Methods for predictive testing of agents to assess induction and inhibition of adverse outcomes and / or disease

Abstract

The invention is a method for predictive testing of materials, allowing to save time on the observation of adverse outcomes such as chronic inflammation, known to lead to the development of most of the slowly progressing diseases, e.g. cancer or autoimmune diseases. The invention uses the observation of the evolution of in vitro models to predict outcomes based on early molecular events that occur first, long before they are shown as long-term effects in tissues. These molecular events are analyzed, measured and / or detected for up to one week, and an in silico time propagation is performed, correlating the early events with the long-term effects, leading to a reliable prediction of the safety / toxicity of the material. The in silico prediction of the observed values ​​in time from their initial values ​​is performed using a system of differential equations constructed from the interaction terms of the interaction matrix (Equation (I)), where scenarios (Equation (II)) are used to select the relevant observed values ​​that match the observed values ​​observed within each in vitro experiment. JPEG2025510982000050.jpg21126

Description

[Technical field]

[0001] The present invention is in the field of investigating or analyzing materials, inter alia, by determining their biological effects due to their chemical or physical or biological properties. The invention is also in the field of safety testing by determining biological events, and further calculating and computing their effects over time. The present invention relates to methods for predictive testing of agents, such as toxins, materials, chemicals, medicines and vaccines, to assess adverse outcomes and / or disease induction and inhibition. [Background technology]

[0002] [Background of the invention and technical problems] Substances (agents), such as materials and particulate matter, compounds, drugs or drug candidates, and vaccines, must be tested to detect possible toxicity, adverse outcomes or similar undesirable scenarios, among which slowly developing adverse outcomes are particularly problematic. Currently, animal testing is the only method that can detect long-term (i.e. chronic) health complications of substances that do not show either immediate toxicity or strong acute inflammatory reactions. That is, animals are exposed and monitored for several months to detect the long-term effects of the substance-material or compound. The tests may be research-driven, focusing on the development of knowledge about the basic mechanisms of the organism, or may be applied, for example, in disease treatment, reproduction, defense studies and toxicology testing, to answer several questions of great practical importance, such as material, drug and vaccine safety and cosmetic testing. The annual use of vertebrates, from zebrafish to non-human primates, is estimated to exceed 100 million. Mice, rats, fish, amphibians and reptiles together account for 85% of research animals.

[0003] However, animal testing has several problems. Because it focuses on detecting end points (adverse outcomes), it is slow, therefore expensive, often unethical, and even inappropriate at times due to its inability to accurately reflect outcomes in humans. For example, Greek and Menache (2013, doi:10.7150 / ijms.5529) pointed out that although approximately 100 vaccines have been shown to prevent HIV in animals, none of them have been effective in humans. On the contrary, Braken (2009, doi:10.1258 / jrsm.2008.08k033) showed that thalidomide causes severe birth defects in humans but not in animals. Nevertheless, the main obstacle in the widespread application of material and chemical safety testing is undoubtedly the long time it takes to reach long-term or chronic disease end points. The required timescales of several months make such testing entirely useless in the case of materials and chemical safety testing, where decisions are needed much earlier in the development process.

[0004] Thus, there is a clear need for animal substitutes, i.e. replacement of animal testing. Public organizations, such as the European Centre for the Validation of Alternative Testing Methods (ECVAM) of the European Commission (EC), the US Coordinating Committee for the Evaluation of Alternative Methods, ZEBET in Germany, and the Japan Centre for the Validation of Alternative Methods, promote the elimination of animal testing and answer regulatory requirements.

[0005] These needs are the driving force behind the development of various testing platforms, all upgraded cell culture systems. The most logical approach seems to be the development of structural copies of the tissues on which the tests are performed. Here, the rationale is the structural similarity between the original tissue and the copies, such as organoids (organ-like testing devices), "organ chips" (patent applications US2020408744, WO2020172670 and WO2009146911), "lung chips" (patent applications GB2585150 and WO2018217882), "gut-on-a-chip" (patent applications US2020224136, US2020283732), "skin chips" (patent application US2020239857). The systems contain cells in two- or three-dimensional systems that mimic organs. These chips can be used to replace animals in in vitro disease research, drug testing, and toxicity testing. Typically, such chips are intended for use in gene expression studies (patent applications US2020224136, WO2020172670), sometimes combined with cell morphology screening and determination of specific enzyme activity or metabolic events (patent applications WO2020172670, JPH04148695, EP2154241, WO2007120699). For example, the Wyss Institute (https: / / wyss.harvard.edu / technology / using-systems-biology-to-find-and-test-new-drugs-faster / ) reports on a novel algorithm used for drug discovery, where the algorithm is arranged to sift through gene expression pattern data of tens of thousands of known drug compounds and identify those that may revert disease state expression patterns to normal. These candidate compounds should then be further tested in cell-based assays to assess in vitro activity. In the final step, successful compounds are tested in both cell-based assays and human organ-chips to demonstrate that they can effectively treat the disease without harmful side effects before moving into clinical trials.Although a good alternative, the time to test for possible adverse outcomes on such organ-on-a-chip systems remains similar to in vivo animal studies (e.g., patent application WO2020172670 describes methods for testing kidney-on-a-chips that last more than a week, with some methods requiring four weeks of exposure), meaning that such studies remain costly and time-consuming.

[0006] Another option is in silico or computer simulation and mathematical modeling, which seeks to investigate and ultimately predict toxicity and drug effects in humans without the use of animals, tissues or cells (Watts and Geoff, 2007, doi:10.1136 / bmj.39058.469491.68; Edelman et al, 2010, doi:10.1002 / wsbm.75. PMC 3157287. PMID 20836040). This is achieved by investigating test compounds at a molecular level, using recent advances in computational power, with the ultimate aim of creating treatments specific to each patient. The latest developments in in silico testing methods are given in a recent report (Mark TD et al. Computational Toxicology 21, 2022, https: / / doi.org / 10.1016 / j.comtox.2022.100213), where the prediction of hazards and exposure relies on QSAR and read-across approaches. The main problem with this approach is the lack of biological tests, since the modeling cannot take into account specific or more complex cellular strategies initiated upon exposure to potentially toxic compounds. Such modeling can only rely on already available information and knowledge (e.g., on AOPs) or on what can be derived from molecular dynamics simulations (which currently can simulate the response of only small volumes of cellular contents). In silico simulations must be carried out on supercomputers, so they are also costly and time-consuming.

[0007] With the above in mind, there is a need for efficient, reliable, rapid and cost-effective methods for testing different materials and compounds that avoid the use of animals or isolated in silico modeling.

[0008] [Latest technology] Although many methods have been proposed and developed as alternatives to animal testing, some of which are mentioned above, this section includes only exemplary methods with the aim of highlighting the novelty of the approach used in the present invention.

[0009] Patent application WO0047761 discloses a method for analyzing the toxicity of chemicals, in which the same cell types, i.e. wild type and mutant strains of yeast cells, bacterial cells, cell lines of human origin, are used to detect oxidative stress, protein damage, cell cycle disruption, energy charge and depletion, microtubule disruption or initiation of metabolic capacity by overexpression of human gene inserts encoding metabolic genes or by incorporation of S9 fractions. In a preferred embodiment of the invention, wild type yeast and individual mutants are administered the desired chemicals and yeast growth is determined using turbidimetry. Dose-response curves are generated and the mutant sensitivity to the compound compared to its parent (relative sensitivity) is calculated. A statistically significant relative sensitivity indicates hypersensitivity of the mutant to the test compound. This approach is significantly different from the present invention, since no mutant strains or calculation of relative sensitivity is required.

[0010] Patent application WO2009146911 describes a self-contained sensor-controlled organ-on-chip device, which allows to establish or maintain organs or organoids as well as stem cell niches in a miniaturized chip format, suitable for live cell imaging and online observation, for example, by two-photon microscopy. The use of the organ-on-chip device in testing the activity, pharmacodynamics and pharmacokinetics of compounds is also described, the testing method comprising the steps of: providing an Organ-Chip device; adding one or more test compounds to the organs / organoids on the chip; evaluating the organ / organoid microscopically and / or determining one or more parameters determinable by one or more sensors;

[0011] However, the document is silent on the details of the test method, parameters or expected microscopic changes in the organs / organoids. Despite this, it is clear that only one time point after exposure is taken to observe the possible effect of the test compound on the organs / organoids, and there is no deep analysis of the impact of this effect on long-term effects, which is a crucial part of this invention.

[0012] Murschhauser et al. (Communications Biology Vol. 2, Article No: 35, 2019) describe a high-throughput microscopy method for single-cell analysis of event-time correlations in nanoparticle-induced cell death. In particular, the method is applied to extract event times from fluorescence time traces of cell death-related markers in automated live cell imaging on single-cell arrays (LISCA), using epithelial A549 lung and Huh7 liver cancer cells as model systems. In paired marker combinations, the chronological order and lag time of lysosomal membrane permeabilization, mitochondrial outer membrane permeabilization and oxidative burst were evaluated after exposure to 58 nm amino-functionalized polystyrene nanoparticles (PS-NH2 nanoparticles). The results show that different cells at different doses of nanoparticles show heterogeneity and interdependence in signaling pathways. The solution does not provide predictions about possible adverse effects of nanoparticles, but rather indicates mechanisms of signaling pathways and cellular responses to various doses of nanoparticles. The authors predict that the high-throughput assay will be a powerful tool for studying mechanistic and temporal heterogeneity in cellular responses to diverse types of perturbations, but they do not show how this could be translated into predicting adverse outcomes, only that they make such a prediction.

[0013] In summary, all currently known methods of material testing and toxicity assessment are based on the measurement or detection of a set of specific events in cells and / or tissues and / or organ chip assemblies after exposure. In contrast, the present invention aims to predict long-term toxicity based on early cellular events and using time propagation. Thus, it upgrades early event monitoring by additional construction of mathematical models, which automatically learn about the initial mechanisms and propagate the observed early event evolution to subsequent adverse outcome predictions.

[0014] Patent application WO2018217882 discloses a microfluidic small airway chip infected with one or more infectious agents (e.g., respiratory viruses) as a model for respiratory disease exacerbations (e.g., asthma exacerbations). Upon exposure, the cells of the lung epithelium within the chip were analyzed for the expression of phenotypes characteristic of asthma. Various phenotypes were observed, the recognition of which was based on the knowledge of asthma exacerbation mechanisms. Although this invention represents a very complete approach to the observation of early molecular changes induced by toxicants in an advanced in vitro system, it lacks an important step forward - it does not predict how these events evolve over time, which our invention addresses. Thus, it does not replace end-point tests, which only monitor the early evolution.

[0015] If the mechanism of event propagation could be known in advance and in vitro model monitoring could be automatically transformed into a mathematical description of the propagation of events, then time propagation could be performed by the model recently suggested by Kokot et al. (2020, doi: 10.1002 / adma.202003913). In this study, the authors first investigated the toxicity of inhaled nanomaterials and reported the discovery of new molecular key events and their causal relationships. In this case, said events were the processes of nanomaterial sequestration (nano-sequestration) and nanomaterial circulation between different lung cell types, which led the cells to enter a vicious circle of chronic inflammation. Similar circulation events (loops) were observed in the Covid-19 pneumonia cases reported by Mould and Janssen (2021, Nature, Vol 590, 553).

[0016] After the mechanism was discovered, the causal chain of events was explained by a simplified theoretical model describing the flow of nanomaterials between four separate compartments: i) extracellular, ii) intracellular, iii) sequestered in the cauliflower, and iv) intramacrophage. The model is defined by several parameters, however, which can be simplified to three main descriptors: a) Percent toxicity of nanomaterials to individual cells (determined by the number of viable macrophages measured in MH-S monocultures after 4 days of exposure); b) Nanomaterial sequestration rate by epithelial cells (calculated from the measured percentage of nanomaterial in cauliflower in LA-4 monocultures after 2 days of exposure), taking into account correction for the toxicity rate of nanomaterial to individual cells; c) Efficiency of signaling and monocyte influx replacing dying macrophages (calculated from measured influx of inflammatory cells (leukocytes) after at least 10 days in vivo), the time point at which chronic events of the response begin to develop, calculation includes correction for the rate of nanomaterial toxicity to individual cells (as well as for the rate of nanomaterial sequestration by epithelial cells).

