Methods for assessing agent predictive tests for eliciting and inhibiting detrimental outcomes and / or diseases

Abstract

The present invention is a method for predictive testing of materials that can save the time to observe harmful outcomes (e.g., chronic inflammation) known to result in the development of most slowly developing diseases (e.g., cancer or autoimmune diseases). The present invention utilizes evolutionary observations of in vitro models to predict outcomes based on the earliest occurring early molecular events, far earlier than those molecular events exhibit long term action in tissue. These molecular events are analyzed, measured and / or detected for up to one week and computer-simulated time propagation is performed in order to correlate early events with long-term effects to make a reliable prediction of the safety / toxicity of the material. A computer simulation prediction of the variation of observable quantities over time is performed starting from an initial value using a differential equation set (formula (I)) constructed from interaction terms of an interaction matrix, where a scene (formula (II)) is used to select a relevant observable quantity to match the observable quantity observed in each in vitro experiment. # imgabs0 #

Description

Technical Field

[0001] The invention is in the field of research or analysis of materials, in particular by determining the biological effects they produce due to their chemical, physical or biological properties. The invention is also in the field of safety testing by determining biological events, and the field of timely information processing and calculation of these effects. The invention relates to a method for predictive testing of agents such as poisons, materials, chemicals, drugs and vaccines to evaluate the induction and inhibition of adverse outcomes and / or diseases. Background Art

[0002] Substances (agents) such as materials and particles, compounds, drugs or drug candidates, and vaccines must all be tested to detect any possible toxicity, harmful outcomes, or similar undesirable situations, of which slowly evolving harmful 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 exhibit immediate toxicity or strong acute inflammatory responses. That is, animals are exposed and monitored for several months to detect the long-term effects of substances (materials or compounds). The tests may be research-driven, focused on developing basic mechanistic knowledge of organisms, or used to answer questions of great practical significance, such as testing disease treatments, breeding, defense research, and toxicology, including materials, drug and vaccine safety, and cosmetic testing. It is estimated that more than 100 million vertebrates are used each year, from zebrafish to non-human primates. Mice, rats, fish, amphibians, and reptiles together account for more than 85% of research animals.

[0003] However, animal testing also has some problems. Because of its focus on detecting endpoints (harmful outcomes), it is slow, expensive, often unethical, and sometimes even irrelevant to human results because it cannot accurately reflect human results in some cases. For example, Greek and Menache (2013, doi:10.7150 / ijms.5529) pointed out that about 100 vaccines have been shown to prevent HIV infection in animals, but none of them work in humans. Conversely, 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 to the widespread use of materials and chemical safety testing is undoubtedly the long time to reach long-term or chronic disease endpoints. In materials and chemical safety testing, the months required make such tests useless because faster decisions are needed during the development process.

[0004] Therefore, there is a clear need for alternatives to animals, that is, alternatives to animal testing. Official agencies such as the European Centre for Validation of Alternative Test Methods (ECVAM) of the European Commission (EC), the Interdepartmental Coordination Committee for the Validation of Alternative Methods in the United States, ZEBET in Germany, and the Center for Validation of Alternative Methods in Japan promote the abolition of animal testing and respond to regulatory requirements.

[0005] These needs are the driving force behind the development of various test platforms, which are all upgraded cell culture systems. The most reasonable approach seems to be to develop a structural replica of the tissue and test it on it. Here, the principle behind it is the structural similarity between the original tissue and replicas such as organoids (organ-like test devices), "organ chips" (patent applications US2020408744, WO2020172670 and WO2009146911), "lung chips" (patent applications GB2585150 and WO2018217882), "intestinal chips" (patent applications US2020224136; US2020283732), and "skin chips" (patent application US2020239857). The system contains cells in a two-dimensional or three-dimensional system that simulates an organ. These chips can replace animals for in vitro disease research, drug testing, and toxicity testing. Typically, such chips are used for gene expression studies (patent applications US2020224136, WO2020172670), sometimes paired with cell morphology screening and determination of specific enzyme activities 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 / ) reported a new algorithm for drug discovery, in which the algorithm is arranged to screen gene expression pattern data for tens of thousands of known drug compounds and identify those that have the potential to restore disease state expression patterns to normal. These candidate compounds should then be further tested by cell-based assays to assess in vitro activity. In the final step, the compounds were successfully tested in cell-based assays and human organ chips to verify that they can effectively treat the disease without harmful side effects before entering clinical trials. While this represents a good alternative, the time required to test for possible harmful outcomes on such organ-on-chip systems is still similar to in vivo animal testing (e.g., patent application WO2020172670 describes methods for testing kidney chips that last more than a week, with some methods requiring even four weeks of exposure), meaning such testing remains expensive and long.

