3A350.28030.43790.50000.66050.17220.6400.52813
P340.62100.35630.78800.79370.56020.9650.86552
C350.49200.66690.70140.76430.43110.3990.29832
V340.3430----0.460.36-
4A350.72230.83870.71570.83210.69120.4510.38086
P350.49700.41800.64740.69850.44530.6810.60233
C340.34860.58690.59040.77810.24700.4100.32917
V340.5358----0.310.25-
5A340.53850.70000.57930.70170.47960.5020.42237
P350.48950.52210.59270.78730.44200.7790.72232
C350.42290.63560.65030.82860.35500.2670.23624
V340.4290----0.400.32-
* A = active training set; P = passive training set; C = calibration set; V = validation set; R2 = determination coefficient; CCC = concordance correlation coefficient; IIC = index of ideality of correlation; CII = correlation intensity index; Q2 = cross-validated R2; RMSE = root mean squared error; MAE = mean absolute error; F = Fischer F-ratio.
Table 4. Statistical characteristics of models of LTT towards Bobwhite quail built up with correlation weights of FLS.
Table 4. Statistical characteristics of models of LTT towards Bobwhite quail built up with correlation weights of FLS.
SplitSet *nR2CCCIICCIIQ2RMSEMAEF
1A350.62350.76810.74570.77080.58770.5040.40055
P330.78000.46140.88320.84800.75270.9040.831110
C350.61170.76520.78200.76580.55120.3430.27352
V350.5140----0.320.25-
2A340.64700.78570.80440.77490.60710.5250.41859
P350.71670.53230.32300.80040.65040.9380.81683
C350.66180.72330.81350.86140.62210.4120.31165
V340.5670----0.510.41-
3A350.14210.24880.28270.747500.6990.6155
P340.61750.29950.78580.80050.55220.8920.76352
C350.70520.79630.83970.83970.66150.2330.18779
V340.6650----0.280.23-
4A350.74920.85670.72890.82960.71990.4280.34099
P350.71000.59580.47450.78750.68180.6340.53481
C340.41070.63080.64080.84790.31300.4010.31522
V340.5208----0.400.31-
5A340.49450.66180.62510.72110.41130.5250.45531
P350.48400.47330.48780.78970.43370.8430.75231
C350.75000.84360.86580.86220.71970.1690.14299
V340.5819----0.360.27-
* A = active training set; P = passive training set; C = calibration set; V = validation set; R2 = determination coefficient; CCC = concordance correlation coefficient; IIC = index of ideality of correlation; CII = correlation intensity index; Q2 = cross-validated R2; RMSE = root mean squared error; MAE = mean absolute error; F = Fischer F-ratio.
Table 5. Mechanistic interpretation of the model for LTT towards Bobwhite quail (split #1, correlation weights of FLS involved to optimal descriptor calculation).
Table 5. Mechanistic interpretation of the model for LTT towards Bobwhite quail (split #1, correlation weights of FLS involved to optimal descriptor calculation).
S or SS or FLS 12345NANPNC
Promoters of increase
[xyyx0].....1.12300.95611.07161.37990.4717292924
c...2.......1.96431.23830.12981.68651.1871231818
C...C.......0.38081.23691.00981.27001.1194191618
S...(.......0.45060.80430.14940.88491.16811298
N...C.......1.30561.85321.88601.32481.22369138
c...O.......0.58931.21240.82370.01480.7202788
n...1.......0.22682.06720.49611.75610.7070643
3...(.......0.42250.88641.34801.53080.7303584
S...C.......0.76060.95141.95501.47940.5854558
P...(.......2.66662.71872.84743.36733.9175421
P...........3.86942.93932.32622.14542.8775421
[xyzyx2]....2.21271.07020.91462.03372.3060413
[xyx4]......0.52181.88971.97961.36291.2598343
n...3.......0.82200.75890.94891.07930.7237311
S...P.......3.30344.89464.46172.52904.1761211
Promoters of decrease
C...(.......−0.4395−0.6585−0.3236−0.2932−0.8184333235
1...........−0.6192−0.4606−0.0605−0.1050−0.4894303232
=...........−0.1914−0.9555−0.6864−0.8934−0.1373302534
O...=.......−0.5426−0.6028−0.0555−0.2830−1.3653292430
c...........−0.4842−0.8222−0.3887−0.4997−0.8995283030
2...(.......−1.1851−1.5421−1.2887−1.3368−1.1119152013
n...c.......−0.5251−1.2618−0.7232−0.4177−0.043412148
[xyzyx1]....−1.9008−0.4257−0.3202−0.6042−0.0287988
n...(.......−0.7403−0.6638−0.4427−0.8044−0.63687104
N...1.......−0.0541−0.8301−1.0608−0.0389−0.7986615
[xyx5]......−2.7243−3.1202−1.5869−1.2596−1.0190664
[xyx7]......−1.4561−0.7690−0.1741−2.0073−0.7584616
C...3.......−0.9602−0.5414−0.4110−0.4759−1.6288556
c...N.......−3.3089−2.0621−1.2194−2.4258−3.3495495
[xyx2]......−1.9693−1.3590−1.3893−1.3086−0.4644423
Table 6. Comparison of different models for toxicity (long-term in our case, acute in the others) toward Bobwhite quail.
Table 6. Comparison of different models for toxicity (long-term in our case, acute in the others) toward Bobwhite quail.
Number of Compounds in Training SetNumber of Compounds in Validation SetR2 for Training SetR2 for Validation Set Reference
41150.67-[21]
2580.88-[21]
115320.950.92[22]
91190.780.65[27]
104340.49 (0.70 calibration)0.67In this work
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Iovine, N.; Toropova, A.P.; Toropov, A.A.; Roncaglioni, A.; Benfenati, E. Simulation of the Long-Term Toxicity Towards Bobwhite Quail (Colinus virginianus) by the Monte Carlo Method. J. Xenobiot. 2025, 15, 3. https://doi.org/10.3390/jox15010003

AMA Style

Iovine N, Toropova AP, Toropov AA, Roncaglioni A, Benfenati E. Simulation of the Long-Term Toxicity Towards Bobwhite Quail (Colinus virginianus) by the Monte Carlo Method. Journal of Xenobiotics. 2025; 15(1):3. https://doi.org/10.3390/jox15010003

Chicago/Turabian Style

Iovine, Nadia, Alla P. Toropova, Andrey A. Toropov, Alessandra Roncaglioni, and Emilio Benfenati. 2025. "Simulation of the Long-Term Toxicity Towards Bobwhite Quail (Colinus virginianus) by the Monte Carlo Method" Journal of Xenobiotics 15, no. 1: 3. https://doi.org/10.3390/jox15010003

APA Style

Iovine, N., Toropova, A. P., Toropov, A. A., Roncaglioni, A., & Benfenati, E. (2025). Simulation of the Long-Term Toxicity Towards Bobwhite Quail (Colinus virginianus) by the Monte Carlo Method. Journal of Xenobiotics, 15(1), 3. https://doi.org/10.3390/jox15010003

Article Metrics

Back to TopTop