[0017] Based on this mechanistic understanding and the corresponding mathematical formulation, including simplified parameterizations, the authors attempted to predict chronic inflammation induced by nanomaterials by these nanomaterial descriptors determined from three single time-point measurements: two in vitro and one in vivo (all measured several days after exposure). Using the above rates, the model was able to reproduce the amount of sequestered nanomaterial in cauliflower, the signaling for immune cell influx, as well as the in vivo time course of total macrophage numbers, which can be used to predict nanomaterial-specific acute-chronic inflammatory outcomes. However, the paper is silent on how the time course of various in vitro and in vivo observable events translates into a prediction of the adverse outcome itself. Moreover, the screening strategy for material safety assessment proposed by Kokot et al. is based on an understanding of the response of an organism to nanomaterial exposure, from the initial contact with the nanomaterial to potential adverse outcomes. This means that known molecular mechanisms are needed to predict the toxicity of proposed test materials, ultimately demanding a multidisciplinary approach, possibly combining advanced imaging, omics, particle labeling and tracking techniques applied in vivo and in vitro with in silico modeling. As a result, for each material, a great deal of effort must be directed towards discovering the mechanism of effect the material has on a particular type of cell, which will undoubtedly lead to lengthy and expensive toxicity testing. [Prior art documents] [Patent documents]

[0018] [Patent Document 1] US Patent Application Publication No. 2020 / 408744 [Patent Document 2] International Patent Application Publication No. 2020 / 172670 Brochure [Patent Document 3] International Patent Application Publication No. 2009 / 146911 Brochure [Patent Document 4] British Patent No. 2585150 [Patent Document 5] International Patent Application Publication No. 2018 / 217882 Brochure [Patent Document 6] US Patent Application Publication No. 2020 / 224136 [Patent Document 7] US Patent Application Publication No. 2020 / 283732 [Patent Document 8] US Patent Application Publication No. 2020 / 239857 [Patent Document 9] Japanese Patent Application Publication No. 4-148695 [Patent Document 10] European Patent Application Publication No. 2154241 [Patent Document 11] International Patent Application Publication No. 2007 / 120699 Brochure [Patent Document 12] International Patent Application Publication No. 00 / 47761 Brochure [Non-patent literature]

[0019] [Non-Patent Document 1] Greek and Menache, 2013, doi:10.7150 / ijms.5529 [Non-Patent Document 2] Braken, 2009, doi:10.1258 / jrsm.2008.08k033 [Non-Patent Document 3] Watts and Geoff,2007,doi:10.1136 / bmj.39058.469491.68 [Non-Patent Document 4] Edelman et al,2010,doi:10.1002 / wsbm.75. PMC 3157287. PMID 20836040 [Non-Patent Document 5] Mark TD et al. Computational Toxicology 21,2022,https: / / doi.org / 10.1016 / j.comtox.2022.100213 [Non-Patent Document 6] Murschhauser et al., Communications Biology, Vol. 2, Article No. 35, 2019 [Non-Patent Document 7] Kokot et al., 2020, doi: 10.1002 / adma.202003913 [Non-Patent Document 8] Mould and Janssen, 2021, Nature, Vol 590, 553 Summary of the Invention [Problem to be solved by the invention]

[0020] [Description of the solution to the technical problem] The main disadvantage of the solution by Kokot et al. and other cited references is that for each new material, a mechanism (mode of action) must first be discovered, after which new observations and their interaction terms can be introduced into the model for prediction of material toxicity. Moreover, known methods based on in vitro and in vivo experiments are performed in a destructive manner, providing information about events at only one time point. The present invention therefore aims to address these significant disadvantages and to provide a significantly upgraded method for predictive testing of materials, chemicals, medicines, vaccines and similar compounds to evaluate their safety, potential adverse outcomes and / or induction of diseases. The present invention upgrades the monitoring of early events with a method that automatically learns about early mechanisms by converting the propagation of early events observed in vitro into a mathematical form that is then used to propagate the evolution of early events to late adverse outcome predictions. The technical problem is solved by the method described in the independent claims, while preferred embodiments of the method are defined in the dependent claims. [Means for solving the problem]

[0021] The main advantage of the invention is that it is not necessary to understand the mechanism (mode of action) of the test compound, while non-destructive time-lapse monitoring of a living in vitro model is converted into a mathematical description of the evolution of observations (key cellular events), the latter being converted into a prediction of an in vivo-observable adverse outcome. The method essentially allows for the automated discovery of the mode of action of a test agent, e.g., a material, a chemical, a medicine, a vaccine or similar substance or particulate matter, on cells. The invention further allows for the determination of the systemic effect of the test agent. Substances that can be tested using the method according to the invention can be selected in the group consisting of materials, nanoparticles, vaccines, compounds, chemicals, medicines, etc. It is not intended that the invention be limited by a particular test agent or by the particular cells in which the agent is tested.

[0022] For a better understanding of the subject matter, the meanings of the following expressions used throughout the text are explained below:

[0023] [Table 1]

[0024] [Table 2]

[0025] Most importantly, the alternative testing method according to the present invention saves time in correlating slowly progressing adverse outcomes with possible inciting agents, by replacing the classical observation of late adverse outcomes with in silico monitoring of early critical event evolution in vitro and further propagation of this evolution over time.

[0026] Such time savings can be easily illustrated in the case of the development of chronic inflammation (which is further known to lead to the development of cancer or autoimmune diseases causing a significant number of deaths worldwide). The time from exposure to said adverse outcome in conventional animal-based tests is usually 14-28 or up to 90 days after exposure. The present invention can achieve this within a few days, and thus much earlier. That is, it first uses cells, cell cultures and / or chip-like devices to estimate and parameterize the initial evolution of a biological system, which is then propagated in silico and outcomes predicted. Thus, the present invention is based on the evolution of molecular events that first occur long before they are manifested as long-term effects in tissues.

[0027] In the method according to the invention, these molecular events are analyzed, measured and / or detected in a time-lapse manner (at least three time points) and in silico time propagation is performed, where material testing is only performed for up to one week, preferably up to three days, most preferably up to 50 hours, and early events are associated with long-term effects, leading to reliable predictions about the safety / toxicity of the material. Not all early molecular events are associated with adverse long-term effects, however, the in silico time propagation is sequenced to recognize only relevant early events.

[0028] Thus, the essence of the invention is the in silico prediction of in vitro detectable and in vivo relevant observations, and their initial values.

number

number

[0029] In a preferred embodiment, the time derivative of the quantified observation is

number

number

number

[0030] In another preferred embodiment, the interaction matrix term IM ij is the power parameter p ij and at least two interaction functions IF ij (interaction matrix term IM ij Observation value q of i (which specifies the dependency on

[0031] The method is not limited by the types of observations selected, their number and / or the manner in which they are determined, calculated, or detected.

[0032] In a preferred embodiment, the method according to the invention comprises three main steps, namely: a) a preliminary step of associatively identifying relevant observations, preferably based on prior knowledge, AOPs, in vivo data and previous results obtained by the method; In vitro experiments, including the following steps: b) preparing an in vitro model for preparing key events of relevance and detrimental relevance to a biological system, preferably selected based on the expected function, mode of action and / or entry point of the biological system, wherein the in vitro model comprises at least one cell type; c) a scenario-based time-lapse acquisition for detecting the initial time evolution of observations in said in vitro model, using a minimum of three time points with any time delay longer than 5 minutes (corresponding to the fastest biological situation change) and a total duration of up to one week (preferably less than two days, corresponding to the longest interval allowing a reliable determination of the evolution of events), within a minimum of two scenarios, where a scenario is a subgroup of measurable observations that are structural, chemical, biological, mechanical, physiological, immunological and similar events, at least one scenario representing an unexposed in vitro model and at least one for an exposed in vitro model, where the test substance is applied at a dose capable of inducing at least a minimal effect in the in vitro model (in toxicology it is called LOAEL - lowest observable adverse effect level); d) quantification of the observations to quantify their values ​​in the in vitro model, where at least one observation is indicative of a characteristic of the unexposed in vitro model and at least one observation is related to a characteristic associated with an exposure-related change in the in vitro model; e) generation of a time series of said observations and time derivatives to obtain a vector of observations for said scenario, which defines the time evolution of the in vitro model and its constituent events; · Mechanism determination, including the following steps: f) Construction of the interaction terms of the interaction matrix with interaction functions to define all possible interactions and connections between the observables, both biophysically and biologically, which in turn mathematically determines all the time evolutions of the observables; g) constructing basis functions and super-equations to transform the set of interaction functions from the previous step into a mathematically orthogonal set of functions, which can be used in the next step to numerically parameterize the evolution of the observations; h) parameterization of the interaction terms defining the rate at which the observations change, said parameterization using the scenario-derived experimental data, the orthogonal set of functions defined previously and the selected numerical method to determine a system of equations allowing the numerical time propagation of the observations in a next step and the identification of the optimal interaction terms in a second step; i) a scenario-dependent fit to the in vitro experiments to derive values ​​for each of the observations for each scenario starting from given initial conditions for each observation, using the parameterization of the interaction matrix from the previous step; j) Optional presentation of determined mechanisms that identify optimal interaction terms (modes of action) and explain the evolution of observed early events-observations for the test agent and their contribution to their development into potential adverse outcomes; - Prediction of outcome based on the results obtained using the above steps, the prediction may optionally further comprise the steps of: k) Optional dose biodistribution characterization, replacing costly animal-based determination of dose-related information, such as the No Observed Adverse Effect Level (NOAEL dose) and the Lowest Observed Adverse Effect Level (LOAEL dose), preferably with in vitro-based image-derived biodistribution. l) Dose-dependent long-term propagation of observations, with eventual use of parameterized interaction matrices (determined mode of action) in the long-term propagation of AO-related observations to determine AO-predictions, long before the AO has evolved and been observed in animal studies.

[0033] In the method described by Kokot et al., the observations (events) were not grouped, but a description of the model responsible for the time propagation of events, relying on predefined differential equations and interaction terms - two single in vitro and one in vivo measurements - was used in the prediction based on a known mechanism previously discovered for a known material. The mechanism then allowed the creation of a nomogram-based parameterization, which was already considered as an in silico time propagation. It was not possible to apply the prediction to materials for which a mathematical model had not been evolved from the mode of action.

[0034] On the contrary, the present invention represents a major upgrade of the method by Kokot et al. by eliminating the need for knowledge and understanding of the link between material toxicity, mechanisms of cellular response and individual early cellular events. The predictions enabled by the method by Kokot et al. can only be used for materials that generate toxicity by exactly the same pre-discovered mechanisms with all the interaction terms for the observations associated with the pre-defined in vitro model, whereas the current invention only requires knowledge of which observations are in vivo relevant and which are in vitro observations. In the present invention, the time propagation of the critical events (in vitro systems) used in the prediction of the outcome is achieved using a general system of differential equations (later automatically parameterized using the concept of interaction matrices, a set of scenarios, related to the individual experiments and used to include / exclude individual observations), as well as a specially generated algorithm that can convert the time derivatives of the observations into power parameters of the individual interaction matrix elements (interaction terms).

[0035] The method for predictive testing of agents according to the invention is disclosed in further detail on the basis of non-limiting exemplary embodiments, examples and figures. [Brief description of the drawings]

[0036] [Figure 1] FIG. 1 is a schematic diagram of an alternative testing method made possible by the present invention compared to animal-based endpoint-directed testing. [Diagram 2] 1 is a schematic overview of the main steps a)-l) of the method according to the invention. [Diagram 3] A non-exclusive list of in vitro models that may be used in step b). [Figure 4] 1 is a non-exclusive list of possible methods for time-dependent data acquisition in step c). [Diagram 5] FIG. 11 is an example of an image quantification process with two observations related to cell properties and three observations related to the agent (toxicant) surface used in step d). [Figure 6] FIG. 1 shows possible interaction functions and their parameterizations used in step f), with all biologically relevant, numerically applicable interaction functions grouped within the dashed region, and two additional interaction functions used in the approximation of the quotient interaction function grouped within the dotted region. [Figure 7] 1 shows a generalized interaction matrix with 9 blocks shown together with an exemplary interaction term representation and interaction functions used in step f). [Figure 8] An exemplary fit to the initial 30 h evolution of events corresponding to exposure of the alveolar epithelium with the two selected materials is shown, with all observations indicated with callouts as they appeared in step i). [Figure 9] 1 shows the graphical pictograms used in the mechanism presentation of step j). [Figure 10] A parameterization-based mechanistic presentation of TiO2 nanotube action used in step j) is shown, with a triple encoding of significance - slight effects are almost transparent and appear faded in grey, a colour change from green to magenta represents a positive to negative effect, the affected quantities are in the inner circle while the quantities causing the effect are on the outer circle, and autostimulatory and autoinhibitory effects are depicted with semicircular arrows. [Figure 11] 1 shows a parameterization-based mechanistic representation of multi-walled carbon nanotube action used in step j). [Figure 12] FIG. 1 shows the image-based dose biodistribution calculations used in step k) shown for the case of two-channel image analysis, which aid in the determination of toxicant-surface vs. membrane-surface dose. [Figure 13] FIG. 1 shows an expanded interaction matrix with an additional part that is not defined by the in vitro system but is defined by AOP knowledge and other available in vivo relevant data used in step l), indicated with the dotted region. [Figure 14]A representative example of actual acquired time-lapse 3-channel (2 fluorescence and 1 backscatter) images of a TiO2-exposed in vitro lung model as part of the data acquisition in step c) is shown. [Figure 15] FIG. 13 shows an example of an interaction matrix implementing the interaction functions for the case of chronic inflammation prediction in a TiO 2 nanotube-exposed in vitro lung model as part of step f). [Figure 16] Shown is the simplification of the interaction matrix in step f), where nulled interaction terms are indicated by zero, e.g., biodegradation / biosynthesis (Block D), adsorption / desorption from phenomenon-related structures (Block B), growth interference between cell lines (cross terms in the top left block of Block G), indirect toxicity (cross terms in the top center block of Block T), and relocation / direct exchange of toxicants between cell lines (center block R) are eliminated. [Figure 17] As part of step f), an example of a simplified interaction matrix implementing the interaction functions for the case of chronic inflammation prediction in a TiO2 nanotube-exposed in vitro lung model is shown. [Figure 18] FIG. 13 shows an example of a basis function and associated parameter list corresponding to a simplified version of the interaction matrix for the case of chronic inflammation prediction of a TiO2 nanotube-exposed in vitro lung model for each of the six scenarios appearing in step g). [Figure 19] Parameterization of the generalized (A) and simplified (B) interaction matrices used for prediction of TiO2 nanotubes and derived by SVD, showing the effect of the additional constraint in the box (positive definiteness of the parameters in block R). [Figure 20] An example of the final results (step l) is shown, including a dose distribution histogram, an 800-hour long-term propagation of all observations indicated by callouts, and a long-term prediction of chronic-inflammation related observations (macrophages) associated with an in vitro lung model exposed to TiO2 nanotubes, where the left and right graphs (bottom) correspond to the minimum and maximum local doses, respectively, from the dose histogram (top). [Figure 21]An example of two- and three-channel microscopy images used to predict adverse outcomes for three different agents: a virus-like particle-based vaccine, micro- and nanoplastics, a drug and a nutraceutical, the cellular compartment is always shown in the green channel and the agent (potential toxicant) channel is indicated by the text color on the image. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0037] A primary objective of the current invention is to provide a means to transition from current toxicity testing concepts that rely on adverse outcome testing in animals to new tests that can rely on the observation of early events that occur shortly after a molecular initiating event (the first response to exposure) and much earlier than the adverse outcome. As such, the current invention can significantly shorten the time required to conduct a test.

[0038] FIG. 1 shows a comparison of the time required for testing between known methods of (animal-based) in vivo testing (where long-term outcome is only analyzed 14, 28 or 90 days after exposure), with the method according to the invention allowing the detection of early molecular events indicative of long-term outcome much earlier, for example up to 48 hours after exposure.

[0039] The current invention is a method for predictive testing comprising 12 clearly defined steps, some of which are optional and included to improve reporting of the test to the user. Implementation of the entire method is illustrated in a non-limiting manner in Examples 1 and 2. Various implementations of the individual steps are shown in Examples 3, 4 and 5.