[0006] Another option is computer simulation and mathematical modeling, which aims to study and ultimately predict the toxicity and effects of drugs on 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.PMID20836040). This is achieved by using the latest advances in computing power to study test compounds at the molecular level, with the ultimate goal of creating unique treatments for each patient. A recent review (Mark TD et al. Computational Toxicology 21, 2022, https: / / doi.org / 10.1016 / j.comtox.2022.100213) gives the latest advances in computer simulation testing methods, in which predictions of hazards and exposure rely on QSAR and read-across methods. The main problem with this approach is the lack of biological testing, as the modeling may not take into account specific or more complex cellular strategies initiated after exposure to potentially toxic compounds. Such modeling can only rely on existing information and knowledge (e.g. AOPs) or derived from molecular dynamics simulations, which currently can only simulate the response of a small amount of cellular content. Computer simulations are expensive and time-consuming because they must be performed on supercomputers.

[0007] Considering the above, there is a need for an efficient, reliable, rapid, and economical method to test different materials and compounds that avoids the use of animals or isolated computer simulation modeling.

[0008] Prior art

[0009] Although many methods have been proposed and developed as alternatives to animal testing, some of which are described above, this section includes only exemplary methods in order to highlight the novelty of the methods used in the present invention.

[0010] Patent application WO0047761 discloses a method for analyzing the toxicity of chemicals, wherein wild-type and mutant strains of the same cell type (i.e., yeast cells, bacterial cells, cell lines of human origin) are used to detect oxidative stress, protein damage, cell cycle disorder, energy load and depletion, microtubule damage or the onset of metabolic capacity by overexpressing human gene inserts encoding metabolic genes or incorporating S9 parts. In a preferred embodiment of the present invention, the desired chemicals are administered to wild-type yeast and corresponding mutants, and the growth of yeast is measured using turbidity. Dose response curves are generated and the sensitivity (relative sensitivity) of the mutant to the compound relative to its parent is calculated. Statistically significant relative sensitivity indicates that the mutant is hypersensitive to the test compound. This method is very different from the present invention because no mutant strains or calculation of relative sensitivity are required.

[0011] Patent application WO2009146911 describes an organ chip device controlled by an independent sensor, which allows the establishment or maintenance of organs or organoids and stem cell microenvironments in a microchip format, suitable for online observation by live cell imaging and, for example, two-photon microscopy. It also discloses the use of the organ chip device in testing the activity, pharmacodynamics and pharmacokinetics of compounds, wherein the testing method comprises the following steps:

[0012] - Providing organ-on-a-chip devices,

[0013] - Add one or more test compounds to the organ / organoid on a chip,

[0014] - Evaluating the organ / organoid by microscopy and / or determining one or more parameters that can be determined using one or more sensors.

[0015] However, the document does not mention the details of the test method, parameters or expected microscopic changes in the organ / organoid. Nevertheless, it is clear that the possible effect of the test compound on the organ / organoid is only observed at one time point after exposure, without any in-depth analysis of the impact of this effect on the long-term effect, which is a key part of the present invention.

[0016] Murschhauser et al. (Communications Biology volume 2, Article number: 35; 2019) described a high-throughput microscopy method for single-cell analysis of event-time correlations in nanoparticle-induced cell death. Specifically, the method uses epithelial A549 lung cancer cells and Huh7 liver cancer cells as model systems to extract event times from the fluorescence time trajectories of cell death-related markers in automated single-cell array live cell imaging (LISCA). In paired marker combinations, the temporal order and delay time of lysosomal membrane permeability, mitochondrial outer membrane permeability, and oxidative burst after exposure to 58nm amino-functionalized polystyrene nanoparticles (PS-NH2 nanoparticles) were evaluated. The results showed that different cells showed heterogeneity and interdependence in signaling pathways under the action of different doses of nanoparticles. The solution does not make any predictions about the possible harmful outcomes of nanoparticles, but rather shows the mechanisms of signaling pathways and the responses of cells to different doses of nanoparticles. Although the authors anticipate that the high-throughput assay will be a powerful tool to study the mechanistic and temporal heterogeneity of cellular responses to different types of perturbations, they do not show how it can be translated into predictions of deleterious outcomes, let alone make such predictions.