[0040] FIG. 2 shows a preferred embodiment of a method for predictive testing, in which the following steps are performed: a) Associative identification of relevant observations based on prior knowledge, AOPs, in vivo data and previous results obtained by the method In vitro experiment including the following steps: b) preparing a suitable in vitro model based on the observations and knowledge identified in step a); c) scenario-based time-lapse acquisition of events in the in vitro model prepared in step b), having at least three time points at any time delay longer than 5 minutes and within a minimum of two scenarios, where a scenario is a group of measurable observations that are structural, chemical, biological, mechanical, physiological, immunological and similar events; d) quantification of the observations to quantify their values ​​in the in vitro model, at least one observation representing a characteristic of the unexposed in vitro model and at least one observation relating to a characteristic associated with an exposure-related change in the in vitro model, all of which observations are obtained in step c); e) generating a time series and time derivatives of the observations determined in step d) to obtain the time evolution of the observations for said scenarios; · Mechanism determination including the following steps: f) constructing the interaction terms of the interaction matrix by the interaction functions to define all possible interactions and couplings between the biophysical and biological observables, which in turn provides the most generalized expression of the relationship between the time derivatives of the observables obtained in step e), thereby mathematically determining all the time evolutions of the observables; g) constructing orthogonal basis functions and super-equations for the interaction matrix of step f) using all scenario-relevant combinations of the variables monitored in step c); h) parameterization of the interaction terms using the time-dependent values ​​and derivatives of the observations of step g) and step e), the obtained parameters defining the rate at which the observations change, said parameterization using the scenario-derived experimental data and using the previously defined set of orthogonal functions and the selected numerical method to determine a system of equations allowing the numerical time propagation of the observations in a next step and the identification of the optimal interaction terms in a second next step; i) a scenario-dependent fit to the in vitro experiments using the initial values ​​of the observations of steps d) and e) and the time evolution of step f) together with the parameterization determined in step h) for each scenario used in step c); j) optionally, a mechanism representation using the parameterization derived in step h); and The prediction optionally further comprises the steps of: predicting an outcome based on the results obtained using the steps above: k) dose biodistribution characterization using image-based dose biodistribution analysis from the images acquired in step c); l) Dose-dependent long-term propagation of the observations identified in step a), the interaction matrix parameterized in step h) and the observation evolution derived in step j) for each of the doses of the distribution histogram determined in step k).

[0041] The end-user-friendly and understandable results of the method for predictive testing of active substances according to the invention are most importantly provided by steps j), k) and l).

[0042] The results of step j) describing the mechanism (mode of action) may be represented by graphical material similar to that of FIG. 10 and may include the following statements: Agent T is highly toxic to the cell type Ci Agent T is not toxic to the cell type Ci Agent T is contact toxic to cell type Ci Agent T is indirectly toxic to cell type Ci when taken up in cell type Cj ·Localized Pli stimulates / inhibits proliferation of the cell type Cj · Cell type Ci increases / decreases the accessibility of agent T in compartment Cj · Cell type Ci induces aggregation / biodegradation of agent T in compartment Cj The agent T is efficiently taken up by the cell type Ci Agent T is efficiently taken up by cell type Cj from cell type Ci Agent T is isolated Phenomenon Plj increases / decreases the accessibility of agent T in compartment Cj Phenomenon Plj induces aggregation / biodegradation of agent T in compartment Cj When the agonist T is in the cell type Cj, it stimulates / inhibits the phenomenon Pli When agonist T is isolated, it stimulates / inhibits the phenomenon Pli · Pli stimulates / inhibits Plj Phenomenon Pli self-stimulates / self-inhibits its action

[0043] The results of step k) describing the biodistribution of the agent may be represented by a graphical material similar to the portion of FIG. 12 and may include the statements of step j) and / or some of the following statements: Agent T is homogeneously dispersed in in vitro models of different sizes, demonstrating that the biological system prefers the dispersed form of Agent T and not its aggregated state. Agent T was detected in large aggregates, indicating that biological systems prefer the aggregated state of agent T Agent T is efficiently sequestered on / in / beside / outside the cell type Ci Agent T is not taken up by the cell type Ci The agent T is distributed within the cell type Ci in the form of many small submicron-sized particles.

[0044] The results of step l) provide a final adverse outcome prediction associated with the agent, which may be represented by graphical material similar to that of portion of FIG. 20, combining statements from steps j) and k), adding additional statements to further define the evolution of the biological system, and / or providing other types of statements. The following statements may be included: Agent T is safe - biological systems do not respond with abnormal evolution Agent T is safe - Agent T does not cause delayed reactions Agent T is safe at doses that might be expected to be encountered under specific conditions Agent T is safe at doses below a certain value Agent T is safe at dose rates below a certain value Agent T induces a weak / strong / very strong acute response (above a certain dose or dose rate) which, however, dissipates over a specific time period Agent T induces an acute response (above a certain dose or dose rate) which can later be amplified (above a certain different dose or dose rate) into a strong subacute / subchronic response and potentially transform into a long-lasting chronic inflammation Agent T caused immediate and severe cellular damage, reflecting its high toxicity. Agonist T induces a weaker but longer-lasting response in terms of inducing a constant increase in the phenomenon / immune response. Agent T induces a weak acute response, however this later amplifies and results in ever-increasing inflammation (cytokine storm) that cannot be resolved.

[0045] The steps of the method for predictive testing are carried out as follows:

[0046] a) Associative identification of related observations The choice of observations depends on existing knowledge about the test substance and the occurrence of possible substance-related outcomes (symptoms).

[0047] For example, at least some of the observations can be selected based on the AOP, previous in vivo data and / or scientific reports and papers. At least one of the observations should preferably correspond to the earliest event observed in vivo or in a patient when adverse outcome evolution is studied.

[0048] For example, if the toxicity assessment deals with inhalation and the lung is the organ of interest since it is the first exposed, observations related to chronic inflammation may be associated with, for example, the following immune cells: the number of monocytes that enter the lungs (as they differentiate into macrophages to replace dead resident macrophages), the number of polymorphonuclear (PMN) cells entering the lungs (because they are most conveniently identified ex vivo and because they correlate well with the influx of monocytes); · The numbers of leukocytes entering the lungs (because their time-dependent rate changes are the most marked and also correlate well with PMN influx).

[0049] b) Preparation of a suitable in vitro model The choice of in vitro model thus initially depends on the observations required to reproduce in vivo relevant events, which are selected in step a). The latter may likewise depend on the test substance, e.g. inhaled particles are tested in an in vitro model relevant to the lungs (either the bronchial or alveolar parts, depending on the size of the substance) and includes pulmonary epithelial cells and / or pulmonary macrophages and / or fibroblasts and / or endothelial cells and / or other immune or other type(s) of cell.

[0050] In vitro models can include any combination of monocultures of cells, or various co-cultures of cells or more complex cell cultures, depending on the details of the test substance. Thus, cells for in vitro models can be selected from the group consisting of at least lung epithelial cells, skin epithelial cells, neurons, endothelial cells, muscle cells, intestinal epithelial cells, mucus cells, parietal cells, chief cells, endocrine cells, immune cells (macrophages, glia, ...), etc.

[0051] The skilled artisan will know the appropriate relevant combination of cells for the analysis of a particular substance. Growth media and conditions will be optimized for the selected in vitro model that mimics the relevant tissue and supports the growth of the selected combination of cells.

[0052] Therefore, the identification of the observed values ​​(step a) directly influences the choice of cell type for constructing the selected in vitro model. Therefore, the construction of an in vitro model includes the following requirements: the selected in vitro model is relevant for the particular agent evaluation, mimicking the relevant target tissue(s) as well as at least a relevant portion of the relevant agent delivery pathway; All cell types or combinations used are structurally or functionally mimicking AOP-associated parts of the target tissue; and -Can express one or more AOP-related important events identified from prior knowledge; All cell types together can respond to the relevant early part of the AOP, which is - precedes the later part of the AOP, which is observed in vivo and develops into the observed adverse effects, and Leads to at least one significant event that has the power to predict an adverse outcome; Here, AOP refers to the general concept of an adverse outcome pathway, including all events that are causally related from a molecular initiating event and all subsequent events to the final adverse outcome, regardless of whether the events are physical, chemical, biological, biophysical, biochemical or any other type.

[0053] In Example 3 and FIG. 3 a non-exclusive list of possible in vitro models for carrying out step b) is given, which are related to different types of in vivo events (chronic inflammation, neurodegeneration, cancer and other slowly progressive diseases of different organs).

[0054] c) Scenario-based time-lapse acquisition After the observations have been identified (step a) and a suitable in vitro model has been constructed that allows mimicking the molecular initiation event and its further early evolution (step b), methods for detecting such molecular events need to be identified and selected.

[0055] The method of obtaining observations of interest in a living in vitro model may be any suitable and widely used method, preferably in time-lapse mode with the required temporal resolution to capture the dynamics, typically on the order of minutes to hours, such as: microscopy, such as bright field, fluorescence wide field or confocal, scattering, Raman, ESEM, FTIR, etc., allowing the observation and quantification of the time evolution of said observables (related to structural or biochemical information, see below) and the determination of their time derivatives, Spectroscopic methods, e.g. Raman, FTIR, UV-VIS, fluorescence, specific stains and similar, that allow the determination of the concentration / amount of compound classes of interest (e.g. proteins, lipids, aroma substances, labeled substances, tagged substances, inorganic substances) in the entire sample, the quantification of their time evolution and the determination of their time derivatives, Immunological methods, e.g. ELISA, antibody isolation and purification, ELISPOT, immunoblotting, immunohistochemistry, immunoprecipitation, immune cell isolation, etc., that allow the determination of the concentration / amount of a particular compound of interest (e.g. a selected cytokine) in the entire sample and the quantification of their time evolution and the determination of their time derivatives, -Omics data, e.g. proteomics, lipidomics, transcriptomics, metabolomics, etc.

[0056] The observed initial event (observation) can be any effect that the test substance has on the cell system, for example a change in a molecular (e.g. lipid, protein, RNA, DNA, etc.) or supramolecular (e.g. membrane, ribosome, cytoskeleton, fibril, vesicle, etc.) or cellular (cell surface, volume, shape, activity, etc.) property that can be quantified, the event is usually selected from the group consisting of: Changes in cell morphology, shape, and mobility, Cell aggregation and / or dissociation, · Changes in gene expression, Increased or decreased lipid expression, Increased or decreased transport of molecules and other structures to or from the extracellular space (vesicles, etc.) -Breaking of tight connections between cells, Actin and tubulin rearrangements, chromatin (de)condensation and other nuclear phase changes, ·Cell lysis, apoptotic or necrotic events or events associated with other cell state changes, · Alterations in the ER, endosomes, lysosomes, mitochondria and / or ribosomes, ·Cell death, Sequestration, phagocytosis or other new structure formation on the cell surface or inside the cell, e.g. sequestration as described by Kokot et al. Transcription, synthesis, expression, relocation and release of attractants from immune and other types of cells, · The binding state, charge, interaction surface chemical composition (intrinsic or extrinsic) of molecules, molecular complexes and supramolecular structures, as well as aggregate structures; the type of interaction exhibited by the interaction surface with any organelle or molecular complex, supramolecular structure or external / exogenous material, Release and / or absorption of molecules, vesicles, etc. from one intercellular / subcellular compartment to another, the surface area of ​​the cells of the selected type (which also corresponds to the number of cells or the concentration of cells, etc.), Surface area of ​​the agent (toxicant) in the selected compartment (also corresponds to mass, concentration, volume, etc.) the amount of signaling or transport molecules or other structures (corresponding to mass, concentration, binding state, etc. of cytokines, chemokines, enzymes, receptors, various types of RNA, exosomes, endosomes, lipid bodies, etc.), Surface area of ​​new or modified structures (also mass, volume, density, lifetime, dimensionality, correlation length, shape descriptors, etc. of supramolecular structures, e.g. vesicular structures, cell or intracellular membranes, intracellular or extracellular fibrils, tight junctions between cells, intercellular gaps, mitochondria or mitochondrial networks, nuclei or subnuclear parts / domains, etc.).

[0057] In Example 4 and in FIG. 4 a non-exclusive list of possible ways allowing time-dependent data acquisition for carrying out step c) is given.

[0058] d) Quantification of Observations Quantification of the observations in control (without substance) and exposed (test substance applied at the selected dose) in vitro models depends on the type of acquired data. If the latter is already in numerical form (e.g. ELISA, transcriptomics, proteomics), it is directly passed to the next step of derivative generation, otherwise, for example when images have been acquired, the analysis is carried out in a two-step method: Masking of parts of the image related to the event (observation), e.g. type of cell line, labeling of cells or cell structures (if appropriate), sensitivity of the method, colocalization information, etc., as well as Quantification of the corresponding observations, where the mask from the previous step is transformed at each time t and for each region of interest (ROI) into quantified observations q, where the surface area is usually obtained by summing the number of pixels in the mask and the quantity is derived by summing the intensity of the appropriate image channel(s) in the mask.

[0059] Optionally, additional histogram or intensity distribution analysis (e.g., FIDA, Fluorescence-Intensity Distribution Analysis) is used to further enhance the quantification procedure. Quantitative results from multiple ROIs can be summed (e.g., surface area of ​​a cell type or a specific structural response), effectively increasing the field of view of a single experiment and thus reducing noise (due to local variations) in the analysis.

[0060] For illustrative purposes, the surface area of ​​a selected cell type or a selected structure type is determined as the number of pixels assigned to the selected cell type or the selected structure type, which requires a binary mask of the corresponding cell / structure. In images with one cell type and a single expression event, such a mask is trivially derived by binarizing the image. In images of more complex cell cultures, such as co-cultures or cultures in which several structures or biochemical events can be expressed simultaneously, multi-labeling is often used (co-localization or off-localization), together with morphological information (size, dimension, shape, etc.) and information from a z-stack to distinguish different events and obtain several masks from the same image (localization above and below the analyzed image). Preferably, when the quantity is associated with structures exhibiting certain geometric or morphological properties, such as size, aspect ratio, density, bulk or surface of the structure, etc., the mask is segmented by one of the available image segmentation algorithms and each segmented object is analyzed in terms of the desired properties. The mask is thus reduced to incorporate only (part of) those objects that meet the desired criteria. For example, a mask of sequestered agents-toxins (e.g., materials that become inaccessible to cell sensing mechanisms or cellular processes) is derived by binarizing the intensity of the channel corresponding to the material, followed by segmentation to find aggregates above a threshold size, and final removal of pixels of the mask that correspond to the surface of the segmented object.