[0017] In summary, all currently known material testing and toxicity assessment methods are based on the measurement or detection of a specific set of events that occur after exposure of cells and / or tissues and / or organ chip components. In contrast, the present invention aims to predict long-term toxicity based on early cellular events using time propagation. Therefore, the present invention upgrades early event monitoring by additionally building a mathematical model that automatically learns early mechanisms and propagates the evolution of observed early events into late harmful outcome predictions.

[0018] Patent application WO2018217882 discloses a microfluidic small airway chip (Airway-on-Chip) that is infected with one or more infectious pathogens (such as respiratory viruses) as a model for exacerbation of respiratory diseases (such as asthma exacerbation). After exposure, the expression of characteristic phenotypes of asthma in lung epithelial cells in the chip is analyzed. Various phenotypes have been observed, but their identification is based on knowledge of the mechanisms of asthma exacerbation. Although the invention represents a very complete method for observing early molecular changes induced by toxicants in advanced in vitro systems, it does lack a key step forward in that it cannot predict how these events evolve over time, which is exactly the problem that the present invention is intended to solve. Therefore, it cannot replace endpoint testing and can only monitor early evolution.

[0019] If one understands the mechanism of event propagation in advance and is able to automatically transform in vitro model monitoring into a mathematical description of event propagation, then time propagation can be accomplished using a model recently proposed by Kokot et al. (2020, doi:10.1002 / adma.202003913). In this work, the authors explored the toxicity of inhaled nanomaterials for the first time and reported the discovery of new molecular key events and their causal relationships, where the events described are the process of nano-quarantining and the circulation of nanomaterials between different lung cell types, which causes cells to fall into a vicious cycle of chronic inflammation. Similar cyclic events (loops) have also been observed in cases of Covid-19 pneumonia, as reported by Mould and Janssen (2021, Nature, Vol 590, 553).

[0020] Following the discovery of these mechanisms, a series of causally related events was described by a simplified theoretical model that describes the flow of nanomaterials between four separate compartments: i) extracellular, ii) intraepithelial, iii) sequestered in broccoli, and iv) intramacrophages. The model is defined by multiple parameters but can be simplified to three key descriptors:

[0021] a) The toxicity rate of nanomaterials to single cells (determined by measuring the number of surviving macrophages in MH-S monocultures after 4 days of exposure);

[0022] b) the rate at which epithelial cells sequester nanomaterials (calculated by measuring the fraction of nanomaterials in broccoli measured in LA-4 monocultures after 2 days of exposure), corrected to account for the toxicity rate of the nanomaterials to individual cells;

[0023] c) The efficiency of signaling and monocyte influx to replace dead macrophages (calculated by the influx of in vivo inflammatory cells (leukocytes) measured after at least 10 days), which is the time point when the chronic events of the reaction start to develop; the calculation includes corrections due to the rate of toxicity of the nanomaterial to individual cells (and due to the rate of sequestration of the nanomaterial by epithelial cells).

[0024] Based on the understanding of the mechanism and the corresponding mathematical formula (including simplified parameterization), the authors attempted to predict the chronic inflammation caused by nanomaterials through nanomaterial descriptors determined from three single time point measurements (two in vitro and one in vivo, both measured a few days after exposure). Using the above rates, the model was able to reproduce the in vivo time course of the amount of sequestered nanomaterials in broccoli, signaling the influx of immune cells, and the total number of macrophages, which can be used to predict the consequences of nanomaterial-specific acute to chronic inflammation. However, the paper does not explain how the time course of various in vitro and in vivo observable events can be translated into predictions of harmful outcomes themselves. In addition, the screening strategy in material safety assessment proposed by Kokot et al. is based on the understanding of the response of the organism to nanomaterial exposure from initial exposure to nanomaterials to potential harmful outcomes. This means that known molecular mechanisms need to predict the toxicity of the test materials as proposed, ultimately requiring a multidisciplinary approach, possibly combining advanced imaging, omics, particle labeling and tracking technologies and computer simulations applied in vivo and in vitro. Therefore, for each material, a lot of effort should be invested in discovering the mechanism of the material's effect on specific types of cells, which will undoubtedly lead to lengthy and expensive toxicity testing. Summary of the invention