[0061] In Example 5 and FIG. 5, five possible quantification processes are presented to non-exclusively describe how images are processed and observations are quantified.

[0062] e) Generation of time series and time derivatives As soon as the observations are quantified at all measurement and analysis time points, derivation of the time series and time derivatives of the observations is performed to provide time-dependent observations and time-dependent time derivatives of the observations. The latter is preferably performed in a noise-reducing manner that takes into account information from all available time points and ROIs, improving the robustness of the results due to limited field of view, cell movement during time-lapse monitoring, changes in labeling and labeling efficiency during monitoring, limited number of observed significant events, detection noise, masking, uncertainties generated during segmentation, etc.

[0063] The noise reduction scheme for determining the time derivatives includes the following assumptions: Each observation (significant event) is mostly slowly varying in time and has at most one maximum or minimum in the time interval of the experimental monitoring; Each of the observations can have at most one segment, which means that it varies fast in the time interval of the experimental monitoring and that the absolute value of the derivative of the observation has at most one maximum or minimum; Any other fast changes are due to experimental noise.

[0064] With that in mind, within a given time interval of experimental monitoring {t1..t max} time t j Observation value q i The measurement value of {q i (t j )} (i.e., the vector denoted by {}) represents a set of fitted values ​​{q i-f (t j )} (the additional index -f indicates the fit value), or by fitting {q i (t j )} can be interpolated by fitting it to a predefined function, for example:

number

[0065] Similarly, the differential value of the observations {dq i / dt(t j )} time sets are the same time points {t j A set of derivatives {dq i-f / dt(t j )}.

[0066] In the next step of mechanism determination, specifically the formation and parameterization of the interaction matrix, a set of experimental data {{t j}, {q i (t j )}, {dq i / dt(t j When referring to a time point, the following values ​​at a given time point, the observations and their time derivatives are used as inputs:

number

[0067] f) Construction of interaction matrix The interaction matrix is ​​constructed by the observation q i The (long-term) time propagation of is achieved using a system of differential equations, roughly written as

number

number

[0068] In some cases, when an interaction function depends on the quotient of two observations q and r, e.g. in concentrations (amount of agent (toxicant) or surface area / compartment volume), it is desirable to convert such an interaction function of a quotient into a product of two interaction functions, where one depends on the first observation and the other on the second observation:

number

[0069] As will be shown below, in practical cases the remnant O(q,r) represents less than 20% of the original value, and the above approximation is considered satisfactory.

[0070] To generalize the interaction function IF(q) to a form independent of the (previously) known mechanism, the following dependence of the interaction term on the observation q is assumed:

number

[0071] The sets (a,b,c) and (o1,o2) from Eq.[6] are related by the following relation:

number

number

[0072] The following six interaction functions (shown in dashed regions in FIG. 6 and identified below by pairs (o1, o2)) represent the most biologically relevant interaction functions (affect their rates): (1,1)-linear dependence on q, ·(1,0)-saturation dependence on q, · Threshold / delay dependence on (0,1)-q, · Saturation & threshold dependence on (0,0)-q, (0,2)-nonlinear / quadratic dependence on q, (-1,0) - a term that does not depend on q.

[0073] The two additional interaction functions (shown within the portion of the dotted region of Figure 6 that extends beyond the dashed region) represent the two functions used in approximating the interaction function using the quotient of two observations q and r referred to in Eq.[5]:

number

number

[0074] By parameterizing the interaction functions by Eq.[6] and their sensitivities by Eq.[4], the interaction terms in Eq.[4] no longer need to be pre-specified (as in Kokot et al.) but could, in principle, be determined using time-dependent experimental data by an inverse problem-solving procedure.

[0075] The scenarios mentioned before are used as a tool to include / exclude relevant / irrelevant observations in each of the experiments. This reduces the list of interaction terms for each scenario, as appropriate in each of the equations. Mathematically, this is done by rewriting the set of differential equations in matrix form:

number

number

[10] Here, a scenario is defined as a vector of 0 and 1 with dimension the number of observations. Multiplication of the interaction matrix by the scenario vector eliminates interaction terms in the differential equations that are not present in a particular experiment.

[0076] For illustrative purposes, also shown in FIG. 7, the observations q are subgrouped into three classes: cell surface area (compartment), amount of agent, i.e. toxicant (associated with compartment), and frequency of occurrence of the selected phenomenon, which logically leads to the evolution of the interaction matrix - a subgrouping of the interaction matrix terms into a 3×3 block form, where each block of the interaction matrix is ​​associated with (see the blocks shown in FIG. 7):

[0077] [Table 3]

[0078] Besides the grouping of interaction terms, Fig. 7 also shows the most probable class of interaction functions that can be associated with each of the interaction matrices, using the notation of Eqs. [4] and [6]. However, the choice of interaction function class may vary from case to case and is optimized for the data in the most general case. Note that some of the interaction terms are zero by definition - they are indicated by 0 in the correct place in the general form of the interaction matrix presented in Fig. 7. The nulled interaction terms are primarily related to the algebraic definition of the total cell surface, the definition of inaccessible toxicants, and the diagonal rearrangement-block related terms. Interaction terms / blocks greater than 1 indicate that the form of the terms differs between the block diagonal and / or the upper and / or lower triangles (also indicated by different fill colors behind). The interaction function dependent representation includes the observations together with an index of 1 or 3 (in line with the definitions in Eqs. [6] and [7]): the first index is always the index of the observation i or j, where i always denotes the row index of the block row and j always denotes the column index of the block-column, the second index is the possible / expected derivative o1, the third index is the possible / expected derivative o2; if the second and third indexes are omitted or denoted by x, they can take any of the possible integer values ​​and can be determined through a parameterization procedure. Multiple occurrences of interaction functions depending on compartment size (F(Ci)) may appear in different forms (with different indexes) in the same interaction term (e.g., one occurrence acts in terms of compartment-related normalization from a concentration-like dependence and the other from another compartment-dependent interaction function) due to the transformation of the quotient-dependent interaction function described before - in such cases the derivatives o1 and o2 are strictly prescribed - they correspond to the transformation rules and are not freely determined by experiment.

[0079] If less complex interactions between the observations (events) are to be detected or less complex biological systems are to be tested, the interaction matrix can be further simplified by omitting or eliminating additional parts or by designating their values ​​as zero (0). Such cases are further described in Example 1 and FIG. 16.

[0080] g) Basis functions and super equation formation The next step is to convert the system of differential equations into a set of terms, i.e. linear products of the interaction matrix elements power parameters and basis functions (which are sums of products of interaction functions), thus preparing the determination of the system-power parameters for automatic parameterization. To carry out the latter, all the differentials dq of the selected observations are i / dt (defined by the left-hand side of the corresponding differential equation - Eq. [3]) is first i,j It can be seen as a linear combination of products of interaction functions weighted by (a combination of Eqs. [3] and [4]).

number

[11]

[0081] In the example of the interaction matrix terms (see the functions listed in Fig. 7), some blocks mix interaction terms and some parameters p i,j It turns out that there are interaction terms greater than one (see the function set presented in Example 1), and these terms are made linearly dependent. To determine the value of such a parameter with the greatest sensitivity, Eq.

[11] must be rewritten as a linear combination of linearly independent equations.

number

[12] In the formula, p i,j Related to B i,j (the sum of the products of the interaction functions) is the union of the observations {q b} as basis functions.

[0082] To be as sensitive as possible, the determination of the parameters according to Eq.

[12] should be performed in a subgroup of equations, where only one subgroup of parameters appears and the other parameters do not (they appear only in the complement of this special subgroup of equations). This means that the other parameters can be determined independently by other equations (their subgroups). Symbolically, each such subgroup of equations is called a super-equation.

[0083] When solving such a super-equation (a system of parametrically coupled algebraic equations), the parameters and basis functions also need to be collected (mathematically speaking, this is translated into the union of the parameters and the union of the basis functions associated with the equations that have been combined into one super-equation). Symbolically, this is called the super-parameter set and the super-basis set.

[0084] Examples of super basis function sets and corresponding parameters are shown in Example 1 and FIG.

[0085] Note that the number of algebraic equations (groups of parameters with one super equation) used in determining the superparameter set is defined by the product: N algebraic equations = N ROI x N time points x N scenaria x N cupled equations

[13]

[0086] h) Parameterization of interaction terms Considering a typical experimental dataset, including data from approximately 10 ROIs over at least 15 time points for each of the six scenarios, and 1-4 coupled equations per super-equation, the number of algebraic equations per task exceeds 100 or even 400 (e.g., for parameters related to toxic relocation). Since the number of parameters appearing in such an algebraic system never reaches such a high figure, the system is always overdetermined and needs to be solved by optimization routines based on maximum likelihood or least squares, singular value decomposition or similar methods.

[0087] Here we show how this system can be solved using singular value decomposition (SVD) due to its simplicity and computational efficiency. However, this does not preclude the implementation of other routines, e.g. linear or non-linear optimization routines. Although SVD is the most computationally efficient, it is easier to constrain the power parameters to a certain range, e.g. positive values, with an optimization routine (see below).

[0088] Independent of the method chosen, an optimization problem is first formulated. For this, data on the values ​​of the observations and their derivatives are first used to construct a matrix M (not to be confused with IM) from the right-hand side of the set of Eqs.

[12] , where: Each row represents a super-equation with many terms, where {t, {q i}, {dq i / dt}} are replaced by their fitted values ​​according to Eq. [2], which correspond to a particular combination of ROI, time point and scenario, Each column is p by Eq.

[12] i,j (i.e., a sequence) and represents the value of one of the basis functions that is a member of the superbasis set.

[0089] The number of columns is equal to the number of parameters that can be specified in one super-equation, while the number of rows is equal to the number of algebraic equations in one super-equation defined by Eq.

[13] .

[0090] On the other hand, the left side of Eq.

[12] , i.e., for each combination of ROI, time point, and scenario, {t, {q i}, {dq i All derivatives of the observations taken from {\displaystyle \mathbb {dt}} are contained in the vector Y. As in Eq.

[12] , Y is equal to the product of the matrix M with the vector of parameters P (Eq.

[14] ):

number

[14]

number

[0091] The optimization problem can now be stated as minimizing, by varying the elements of P (i.e., the search values ​​for the power parameters),

number

[15] .

[0092] When the SVD method is adopted, the matrix M (real, rectangular) can be expressed as a matrix product: M=UWV T

[16] During the ceremony, W is a diagonal matrix with values ​​w ii are the singular values ​​of the matrix M ·U and V are column-normal matrices.

[0093] When minimizing

[15] by SVD, the parameter vector P is

number

[17] where U and V are defined by Eq.

[16] , w is a vector of the diagonal elements of W, also defined by Eq.

[16] , and division by a vector represents element-wise division.

[0094] To keep the parameterization as reliable and meaningful as possible and to prevent erroneous parameterization, two adjustments were performed.

[0095] The first one is only used in the case of SVD, the second one is also implemented when the parameterization is performed using other optimization routines: When using SVD in the minimization

[15] , the rank of the SVD (i.e., the dimension of W) is preferably minimized as follows: SVD is first performed with maximum rank to determine the vector w with all singular values. Then, the rank of the SVD is reduced to some low relative threshold, e.g., the number of singular values ​​that is greater than 5% of the maximum singular value, and SVD is performed again. This effectively reduces the rank of the SVD to the minimum required to satisfactorily describe the noisy experimental data.

[0096] On the other hand, irrespective of the minimization / optimization method chosen (SVD or any other), by restricting the selected parameters to positive values, also improper parameterization is prevented. Due to some physical meaning of the equations, some parameter values ​​must take positive values, for example those describing the mass retention translocation of the toxicant (central block of the central block row). The other elements in the central block row of the interaction matrix in principle allow the reduction or generation of the amount of the toxicant (by biodegradation, aggregation, adsorption, inactivation, passivation, and biosynthesis), and some of the interaction terms can compensate each other during the determination of the parameters, leading to incorrect parameterization, which later results in oscillations in the time propagation of the final observations. Such improper parameterizations often originate from the wrong sign of the parameters, so the maintenance of their positive values ​​is important. If a positive prescribed parameter (for example the relocation parameter in the central block of the interaction matrix) becomes negative after the first run adjustment 1), it is locked to 0 and the determination of the parameters using SVD is performed again. In practice, the whole determination procedure becomes stable within three iterations, even with very complex interaction matrices.

[0097] In Example 1, the parameters that are affected by the above approach of preventing improper parameterization by restricting values ​​to positive defined values ​​are shown as boxes in FIG.

[0098] i) Scenario-dependent fit to in vitro experiments The parameterization of the interaction matrix corresponds to the determination of the derivatives of the observations. However, the values ​​of the observations are determined from their corresponding derivatives using the corresponding initial conditions.

[0099] Once all interaction matrices are parameterized (all differential equation terms are specified numerically), the (initial) evolution of the observations is specified for each of the scenarios using the initial conditions as follows: if a mathematical model is used to describe the initial evolution of the in vitro model, the initial conditions may be equal to the experimental initial conditions. Conversely, if a mathematical model is used in the final step, where the observations are time propagated and the real in vivo situation is mimicked, the initial conditions may also differ from the experimental initial conditions.

[0100] The simplest method for determining initial values ​​for observations is to average the earliest data points for each corresponding observation (averaging over ROIs, if applicable), however initial conditions may also be obtained from an in vivo system (if applicable or measurable) or from available data sources.

[0101] An exemplary fit (corresponding to Example 1) to the initial time evolution of all the observations monitored in the in vitro model is shown in Figure 8. Here we present the measured time evolution (dots) of all the observations monitored in the lung in vitro model (epithelial cells + lung macrophages) for all scenarios (3 control + 3 exposure related) for the two materials together with their in silico fits (lines) to confirm the quality of the translation of the in vitro observed dynamics to the in silico parameterization.