[0025] According to the solutions of Kokot et al. and other references, the main disadvantage is that for each new material, the mechanism (mode of action) must first be discovered before the new observables and their interaction terms can be introduced into the model for predicting the toxicity of the material. Furthermore, the known methods based on in vitro and in vivo experiments are carried out in a destructive manner and only convey information about events at one point in time. Therefore, the purpose of the present invention is to solve these significant disadvantages and provide a significantly upgraded method for predictive testing of materials, chemicals, drugs, vaccines and similar compounds to assess their safety, possible adverse outcomes and / or the induction of diseases. The present invention upgrades the monitoring of early events by converting the propagation of early events observed in vitro into a mathematical form, which is then used to automatically learn early mechanisms by propagating the evolution of early events into predictions of late adverse outcomes. The technical problem is solved by the method described in the independent claim, while the preferred embodiments of the method are defined in the dependent claims.

[0026] The main advantage of the present invention is that there is no need to understand the mechanism (mode of action) of the test compound, and the non-destructive time-delay monitoring of the in vitro living model is converted into a mathematical description of the evolution of observable variables (key cell events), and the latter is converted into a prediction of the harmful outcomes observable in vivo. The method essentially allows the automatic discovery of the mode of action of the test agent, such as particulate matter on materials, chemicals, drugs, vaccines or similar substances or cells. The present invention also allows the determination of the systemic effects of the tested substance. The material that can be tested by the method of the present invention can be selected from the group consisting of materials, nanoparticles, vaccines, compounds, chemicals, drugs, etc. The present invention is not intended to be limited by a specific test agent or a specific cell that tests the agent.

[0027] For a better understanding of the subject matter, the meanings of the following expressions used in the text are explained below.

[0028]

[0029]

[0030] Most importantly, the alternative testing method of the present invention saves time in correlating slowly evolving adverse outcomes with possible causative factors by replacing the traditional observation of late adverse outcomes with in vitro monitoring of the evolution of early key events and further propagating this evolution in computer simulations.

[0031] This time saving can be easily illustrated in the case of the development of chronic inflammation, which is known to further lead to the development of cancer or autoimmune diseases, causing a large number of deaths worldwide. In conventional animal experiments, the time from exposure to the appearance of the above-mentioned harmful outcomes is generally 14 to 28 days, and even up to 90 days. The present invention can achieve this goal in less than a few days, thus significantly advancing the time. That is, it first uses cells, cell cultures and / or chip-like devices to infer and parameterize the early evolution of biological systems, and then predicts the outcomes through computer simulation propagation. Therefore, the present invention is based on the evolution of molecular events that occur first, long before they manifest long-term effects in tissues.

[0032] In the methods of the present invention, these molecular events are analyzed, measured and / or detected in a time-delayed manner (at at least three time points) and computer simulated time propagation is performed, wherein material testing is only performed for up to 1 week, preferably up to 3 days, and most preferably up to 50 hours, in order to correlate early events with long-term effects, thereby making reliable predictions about the safety / toxicity of the material. Not all early molecular events are associated with long-term adverse outcomes; however, the computer simulated time propagation is arranged to identify only relevant early events.

[0033] Thus, the essence of the present invention is to perform computer simulation predictions of in vitro detectable and in vivo related observable quantities, which are derived from their initial values ​​using a system of differential equations. The evolution of the early observables detectable in vitro can be automatically transformed into a set of parameterizable predefined interaction terms that define the set of differential equations suitable for the desired time propagation. These equations are preferably expressed by the interaction matrix The interaction matrix is ​​defined by the interaction terms, which are determined in in vitro experiments using the concept of scenarios.