[0102] j) Mechanism presentation The determined parameterization defines the contribution of the interaction matrix elements (i.e., the time derivatives in the system of differential equations) (a mechanism of action can be presented if desired). Since the method according to the invention determines how pairs of observations affect each other's value changes in time, the magnitude of the effect is parameterized by power parameters in the interaction matrix, and generalized interaction matrices contain a large number of such parameters, which are therefore difficult to investigate quickly.

[0103] To allow rapid exploration of the parameterization of the interaction matrix and easier identification of the dominant mode of action in individual material exposures, the present invention also optionally provides a method for mechanism presentation.

[0104] Using this step, key mode of action suggestions are provided to allow efficient comparison of mechanistic assessments between materials or individual material exposures. A preferred mechanistic suggestion (see Figures 9, 10 and 11) is defined as follows: The quantities are represented by the pictograms presented in Figure 9. Effects are indicated by triple-encoded arrows (see examples in Figures 10 and 11, which belong to the mode of action determined for exposures in Examples 1 and 2) - larger effects are coded by more opaque arrows with saturated colors, smaller effects are coded by thinner almost transparent arrows with less saturated colors, slight effects are barely visible due to almost clear transparency and gray, and the positive and negative signs of the effects are coded by green and magenta colors, respectively; The quantities that are affected (those on the left side of the differential equation) are in the inner circle and their pictograms are filled in grey, while the quantities that cause the effect are on the outer circle with open pictograms, · Self-stimulation and self-inhibition effects start with the same pictogram (of the quantity indicated by the self-stimulation or self-inhibition effect) and are indicated by a semicircular arrow targeting it.

[0105] Thus, Figure 9 shows the graphical symbols (pictograms) used to represent the elements of a mechanism report. The particular case shown corresponds to the co-culture of two cell lines (pulmonary epithelial cells and pulmonary macrophages) and the two events are chemokines 12 (CXCL12) and 3 (CCL3) excreted by epithelial cells and macrophages, respectively:

[0106] According to the scenario concept, it is clear that some of the blocks determine the function of the biological processor itself (biological control), while other blocks determine the response of the processor to exposure to agents (toxicants).

[0107] k) Dose biodistribution characterization Because animal-based disease prediction is based on prolonged observation of endpoints (symptoms, adverse outcomes, disease, etc.), determination of dose-related information, e.g., no observed adverse effect level (NOAEL dose) and lowest observed adverse effect level (LOAEL dose), is longer and more costly.

[0108] On the other hand, in the current invention, disease prediction is realized by in vitro monitoring of early-events coupled with in silico time propagation, which is obviously much faster and consequently much cheaper than the original animal-based tests. First, the observation of early-key events required for prediction is 10-30 times faster (compared to the time required to observe the endpoints required for prediction in animal-based tests). Second, the local biorelevance / biodistribution dose can be derived directly from the image of the exposed in vitro model in terms of dose distribution histograms within a single exposure experiment (opposite to animal-based experiments, where each different dose requires its own exposure experiment and subgroup of animals to correlate with disease induction efficiency).

[0109] In this invention, the dose biodistribution characterization (derived within an in vitro-in silico approach) thus relies on the experimental fact that under real conditions test agents (toxicants / materials / chemicals / etc.) cannot be delivered uniformly in biological systems, regardless of the delivery method. As early events at the cellular and subcellular levels are monitored and co-localized with the distribution of the test agent (toxicant), the method allows direct analysis of dose response from image-based dose distributions.

[0110] For that purpose, the dose biodistribution is evaluated as follows: Images from channels that deliver or are associated with information about a toxin are segmented (by one of the available algorithms for image segmentation) and a list of agent (toxin) related objects is created; Optionally, the list of segmented objects can be additionally modified by excluding objects based on their size or other descriptors, the latter being calculated by one of the available algorithms taking into account their intensity, dimensionality, aspect ratio, edge length / surface, skeletal information, mass or cross-sectional size, etc. Optionally, the object may be further divided into the object's surface (accessible agents) and the object's bulk (inaccessible agents) if required or assumed by the model; For each segmented object in the list, its dose equivalent is calculated as: Dose in terms of mass or surface area of ​​the agent (sum of intensities of pixels within the segmented object, if the intensity does not depend on the labeling / identification process - useful for e.g. Raman-based determination) Model-corrected dose in terms of mass or surface area of ​​the agent (if the intensity depends on the labeling / identification process - the sum of the number of pixels in the segmented object - valid for e.g. fluorescence-based or scattering-based determinations, the model requires a high degree of calibration for information density, scattering intensity, etc.) Dose in terms of interaction dose (sum of intensities of only those pixels that are in contact with the biological system of interest, which may be only the surface of the object, or may be part of the surface, e.g. surfaces that contact / colocalize with membranes, proteins, RNA, etc.) From the list of objects and their associated doses, a dose histogram is created showing the number of objects exhibiting a certain dose (bins, subranges), where the number of dose bins is defined based on the total number of objects in the analysis.

[0111] Since effect propagation depends on the local dose, multiple ROI images are used to determine a histogram of the dose distribution.

[0112] Various combinations of dose concepts are used in the predictive calculations: For barrier-like organs (with large surface vs. volume, e.g. lung or liver), surface (of material) vs. surface (of cells) local dose is used for materials and mass (of compound) vs. surface (of cells) is used for insoluble compounds; For voluminous organs (small surface vs. thick volume, e.g. muscle or brain), surface (of material) vs. mass (of organ / body) is used for material and mass (of compound) vs. mass (of organ / body) is used for soluble compounds; A different dose concept will be used for the submission of the final report (due to exposure-related issues): For barrier-like organs, this is recalculated as mass vs. surface for insoluble compounds, and later as mass vs. time (usually related to exposure time, since these organs may be exposed multiple times), For voluminous organs, for soluble compounds mass vs. organ / body mass is constantly recalculated to mass vs. time.

[0113] The above calculations of dose biodistribution are also shown in FIG.

[0114] Optionally, the observed dose (D) (for each bin of the submitted histogram) can be related to a dose rate (dD / dt), which is frequently measured in real-world environments and is consistent with the critical exposure time t c can be derived, t c =D / (dD / dt)

[18] This is required for a dose D to accumulate in the environment or organism at a measurement rate dD / dt.

[0115] l) Long-term propagation of observed values ​​- AO prediction After the in vitro-based parameterization has been fully derived, the final prediction of adverse outcome onset is calculated as the long-term time propagation of the propagation of initial observations following exposure to the material / chemical / drug / agent of interest.

[0116] In real organisms, long-term effects essentially involve systemic responses in addition to local tissue responses, so adverse outcome prediction should involve coupling of local tissue evolution, determined and parameterized by in vitro model early evolution monitoring (even the most complex ones), with systemic effects (including the interplay of in vivo responses with the organism's local tissues).

[0117] This coupling ultimately translates the evolution of initial local events into responses that can similarly propagate effects in the organism, making the final adverse outcome / symptom / disease predictions much more relevant. Such approaches employ complex knowledge from AOPs or mode of action databases, which collect events that critically define the evolution of adverse outcomes in terms of the organism's complex responses.

[0118] In the present invention, this knowledge is implemented by extending the interaction matrix (see the dotted area shown in FIG. 13). In addition to the already parameterized part of the interaction matrix defined through the in vitro system, the extended interaction matrix thus implements additional information based on the AOP represented by the local-to-systemic coupling (bottom horizontal part) and the systemic-to-local coupling (right vertical part). The additional part of the interaction matrix provides a means to couple local events to systemic events and vice versa. By definition, the former occurs within the in vitro model and the latter occurs outside the in vitro model. For example, if a cytokine is released within the in vitro model, it is considered a local event. However, if it leaks from a local tissue into the bloodstream, it is considered a systemic event outside the in vitro model and would be used to model the alveolar epithelium. As it is related to a local event, the present invention models it under the local-to-systemic coupling region of the interaction matrix. Note that such leakage would remain a local event if the in vitro model also included a vascular endothelial barrier. On the other hand, systemically released signaling molecules recruit new immune cells to enter the alveolar tissue (e.g., monocytes or PMN cells entering the lung from the blood in response to local immune cell attractant release). Such events are considered systemic, however, they directly affect local observations (number of immune cells in the local tissue). In the present invention, it is therefore encompassed by systemic vs. local coupling. Logically, the extended part of the interaction matrix is ​​relevant to adverse outcome prediction and depends on the in vivo-related observations defined at the beginning of this invention.

[0119] The interaction terms in the extended portion of the interaction matrix can have any form and can depend on any combination of observations from within the in vitro model as well as observations from outside the in vitro model.

[0120] Validation of the method according to the invention is carried out by comparing the results obtained with previously published data.

[0121] The invention provides a solution for the safety and / or toxicity prediction of potential toxicants, such as materials, chemicals, medicines, vaccines and similar substances and agents, which can predict adverse outcomes in advance with respect to their evolution in vivo within animal-based tests, thus making it (much) faster, more cost-effective and easily applicable to various experimental setups. EXAMPLES

[0122] The following examples are provided to illustrate various embodiments of the invention and its uses, but should not be construed as limiting the scope of the invention.

[0123] Example 1 - Comparison of methods and results published by Kokot et al. (2020) for predicting chronic inflammation with TiO2 nanotubes This first example aims to illustrate the ability of the invention to independently determine an initial mode of action and translate the latter into a safety assessment using only in vitro model monitoring. This example concerns long-term toxicity prediction for a material (TiO2 nanotubes), where mechanistic studies have been published in Kokot et al. Adv. Mater. 2020 (so the mechanisms revealed here can be directly compared to the published ones) and where in vivo data are available for validation (so predictions can be directly compared to actual in vivo data).

[0124] This example illustrates all the steps of the method according to a preferred embodiment of the invention (FIG. 2).

[0125] a) Associative identification of related observations: Based on published in vivo data related to inhalation of TiO2 nanotubes (H2020 project SmartNanoTox reports and publications, e.g., Adv Mater. 2020 doi: 10.1002 / adma.202003913, Toxicol. Appl. Pharmacol. 2020 doi: 10.1016 / j.taap.2019.114830), the in vivo-relevant observations for predicting chronic inflammation are: The surface of the epithelial cells that make up the alveolar surface, which are the first candidates to be affected during exposure; - Macrophage numbers, because after pulmonary exposure, monocytes enter the lung, differentiate into macrophages, and replace dead resident macrophages, and polymorphonuclear (PMN) cells enter the lung and their influx strongly correlates with the influx of monocytes. - Surfaces of isolated materials, since the process of isolation is a direct result of attractive interactions between many materials and many biomolecules (confirmed by MD studies and high-resolution in vitro and in vivo microscopy).

[0126] In addition, the in vivo relevant observations in this case (confirmed by a transcriptomics study of the expression of 50.000 genes for several material exposures) are also the following cytokines: · CCL3 - macrophage inflammatory protein (MIP)-1α / CCL3 - an inflammatory chemokine produced by cells during infection or inflammation, detected by Mouse CCL3 ELISA Kit (Proteintech, KE10023); Cxcl12 / CXCL12 - a constitutive chemokine involved in lung, brain, and arthritis, detected by Mouse CXCL12 ELISA Kit (Proteintech, KE10049); · CCL4 / CCL4, also known as macrophage inflammatory protein-1β (MIP-1β) - a CC chemokine with specificity for the CCR5 receptor, a chemoattractant for natural killer cells, monocytes, and various other immune cells, detected by the Mouse CCL4 ELISA Kit (Proteintech, KE10030)

[0127] b) Preparation of in vitro model A lung-mimetic in vitro model was constructed from lung epithelial cells (their surface represented by observable compartment 1 (Epi)-C1) and lung macrophages (their surface represented by observable compartment 2 (Imu)-C2) to reproduce the pulmonary alveolar tissue exposed to sub-3-micron sized particles / substances that are not cleared by the mucociliary escalator (the dimensions of the TiO2 nanotubes are 10 nm diameter and 100-200 nm length).

[0128] In vitro models that mimic the lung include:

[0129] [Table 4]

[0130] Cell lines are checked weekly for morphological changes, viability, and metabolic activity. Only cells that fall within the quality control range are allowed into the experiments.

[0131] c) Scenario-based time-lapse acquisition Methods for acquiring high-throughput in vitro time-lapse data: Fluorescence confocal microscopy using partitioning-based labeling and cell-transfection-based labeling with excitation / detection bands: 488 / 500-550+561 / 570-630+640 / 650-710nm; Confocal backscattering (dark field) microscopy with excitation / detection at 488 / 488 nm · ELISA for detection of CCl3, CCl4, and Cxcl12.

[0132] Additional auxiliary measures used to identify the observations: Helium ion microscopy (HIM), Scanning Electron Microscopy (SEM), Raman microscopy, Transcriptomics, Proteomics of the coronome.

[0133] Number of time points: 14 for microscopy and 5 for ELISA Number of ROIs in each scenario: 10 Number of slices per z-stack: 4 (3 micron height difference) Total duration of acquisition: 30 hours Observations used in this prediction:

[0134] [Table 5] Number of scenarios: 6

[0135] Scenario description:

[0136] [Table 6]

[0137] If applicable (Scenarios with exposure: EpiTox, ImuTox, EpiImuTox), exposure related data are as follows:

[0138] [Table 7]

[0139] The solution of the test material was added dropwise to the in vitro model to cover the entire surface of the in vitro model. The volume applied to the cells never exceeded 10% of the medium volume. The total exposure period was 30 hours, and all the tests were carried out at different time points within 30 hours. Different interaction doses were calculated from different regions of interest (ROI).

[0140] Representative examples of acquired images (fluorescence and backscattering microscopy) of this experiment are presented in Figure 14. Here, we show a portion of images from a time-lapse 3-channel (two fluorescence and one backscattering) monitoring of a TiO2-exposed in vitro lung model. As can be seen, these images and the choice of acquisition methods mentioned above allow the identification of objects and compartments, their segmentation, further quantification, and colocalization of events.

[0141] The fluorescent probes used in this particular experiment were: CellMask Orange for labeling all cells, · A cell line transfected with GFP to label only epithelial cells.

[0142] The nanomaterials were tracked by confocal backscattering microscopy.

[0143] d) Quantification of observed values ​​through image analysis Epithelial cell masks (C1) were created directly from the transfected cell line channel images by binarization (thresholding above 3x the average dark noise at cell-free locations). To eliminate errors from cell overgrowth, masks from different z-slices were down-projected (final mask pixel value equal to the union of the individual mask pixel values).

[0144] An immune cell mask (C2) was created from the mask corresponding to the CellMask Orange channel and the mask corresponding to the transfected cell line channel, again by binarization (thresholding), as distinct. To eliminate errors from cell overgrowth, masks from different z-slices were downprojected before subtraction (final mask pixel value equals the union of the individual mask pixel values).

[0145] A particle mask was derived from the backscattered image at the lower z-slice. The image of the top z-slice was used to derive a mask for material outside the cells (ToC0). The particle mask was segmented to identify objects with a diameter greater than 3 pixels. The list of objects was filtered by object size to exclude objects smaller than 1 micron. A new material mask was then constructed considering that only the masks of larger objects with a 3 pixel border were excluded. This is shown as ToCQ mask (isolation).

[0146] After subtracting the ToCQ mask from the original material mask, the remaining mask (ToC) was used to derive material masks within epithelial cells (ToC1) and immune cells (ToC2) by multiplying the ToC mask with the corresponding cell masks C1 and C2 from the previous step.

[0147] The intensity of material in a particular compartment was derived by summing the intensities of the pixels in the corresponding masks ToC0, ToC1, ToC2, ToCQ.

[0148] The process adopted is shown in FIG.

[0149] e) Generation of time series and time derivatives {{t j}, {q i (t j )}, {dq i / dt(t j The entire set of experimental data in the form of {{t j}, {q i-f (t j )}, {dq i-f / dt(t j )}}, {q i (t j )} to {q i-f (t j )} using Eq. 1. The original set of experimental data can be seen as dots on Fig. 5.

[0150] f) Construction of interaction matrix To define the time propagation of the vector of the observations (the definition of the observations in step c) from the time propagation definition in Eq. 10 above, the interaction matrix elements are constructed (Eq. 4 and the definition in Fig. 7) through the individual interaction functions (using the list of possible functions defined through Eqs. 6 and 7). The entire automatically constructed interaction matrix used in the prediction of TiO2 nanotubes is presented in Fig. 15 (due to its large size, the elements are wrapped, grid lines are added, and the power parameter p X,ij are already named after their function and block location within the 3 × 3 block of the interaction matrix).

[0151] Since TiO2 nanotubes are insoluble in nature, the interaction matrix can be further simplified - mostly for presentation purposes (see FIG. 17) - but can still be used in predicting TiO2 nanotubes. In this case, blocks D and B are nulled and some other minor simplifications are performed as well (see FIG. 16).

[0152] g) Basis functions and super-equation construction The basis functions and superequation were generated using Eq.12 above.

[0153] Here, an example of a basis function list (together with the corresponding time derivatives and power parameters of the observations), sorted by equation (row), is shown in Figure 18. This image shows the basis function list for a simplified version of the interaction matrix - note that the basis functions for the generalized interaction matrix are too large in size to be presented here.

[0154] h) Parameterization of interaction terms The additional intervention of the algorithm in the case of constraining the parameterization of the interaction matrix to positive and specified values ​​(if asserted) is indicated by a box. The other nulled terms correspond to the terms nulled by the definition of the interaction matrix.

[0155] If the interaction matrix is ​​simplified (some terms are nulled as shown in Figs. 16 and 17), the resulting parameterization of the generalized interaction matrix, which can also be used in the prediction of TiO2 nanotubes, is presented in Fig. 19B. The additional intervention of the algorithm in the case of constraining the parameterization of the interaction matrix to positive and specified values ​​(if stated) is again indicated by a box.

[0156] i) Scenario-dependent fit to in vitro experiments With the parameterized interaction matrix, the fits to the initial evolution of the observations for the in vitro monitoring of the alveolar model of lungs exposed to TiO2-nanotubes can now be plotted against time for all six scenarios. The plots are presented in Figure 8 (top and center rows correspond to TiO2-nanotube exposure-related scenarios).

[0157] j) Interaction mechanism presentation By using the graphical pictograms defined in Figure 9 (each pictogram represents a selected observation), the mechanism by which the observations interact with each other in the case of a pulmonary alveolar model exposed to TiO2-nanotubes is shown diagrammatically in Figure 10. The presentation of the mechanism is based on a block configuration and a parameterized interaction matrix (see results in step h) using the concepts previously described - briefly summarized here: Effects are indicated by arrows with triple encoding (transparency, color and darkness) - those that are almost transparent, grey and light are not significant The quantity to be affected is in the inner circle, and the quantity causing the effect is on the outer circle. · Self-stimulatory and self-inhibitory effects are indicated by semicircular arrows.

[0158] k) Dose biodistribution characterization In this example, intra-body distribution of toxin is determined via image analysis as local dose (here shown in units of toxin area using ToC0, ToC1 and ToC2 observations per cell area using C1, C2).

[0159] Using 10 ROI images from the lower z-slice, a histogram of the dose distribution is determined.

[0160] A local 3D stack (for one ROI) is used to calibrate the surface to volume concentration transformation.

[0161] In this example, surface-to-surface dose is used in the calculation, which is understood by all compartment definitions (Ci-cell surface, ToCi-toxin surface).

[0162] In this example, mass versus surface dose (in the lung) is shown.

[0163] In this example, the cell-related image channel provides the intensity (density) of the labeled membrane. Since it is dependent on the label, it must be separated from the labeling efficiency and all experimental factors in order to provide the required information about the local membrane surface. It is transformed taking into account: The required membrane surface / pixel dS / dNpx can be rewritten as the inverse of the product of the labeling efficiency dI / dS and the actual measured intensity histogram dNpx / dI dS / dNpx=1 / (dI / dS dNpx / dI) It can be shown experimentally that the labeling efficiency function varies very slowly with intensity due to the subdivision of the labels in the system (which always try to dilute them locally), and for this reason this function can be approximated as a constant, which can nevertheless be calibrated experimentally. Thus, the inverse of the intensity histogram used by the above equation represents the increase in surface with increasing intensity and generally defines the conversion from intensity to surface, which reflects the fact that at low intensities the surface is in fact independent of intensity due to underlabeling, whereas at higher intensities the surface increases strongly, since overlabeling is always avoided experimentally, and thus the latter reflects only the dense packing of local membranes (such as in the case of lipid bodies and local endoplasmic reticulum stacks); Using the above considerations, the inverse intensity histogram is empirically fitted with a transformed potential function with an exponent larger than 2 (lower intensity cutoff to avoid the use of noisy pixels) and this fit is used to convert the intensity to a surface using the above equation, and the gradient is calibrated through the experimentally derived average distance between packed membranes in the cell.

[0164] The material-related dose is calculated from the toxin-related image channel. Because it is a backscattered image, the intensity here can be approximated as proportional to the toxin (material) surface.

[0165] The local dose distribution is finally calculated for each pixel as the ratio between the local surface of the material and the local film surface.

[0166] An example of how a dose histogram is calculated is shown in FIG.

[0167] l) Dose-dependent long-term propagation of observed values For chronic-inflammation predictions related to inhalation of TiO2 nanotubes, as in this example, the interaction matrix is ​​finally extended by introducing: One local vs systemic binding (one new row, Ps1) showing that chemokine 3 is leaking out of the in vitro model (Pl2) into the bloodstream (there are already three local terms related to chemokines 3, 12 and their interference in the in vitro model), since the leakage depends on the formation of gaps between or within the epithelial cells, which are related to the toxic area within the epithelial cells (ToC1), the final form of the local vs systemic binding term is

number

number

[0168] Finally, the entire extended interaction matrix is ​​assembled (in vitro parameterized part + in vivo predefined local vs. systemic and systemic vs. local coupling). Values ​​of additional parameters (where:

number

number

[0169] This case (exposure for testing TiO2 nanotubes) is used for To prove that the present invention is able to independently reproduce the same mechanism of rearrangement (block R of the interaction matrix, Figure 10) representing the same circulation of nanomaterials published by H. Kokot et al., and To confirm accurate values ​​of local vs. systemic and systemic vs. local binding from in vivo data (PMN influx measurements on days 1 and 28 and monocyte influx measured in the first 4 hours).

[0170] Example 2 - Illustrates the possibility of investigating and interpreting differences in chronic inflammation induction mechanisms between metal-oxide nanotubes (for which mechanisms are known) and carbon nanotubes (for which mechanisms are not available but in vivo data have been verified).

[0171] After the parameterization of the interaction matrix is ​​performed for both materials -metal-oxide nanotubes (TiO2) and carbon nanotubes (MWCNT) in step h), the mechanism of adverse outcome induction is presented using the mechanism presentation concept from step j) in Figures 10 and 11 for TiO2 and MWCNT, respectively.

[0172] Using the same in vitro system and in vivo-related observations as in Example 1, and ignoring blocks G and M (which directly describe the in vitro system itself), the user first interprets the direct toxic effect: Toxin-regulated proliferation - Block T, · Toxic Relocation – Block R, Toxic regulated phenomena - Block S.

[0173] Block T represents the different kinds of toxicity of toxicants to all cell types used in the in vitro model, in this case the effect on the pulmonary alveolar epithelium of two types of nanomaterials with different sizes, chemical compositions, surface properties, etc., including direct toxicity (toxicants taken up inside / in a specific cell type), indirect toxicity (toxicants taken up in one cell type or compartment and affecting other cell types or compartments) and contact toxicity (toxicants are outside the cell and affect a specific cell type). A closer look at Figures 10 and 11 shows that metal-oxide nanotubes are slightly toxic to both cell types when taken up by macrophages (one direct and one indirect toxic effect). On the other hand, there is no such toxic effect in MWCNTs. They stimulate rather than harm the proliferation of some cells.

[0174] Block R describes the relocation of toxicants between cell types and compartments. A closer look at Figures 10 and 11 reveals that metal-oxide nanotubes accumulate gradually in epithelial cells, mostly sequestered. On the other hand, MWCNTs rather accumulate sequestered and weakly, but there is a balanced mix between accessible and sequestered (inaccessible) forms of these nanotubes.

[0175] Block S shows a phenomenon, in our case the release of two early cytokines CCl3 (Pl2) and Cxcl12 (Pl1). Normally they are excreted by macrophages (C2) and epithelial cells (C1), as can be seen from block M. Looking closely at figures 10 and 11, we see that after exposure, free metal-oxide nanotubes (indirect) as well as those internalized inside epithelial cells (direct) stimulate the excretion of Cxcl12 (normally, it is not excreted). On the contrary, MWCNTs inhibit the excretion of Cxcl12. In the case of CCl3, the situation is much more similar in both cases - note that this cytokine is always excreted under normal conditions.

[0176] The second step considers second-order effects and addresses: ·Phenomenon-regulated proliferation - Block C Phenomenon Interference - Block 2

[0177] Block C mimics the effect of cytokines on in vitro model proliferation. Under normal conditions, they maintain stable cultures. A closer look at Figures 10 and 11 reveals that upon exposure to metal-oxide nanotubes, CCl3 appears to inhibit proliferation of both cell types, while Cxcl12 stimulates proliferation of both cell types. In the case of MWCNTs, the effect of CCl3-based inhibition disappears.

[0178] Block 2 represents the quadratic effect of cytokines on cytokine release. A closer look at Figures 10 and 11 reveals that in our case, no differences were observed between the two exposures.

[0179] Finally, the most subtle effects can be seen if the interaction matrix is ​​allowed to include non-zero blocks B and D. In this case, these effects are: · Regulation of the phenomenon of toxic substance accessibility - Block B and · Toxin-accessible cell regulation - Block D

[0180] Block B describes the change in the accessibility of the toxicant due to the action of the cells themselves, e.g., biodegradation and bioinduced aggregation. A closer look at Figures 10 and 11 reveals that in our case, exposure to metal-oxides is not associated with a significant change in accessibility, whereas exposure to MWCNTs leads to a slight shift in accessibility between free and sequestered form (biological effect of form).

[0181] Block D leads to a change in the accessibility of the toxicant, in our case for cytokines, due to the action of the phenomenon. A closer look at Figures 10 and 11 reveals that in our case, for MWCNTs, a mixed but stronger effect appears, meaning that the latter interferes with biological signaling (adsorbed to or released from the nanotube surface).

[0182] Example 3 - Illustrating the possibilities of using different in vitro models in step b) Thus, the selection of an in vitro model will depend primarily on the adverse outcome likely to be induced by the agent, as well as the observations related to the events observed in vivo that lead to the identified adverse outcome. In vitro model selection may also similarly depend on the test substance, e.g., inhaled particles are tested in a lung-relevant in vitro model.

[0183] The simplest approach is to simply construct an in vitro model from cells that compose the organ affected by or associated with the adverse outcome or organ that is first exposed to the agent of choice, but in vitro models that are needed to address secondary exposures or more complex situations that are likely to be mechanistically plausible also exist.

[0184] For example, inhalation-based exposures logically lead to in vitro models that resemble the function of the lung. However, these may include various in vitro models, such as in vitro models mimicking the alveolar epithelial layer (air-liquid interface, exposed to smaller particles that enter the alveoli - Figure 3 - Section Lung, Epi cells), the immune system within the alveoli (Figure 3 - Section Lung, Imu cells), the bronchial epithelium (exposed to larger particles that do not reach the alveoli), any combination of these in vitro models (Figure 3 - Section Lung, Epi+Imu) and their further complexes with fibroblasts (Figure 3 - Section Lung, Epi+Imu+Fibro) (introducing many other characteristics to the tissue) (the ability to generate excessive amounts of extracellular matrix).

[0185] Inhalation-based exposure may also affect neural tissues, e.g. the olfactory barrier, which allows the direct transport of some substances into the central nervous system and is suspected to lead to diseases such as neurodegeneration with very high socio-economic impacts (Figure 3-section brain, Epi), which can be logically extended by immune cells of neural tissue - glial cells (Figure 3-section brain, Epi+Imu).

[0186] The immune system can be further enhanced by adding a variety of other immune cells to any of the in vitro models (Figure 3 - section Immune system, dendritic cells or T-cells).

[0187] To address secondary exposure, even with the inhalation route as the primary delivery route, the simplest in vitro model to address secondary exposure is the vascular endothelium with additional immune system components (monocytes, etc.). Naturally, more distant organs may be affected, leading to an in vitro model mimicking the epithelial layer in the kidney (Figure 3 - Section Kidney). Since some of the agent may accumulate in the organ, serious long-term health complications may occur, namely cancer - something that has a large societal burden, leading to in vitro models to address early changes in those systems (Figure 3 - Sections Breast, Umbilical Cord, Uterus).

[0188] Example 4 - Illustrating the various possibilities in the acquisition process of step c) The choice of methods used in the acquisition process depends on the type of toxicant as well as the type of observations identified in step a). Any combination of methods selected to perform step c) must allow the time-dependent detection of all desired specific events in the in vitro model selected in step b).

[0189] Figure 4 shows a (non-exclusive) exemplary list of such possible methods (described below), which are always combined with respect to the type of observations used in the acquisition process.

[0190] For observations related to cell properties (identity, shape, surface, dimension, structure, function, etc.), the following methods are preferred: Fluorescence microscopy (confocal or wide-field versions), using specific fluorescent labels to localize specific cell types or specific targets within cells; Fluorescence microscopy (confocal or wide-field versions) using partition-based fluorescent labeling to localize all organelles or compartments with the same physicochemical properties, e.g. the plasma membrane, all membranes, the cytoplasm, the actin cytoskeleton, mitochondria, etc. Bright-field (transmitted) light microscopy, when lower resolution is sufficient; Nonlinear microscopy (second or third harmonic generation) as a non-fluorescent label-free imaging method, When higher resolution is required, super-resolution fluorescence techniques such as stimulated emission depletion (STED) microscopy; If localization needs to be performed with even higher resolution, electron (SEM, TEM) and ion microscopy (HIM) (to resolve details finer than 10-30 nm), however in this case more advanced sample holder or chamber condition control concepts need to be used to allow time-lapse acquisition or imaging in physiological conditions.

[0191] When observations relate to a particular local environment, the following methods may be used: Fluorescence Lifetime Imaging (FLIM) Hyperspectral imaging (microspectroscopy) However, all of these require advanced image decomposition techniques.

[0192] When observations relate to the dynamics of local biocomponents or toxicants (diffusion, mobility, transport, etc.), high-speed time-lapse techniques are required, such as: Fluorescence Correlation Spectroscopy (FCS), Fluorescence Cross-Correlation Spectroscopy (FCCS), Time-lapse imaging with AI-based tracking and identification, etc. But all require advanced image / signal analysis.

[0193] For observations related to tracking toxicants or local cellular compartments / molecular / supramolecular complexes identified by specific chemical properties and related to their concentration, amount, mass, surface, charge, etc., different imaging techniques may be preferred: Fluorescence microscopy techniques for any toxicological object that is naturally fluorescent or can be fluorescently labeled prior to in vitro exposure experiments (some drugs and supplements, organic and inorganic particles, supramolecular complexes, e.g. delivery vehicles, vehicles for vaccines, viruses, aged plastics, etc. - Figure 21 shows four such systems - vaccines based on virus-like particles, micro- and nanoplastics, drugs and nutraceuticals), Backscattered optical microscopy for toxic objects (inorganic particles, some plastics, etc.) that are larger than 1 / 10 of the wavelength of light, are locally crystalline, and exhibit large electron density or period electron density profiles; For toxicants that cannot be fluorescently labeled (most small molecules) or are not naturally fluorescent (many drugs), Fourier transform infrared microscopy (FTIR), Raman microscopy (the most sensitive version is stimulated Raman microscopy), X-ray induced fluorescence (XRF), secondary ion mass spectrometry (SIMS), proton induced X-ray emission (PIXE) and similar micro- and hyperspectral microscopy; Electron (SEM, TEM) or ion microscopy (HIM) for all objects that are at least a few nanometers in size, however 3D localization is performed by additional edging or cutting techniques (e.g. microtome TEM, cryo-TEM, FIB SEM, etc.).

[0194] For observations related to specific (DNA) transcription or (protein) expression, various omics approaches and other specific detection techniques are used: Transcriptomics, Proteomics, Lipidomics, ELISA.

[0195] Example 5 - Illustrating the various possibilities in the quantification process of step d) Figure 5 provides an example of five possible quantification processes showing how images are processed and observations quantified. In this case, two observations relate to cell properties (cell surface, C1 and C2, of epithelial and immune cells, respectively) and three relate to toxin surface concentrations in different compartments (toxin surface in cell type 1-epithelial cells-ToC1, toxin surface in cell type 2-immune cells-ToC2, and toxin surface that is sequestered, i.e., inaccessible-ToCQ).

[0196] The examples provided illustrate the process used to quantify observations from: Image-derived masks z-stack derived mask A mask derived by multiplication, addition or subtraction of several masks Masks derived by object segmentation and filtering A mask derived by mathematical processing of all the above masks

[0197] Here, a first step is used to convert the z-stack of 3D cell location-fluorescence images into appropriate masks for epithelial cells (C1) and immune cells (C2). Two probes are used to distinguish between the two cell types - the first probe P1 identifies all cells (Cs) and the other P2 depicts only epithelial cells. Thus, the first mask C1 is derived as the maximum projection of the P1 z-image-stack and assigned directly to the C1 mask. The second mask C2 results as the maximum projection of the P2 z-image-stack from which the P1 z-image-stack is subtracted.

[0198] In the second step, the backscattered image BS of the material (toxin) is first thresholded to provide a total material mask (To). The mask is segmented into a list identifying larger objects (aggregates) of the toxin. The interface regions (surfaces) of these aggregates are then eliminated and a mask for the isolated toxin ToCQ is derived. A mask of all non-aggregated material is provided to the total material mask To as the complement of the isolated material ToCQ mask, and masks of the amount of toxin internalized in the corresponding cell types, ToC1 and ToC2, respectively, are derived as the product of the non-aggregated material mask with the specific cell type mask (C1 and C2).

[0199] After all the masks are created, the amount of toxin is derived by simply multiplying the toxin intensity image and the corresponding mask.

Claims

1. A system of differential equations is used to estimate at least two observations (q ｉ 1. A method for predictive testing of agents to predict the likelihood of an agent-associated adverse outcome based on the time propagation of the evolution of Observation value (q ｉ ) is any property that changes during a set of in vitro experiments, Each observation (q ｉ ) at a certain time point {t ｍ }, as determined experimentally in vitro; At least one observation (q ｉ ) is detectable in vivo or ex vivo following exposure to said test agent and can be causally linked to a known adverse effect that is associated with a key event that inevitably leads to a next event in the chain of events from exposure to an adverse outcome (AO); - at least one of said in vitro experiments comprises exposure of an in vitro model to a selected agent for which a prediction is sought, The observed value (q ｉ ) is analyzed, measured, and / or detected in a time-lapse manner, i.e., at least three time points, and each observation (q ｉ ) is obtained, In vitro detectable observations (q ｉ ) is propagated in silico over time for each observation (q ｉ ) is defined as the measurement evolution of the observations (q ｉ A set of parameterizable interaction terms I M that define a system of differential equations suitable for the required time propagation of ｉｊ This is achieved by converting ・Each interaction term IM ｉｊ is constructed from a predefined class of interaction functions and is associated with at least one other observation (q ｊ ) and the observed value (q ｉ ) changes, ・Interaction matrix [0010] The interaction term assembled into ｉ ), During the parameterization of the interaction matrix, the time points, the observations and their time derivatives at given time points are used as inputs. A method comprising:

2. The system of differential equations may be adapted to describe the evolution of the entire in vitro system or only a part of it, such as a scenario [0025] Using the concept of [0030] Scenario Vector [0045] The scenario vector is adjusted by selecting the interaction term associated with the particular experiment via multiplication by [0050] where u ｉ is a single value 0 or 1, and u ｉ A non-zero value of q is a measurable observation {q ｉ 2. The method for predictive testing of agents according to claim 1, wherein the method is used to select a

3. At least two in vitro detectable, in vivo relevant observations q ｉ The in silico predictions of [006] From the scenario [0070] The interaction matrix selected by [0080] The determinant defined by the interaction terms of [0090] 3. A method for predictive testing of agents according to claim 1 or 2, achieved by their propagation in time using a system of differential equations defined by:

4. The interaction matrix term IM ｉｊ is the power parameter p ｉｊ and the interaction matrix term I ｉｊ The observed value q ｋ Each interaction matrix element is composed of a product of at least two interaction functions IF that define the dependence on [0010] and the power parameter p ｉ，ｊ is the observed value q of each interaction term ｉ A method for predictive testing of an agent according to any one of claims 1 to 3, which determines the effect on the rate of change of

5. A method for predictive testing of agents according to any one of claims 1 to 4, wherein for the parameterisation of the interaction matrix, the inputs are: ##EQU00011##

6. The interaction function IF(q) is generalized to take a form independent of (a priori) known mechanisms, and the following dependence of the interaction term on the observed value q is assumed: ##EQU00012## where a, b, and c have simple integer values ​​0, 1, 2, ..., and o １ and ２ 6. A method for predictive testing of agents according to any one of claims 1 to 5, wherein the q-value is transformed into a function dIF / dq, which has integer values ​​of -1, 0, 1, 2 ..., defining the derivatives dIF / dq at small and large values ​​of q, i.e. determining the sensitivity of the interaction term to smaller and / or higher values ​​of the observed value q.

7. The sets (a, b, c) and (o １ , o ２ 7. The method for predictive testing of agents according to claim 6, wherein the interaction functions are related to the most biologically relevant interaction functions: Linear dependence on (1,1)-q, Saturation dependence on (1,0)-q, Threshold / delay dependence on (0,1)-q, - Saturation & threshold dependence on (0,0)-q, (0,2)-nonlinear / quadratic dependence on q, (-1,0)-a term that does not depend on q.

8. To determine the parameters, a linear combination of linearly independent equations is used. ##EQU00013## ［１２］ In the formula, p ｉ，ｊ Related to B ｉ，ｊ (the sum of the products of the interaction functions) is the observations {q ｂ 8. A method for predictive testing of agents according to any one of claims 1 to 7, wherein the basis functions depend on bonds of

9. The parameterization using Eq. [12] is ｉｊ 9. The method for predictive testing of agents as claimed in claim 8, wherein only the subgroup of equations of Eq. (a) appears, and the other parameters do not appear - they appear in the complement of this special subgroup of equations, the method being carried out on a subgroup of equations.

10. The system of equations according to claim 8 is solved using any suitable optimization routine, for example a linear or non-linear optimization routine, preferably using singular value decomposition (SVD), to find the optimal solution for the time point t ｍ , observation q ｉ (t ｍ 10. A method for predictive testing of agents according to claim 8 or 9, wherein the α- and β-values ​​and their corresponding derivatives are replaced by their measured values ​​or values ​​according to claim 8.

11. 11. A method for predictive testing of agents as claimed in claim 10, wherein SVD is used and the determined parameters are restricted to positive values.

12. A method for predictive testing of agents according to any one of claims 1 to 11, wherein monitoring of the in vitro test is carried out for up to 1 week, preferably up to 3 days, most preferably up to 50 hours.

13. 13. A method for predictive testing of agents as claimed in claim 12, wherein observations are taken at least three different time points and with any delay, the minimum delay being 5 minutes.

14. The method for predictive testing of agents according to any one of claims 1 to 13, comprising the steps of: a) a preliminary step for associatively identifying relevant observations; b) preparing an in vitro model for preparing key events of relevance and detrimental relevance to a biological system, preferably selected based on the expected function, mode of action and / or entry point of the biological system, wherein the in vitro model comprises at least one cell type; c) a scenario-based time-lapse acquisition for detecting the initial time evolution of said observations in said in vitro model in a minimum of two scenarios, with at least three time points at any time delay longer than 5 minutes, where at least one scenario represents an unexposed in vitro model and at least one scenario is for an exposed in vitro model, where a test substance is applied at a dose capable of inducing at least a minimal effect in the in vitro model; d) quantification of said observations to quantify their values ​​in an in vitro model, wherein at least one observation is indicative of a characteristic of the unexposed in vitro model and at least one observation is related to a characteristic associated with an exposure-related change in the in vitro model; e) generation of a time series of the observations and their time derivatives to obtain a vector of observations for the scenario, which defines the time evolution of the in vitro model and its constituent events; f) constructing the interaction terms of an interaction matrix with interaction functions to define all possible interactions and connections between the observables, both biophysically and biologically, which in turn mathematically determines the time evolution of all the observables; g) Construction of basis functions and super-equations to transform the set of interaction functions from the previous step into a mathematically orthogonal set of functions, which are used in the next step to numerically parameterize the evolution of the observations; h) parameterization of said interaction terms defining the rate at which the observations change, said parameterization determining a system of equations using scenario-derived experimental data, a previously defined orthogonal set of functions and a selected numerical method, allowing the numerical time propagation of said observations in a next step and the identification of optimal interaction terms in a second step. i) a scenario-dependent fit to in vitro experiments using the parameterization of the interaction matrix of the previous step to derive values ​​of each of the observations for each scenario starting from given initial conditions for each observation. j) identifying optimal interaction terms and optional presentation of determined mechanisms explaining the evolution of observed early events-observations for the test agent and their contribution to their development into potential adverse outcomes; k) optional dose biodistribution characterization, and / or l) Dose-dependent long-term propagation of observations for eventual use of the parameterized interaction matrix in propagating said AO-related observations over time to determine AO-predictions.

15. The interaction function IF(q ｉ , o １ , o ２ ) is the interaction matrix term I ｉｊ The observed value q ｉ and the interaction function IF(q ｉ )q ｉ Small values ​​of and q ｉ For large values ​​of １ and ２ The sensitivity through the interaction matrix term I ｉｊ is the power parameter p ｉｊ and at least two interaction functions IF(q ｋ ), where k is the index of row i or column j, and the interaction matrix term I ｉｊ 15. A method for predictive testing of agents as claimed in claim 14, wherein the α-value depends on the group / position of the observation.

16. Basis function B ｉｊ is the power parameter p ｉｊ and the associated interaction function IF(q ｋ ) and the sum of the products of the individual observations q ｉ 16. A method for predictive testing of active substances according to claim 14 or 15, comprising creating a basis for linearly independent functions capable of describing the time derivative of

17. A super-equation denotes a subgroup of differential equations, i.e., a group of rows of the interaction matrix, and has a power parameter p ｉｊ 17. A method for predictive testing of agents according to any one of claims 14 to 16, wherein only a subgroup of isoforms of isoforms of the same order are present and no other power parameters are present.

18. 18. A method for predictive testing of agents according to any one of claims 1 to 17, wherein the selection of observed values ​​depends on prior knowledge, e.g. based on AOPs, previous in vivo data and / or scientific reports and papers, regarding said test substance and possible substance-related outcomes or onset of symptoms.

19. 19. A method for predictive testing of agents according to any one of claims 1 to 18, wherein the observed value is a quantifiable property of a molecule, e.g. lipid, protein, RNA, DNA, etc., or a supramolecule, e.g. membrane, ribosome, cytoskeleton, fibre, vesicle, etc., or a cell, e.g. cell surface, volume, shape, activity, changes in cell compartment surface, volume, shape, etc.

20. 20. The method for predictive testing of agents according to claim 19, wherein the event is typically selected from the group comprising: - Changes in cell morphology, shape, and mobility; - cell aggregation and / or dissociation, - Changes in gene expression, Increased or decreased lipid expression, Increased or decreased transport of molecules and other structures to or from the extracellular space (vesicles, etc.), - Breaking of tight connections between cells, - actin and tubulin rearrangements, chromatin (de)condensation and other nuclear phase changes, ・Cell lysis, - apoptotic or necrotic events or events associated with other cell state changes; - alterations in the ER, endosomes, lysosomes, mitochondria and / or ribosomes; ・Cell death, Sequestration, phagocytosis or other new structure formation on the cell surface or inside the cell, e.g. sequestration is described by Kokot et al. - transcription, synthesis, expression, relocation and release of attractants from immune and other types of cells; - binding state, charge, interaction surface chemical composition (intrinsic or extrinsic) of molecules, molecular complexes and supramolecular structures, as well as aggregate structures; - the type of interaction exhibited by the interaction surface with any organelle or molecular complex, supramolecular structure or external / exogenous material; Release and / or absorption of molecules, vesicles, etc. from one intercellular / subcellular compartment to another; - surface area of ​​cells of the selected type (also corresponds to cell number or cell concentration, etc.); The surface area of ​​the toxin in the selected compartment (also corresponds to mass, concentration, volume, etc.) - the amount of signaling or transport molecules or other structures (corresponding to mass, concentration, binding state, etc. of cytokines, chemokines, enzymes, receptors, various types of RNA, exosomes, endosomes, lipid bodies, etc.); Surface area of ​​new or modified structures (also corresponding to mass, volume, density, lifetime, dimensionality, correlation length, shape descriptors, etc. of supramolecular structures, e.g. vesicular structures, cell or intracellular membranes, intracellular or extracellular fibrils, tight junctions between cells, intercellular gaps, mitochondria or mitochondrial networks, nuclei or subnuclear parts / domains, etc.).

21. The method for obtaining the observed values ​​is any suitable and widely used method, preferably a method for predictive testing of agents according to any one of claims 1 to 20, which is: - a microscopy method, such as bright field, fluorescence wide field or confocal, scattering, Raman, ESEM, FTIR, etc., that allows the observation and quantification of the time evolution of said observables (related to structural or biochemical information, see below) and the determination of their time derivatives, Spectroscopic methods, e.g. Raman, FTIR, UV-VIS, fluorescence, specific stains and similar, that allow the determination of the concentration / amount of compound classes of interest, e.g. proteins, lipids, aroma substances, labeled substances, tagged substances, inorganic substances, in the whole sample, and the quantification of their time evolution and the determination of their time derivatives; Immunological methods, such as ELISA, antibody isolation and purification, ELISPOT, immunoblotting, immunohistochemistry, immunoprecipitation, immune cell isolation, etc., that allow the determination of the concentration / amount of specific compounds of interest, e.g. selected cytokines, in the entire sample, and the quantification of their time evolution and the determination of their time derivatives; - Omics data, e.g. proteomics, lipidomics, transcriptomics, metabolomics, etc.

22. A method for predictive testing of agents according to any one of claims 1 to 21, wherein at least some of the observations are selected based on AOPs, previous in vivo data and / or scientific reports and papers.

23. 23. A method for predictive testing of agents according to any one of claims 1 to 22, wherein at least one of the observations must preferably correspond to an earliest event observed in vivo or in a patient if adverse outcome evolution is studied.

24. A method for predictive testing of agents according to any one of claims 1 to 23, wherein the observations q are sub-grouped into three classes: cellular compartments, amounts of agents associated with the compartments, i.e. toxicants, and amounts associated with the phenomenon.

25. 25. A method for predictive testing of agents according to claim 24, wherein when the toxicity assessment deals with inhalation and the lung is the target organ, e.g. since it is the first exposed, the observations related to chronic inflammation are associated with immune cells, e.g. - the number of monocytes that enter the lungs, - the number of polymorphonuclear (PMN) cells entering the lungs; -The number of white blood cells entering the lungs.

26. the selected in vitro model is relevant for the particular agent evaluation that mimics the relevant target tissue(s) as well as at least a relevant portion of the relevant agent delivery pathway; All cell types or combinations used are structurally or functionally mimicking an AOP-associated portion of said target tissue; and - capable of expressing one or more AOP-related significant events identified from prior knowledge; All cell types together can respond to the relevant early part of the AOP, which is - observed in vivo, preceding the later part of the AOP, and progressing to the observed adverse effects; and - Leads to at least one significant event that has the power to predict said adverse outcome; 26. A method for predictive testing of agents according to any one of claims 1 to 25, wherein said AOP represents a general concept of an adverse outcome pathway and includes all events that are causally linked from a molecular initiating event and all subsequent events to said final adverse outcome, irrespective of whether said events are physical, chemical, biological, biophysical, biochemical or any other type.

27. 27. A method for predictive testing of agents according to any one of claims 1 to 26, wherein the in vitro model comprises any combination of monocultures of cells or various co-cultures of cells or more complex cell cultures depending on the details of the test substance.

28. 28. The method for predictive testing of agents according to claim 27, wherein the cells for the in vitro model are selected from the group consisting of at least lung epithelial cells, skin epithelial cells, neurons, endothelial cells, muscle cells, intestinal epithelial cells, mucus cells, parietal cells, chief cells, endocrine cells, immune cells such as macrophages, glia, etc.

29. Quantification of the observations in control (no substance) and exposed (test substance applied at the selected dose) in vitro models depends on the type of data obtained: If the acquired data is already in numerical form, e.g. for ELISA, transcriptomics, and / or proteomics, it is used directly for derivative generation; If the acquired data is not in numerical form, e.g. if an image is acquired, the analysis is carried out in a two-stage method: Masking of parts of the image associated with said event-observations, e.g. cell line type, cell or cell structure labeling (if appropriate), method sensitivity, colocalization information, etc., and Quantification of the corresponding observations, where the mask from the previous step is transformed into quantified observations q at each time t and for each region of interest (ROI), typically the surface area is obtained by summing the number of pixels in the mask and the quantity is derived by summing the intensities of the appropriate image channel(s) within the mask.

30. 30. A method for predictive testing of agents as described in claim 29, wherein additional histogram or intensity distribution analysis (e.g. FIDA, fluorescence-intensity distribution analysis) is used to further enhance the quantification procedure.

31. 31. A method for predictive testing of agents as described in claim 30, wherein quantification results from multiple ROIs are summed.

32. The method for predictive testing of agents according to any one of claims 1 to 31, wherein the noise reduction mode of time derivative determination comprises the following assumptions: Each of said observations (significant events) is mostly slowly varying in time and has at most one maximum or minimum in said time interval of experimental monitoring; each of said observations has at most one segment, which means that it varies fast in said time interval of experimental monitoring and that the absolute value of the derivative of said observations has at most one maximum or minimum; Any other fast changes are due to experimental noise, And within a given time interval of experimental monitoring {t １ . . t ｍａｘ } time t ｊ Observation value q ｉ The measured value of {q ｉ (t ｊ )} (i.e., the vector, denoted by {}) is a set of fitted values ​​{q ｉ－ｆ (t ｊ )} (the additional index -f indicates the fit value), or the fitting {q ｉ (t ｊ )) to a predefined function, e.g., ##EQU14## Similarly, the differential value of the observation {dq ｉ / dt(t ｊ The time set of {t ｊ A set of differential values ​​{dq ｉ－ｆ / dt(t ｊ ))

33. A method for predictive testing of agents according to any one of claims 1 to 32, wherein the test agents are delivered in any manner, preferably homogeneously, into a biological system.

34. A method for predictive testing of an agent according to any one of claims 1 to 33, wherein dose biodistribution is assessed as follows: Images from the channels that deliver or are associated with information about said toxin are segmented (by one of the available algorithms for image segmentation) and a list of agent (toxin) related objects is created; Optionally, the list of segmented objects is additionally modified by removing objects based on their size or other descriptors, the latter calculated by one of the available algorithms taking into account intensity, dimensions, aspect ratio, edge length / surface, skeletal information, mass or cross-sectional size, etc. Optionally, the object may be further divided into a surface of the object (accessible agents) and a bulk of the object (inaccessible agents) if required or assumed by the model; For each of the segmented objects in the list, its dose equivalent is calculated as: Dose in terms of mass or surface area of ​​the agent (if the intensity is independent of the labeling / identification process, the sum of the intensities of the pixels within the segmented object - useful for e.g. Raman-based determinations) Models in terms of mass or surface area of ​​the agent - corrected dose (if the intensity depends on the labeling / identification process - valid for example for a sum of the number of pixels in the segmented object - fluorescence-based or scattering-based determination, the models require advanced calibration for information density, scattering intensity, etc.) Dose in terms of interaction dose (sum of intensities of only those pixels that are in contact with the biological system of interest, which may be only the surface of the object or may be part of the surface, e.g. surfaces that contact / co-localize with membranes, proteins, RNA, etc.) From the list of objects and their associated doses, a dose histogram is created (bins, subranges) showing the number of objects exhibiting a certain dose, where the number of dose bins is defined based on the total number of objects in the analysis.

35. A method for predictive testing of agents according to any one of claims 1 to 34, wherein multiple ROI images are used to determine said histogram of dose distribution.

36. A method for predictive testing of an agent according to any one of claims 1 to 35, in which various combinations of dose concepts are used for the calculation and / or presentation of the prediction, for example: For barrier-like organs with large surface to volume, e.g., lungs or liver, the surface of the material to the surface of the cells local dose is used for the material, and the mass of the compound to the surface of the cells is used for insoluble compounds; For large organs with small surface to volume and thick volumes, e.g. muscle or brain, the surface of the material to the mass of the organ / body is used for the material and the mass of the compound to the mass of the organ / body is used for soluble compounds; For barrier-like organs, the mass of compound versus the surface of the lung was used for insoluble compounds, later recalculated to the mass of compound versus time of exposure. For voluminous organs, for soluble compounds the compound mass vs. organ / body mass is constantly recalculated to compound mass vs. time of exposure.

37. 37. A method for predictive testing of agents according to any one of claims 1 to 36, wherein the already parameterised portion of the interaction matrix is ​​extended to carry additional biological response information at the level of tissue based on AOPs or other in vivo data sources representing local vs systemic and systemic vs local binding, and the additional portion of the interaction matrix provides a means to couple local events to systemic events and vice versa.

38. The method for predictive testing of agents according to any one of claims 1 to 37, wherein the end result describing the mechanism (mode of action) comprises any of the following statements: Agent T is highly toxic to cell type Ci; Agent T is not toxic to cell type Ci; Agent T exhibits contact toxicity to cell type Ci; Agent T is indirectly toxic to cell type Ci when taken up in cell type Cj; ・Local phenomenon Pl ｉ stimulates / inhibits proliferation of the cell type Cj; ・Cell type C ｉ increases / decreases the accessibility of agent T in compartment Cj, ・Cell type C ｉ induces aggregation / biodegradation of agent T in compartment Cj; - the agent T is efficiently taken up by the cell type Ci; Agent T is efficiently taken up by cell type Cj from cell type Ci; Agent T is sequestered; ・Phenomenon Pl ｊ increases / decreases the accessibility of agent T in compartment Cj, ・Phenomenon Pl ｊ induces aggregation / biodegradation of agent T in compartment Cj; Agent T, when present in cell type Cj, stimulates / inhibits event Pli; Agonist T stimulates / inhibits the phenomenon Pli when sequestered; ・Phenomenon Pl ｉ stimulates / inhibits the phenomenon Plj, ・Phenomenon Pl ｉ self-stimulates / self-inhibits its action.

39. 39. The method for predictive testing of agents according to claim 38, wherein the end result describing the biodistribution of an agent further comprises, for example, the following additional statements: Agent T is homogeneously dispersed in in vitro models of different sizes, demonstrating that the biological system prefers the dispersed form of Agent T and dislikes its aggregated state; Agent T is detected in large aggregates, indicating that the biological system prefers the aggregated state of Agent T; Agent T is efficiently sequestered on / in / beside / outside cell type Ci; - the agent T is not taken up by said cell type Ci, - The agent T is dispersed within the cell type Ci in the form of many small submicron sized particles.

40. The method for predictive testing of agents according to claim 39, wherein the final result provides a final adverse outcome prediction associated with the agent, further combining the statements of claims 38 and 39 with additional statements to further define the evolution of the biological system and / or provide other types of statements, such as the following statements: Agent T is safe – the biological system does not respond with abnormal evolution; Agent T is safe - Agent T does not cause delayed reactions; Agent T is safe at doses expected to be encountered in a particular condition; Agent T is safe at doses below a certain value; Agent T is safe at dose rates below a certain value; Agent T induces a weak / strong / very strong acute response above a certain dose or dose rate, which, however, dissipates within a specific time period; Agent T induces an acute response above a certain dose or dose rate, which at a certain different dose or dose rate can later be amplified and turn into a strong subacute / subchronic response and long-lasting chronic inflammation. Agent T caused immediate and severe cellular damage, reflecting its high toxicity. Agent T induces a weaker but longer-lasting response in terms of inducing a constant increase in the phenomenon / immune response; Agent T induces a weak acute response, however this later amplifies into ever increasing inflammation (cytokine storm) that cannot be resolved.

Images