[0034] In a preferred embodiment, the time derivatives of the quantized observables are used to construct an interaction matrix that transforms the time derivative vector of the observables into The observable vector with a given scenario S = {u1,u2,...} Contact, where u i is a unit value of 0 or 1, indicating whether a single observable q is selected in scenario S i .

[0035] In another preferred embodiment, the interaction matrix term IM ij Including the power parameter p ij and at least two interaction functions IF ij , the interaction function IF ij The interaction matrix term IM is defined ij For the observable quantity q i dependency.

[0036] The method is not limited by the type of observable selected, its value and / or the method of determination, calculation or detection.

[0037] In a preferred embodiment, the method according to the present invention comprises three main steps, namely:

[0038] a) a preparation phase, preferably based on prior knowledge, AOPs, in vivo data and previous results obtained by the method, to identify relevant observables,

[0039] - In vitro experiments, including the following steps:

[0040] b) preparing an in vitro model to prepare a living system associated with a critical event and associated with a harmful effect, preferably selected based on the expected function, mode of action and / or entry point in the living body, wherein the in vitro model comprises at least one cell type,

[0041] c) scenario-based time-lapse acquisition to detect the early time evolution of said observables in said in vitro model, with at least 3 time points in at least two scenarios with any time delay of more than 5 min (corresponding to the fastest biological background changes) and a total duration of up to 1 week (preferably less than 2 days; corresponding to the longest interval over which the evolution of events can be reliably determined), where scenarios are subgroups of measurable observables, which are structural, chemical, biological, mechanical, physiological, immunological and similar events, where at least one scenario refers to an unexposed in vitro model and at least one scenario refers to an exposed in vitro model, where the test substance is applied at a dose capable of causing at least a minimal effect in the in vitro model (known in toxicology as LOAEL - Minimum Observable Adverse Effect Level),

[0042] d) quantifying observables to quantify their values ​​in the in vitro model, wherein at least one observable refers to a property of the unexposed in vitro model and at least one observable relates to a property that is associated with exposure-related changes in the in vitro model,

[0043] e) generating time series and time derivatives of said observables to obtain a vector of observables of said scenario, which vector defines the time evolution of the in vitro model and the included events,

[0044] - Mechanism determination, including the following steps:

[0045] f) constructing the interaction terms of the interaction matrix through interaction functions to define all possible interactions and couplings between observables biophysically and biologically, and thus mathematically determine the time evolution of all observables,

[0046] g) constructing basis functions and super equations to transform the set of interaction functions of the previous step into a set of mathematically orthogonal functions that can be used to numerically parameterize the evolution of observables in the next step,

[0047] h) parameterizing the interaction terms defining the rates of change of the observables, said parameterization using the experimental data derived from the scenario, the set of orthogonal functions defined previously and the chosen numerical method to determine the system of equations and to implement in a next step the numerical time propagation of the observables and to identify in a second step described below the most relevant interaction terms,

[0048] i) perform scenario-dependent fitting of the in vitro experiments to derive the value of each observed variable under each scenario, using the interaction matrix parameterized in the previous step and starting from the given initial conditions for each observable,

[0049] j) optionally present the identified mechanisms to identify the most relevant interaction terms (modes of action) and describe their contribution to the evolution of observed early events (observable values ​​of the test agent and their progression to potentially deleterious outcomes),

[0050] - Predicting the outcome based on the results obtained in the above steps, wherein the prediction may also optionally include the following steps:

[0051] k) alternative dose-biodistribution characterization to replace expensive animal-based determination of dose-related information, such as no observable adverse effect level (NOAEL) and least observable adverse effect level (LOAEL), preferably through in vitro image-derived biodistribution,

[0052] l) Dose-dependent long-term propagation of observables to ultimately determine AO ​​predictions using parameterized interaction matrices (defined modes of action) in long-term propagation of AO-related observables, long before AO evolves and is observed in animal experiments.

[0053] In the method described by Kokot et al., observables (events) are not grouped, the description of the model responsible for the propagation of events over time relies on predefined differential equations and interaction terms, using two separate in vitro measurements and one in vivo measurement to make predictions based on a known mechanism previously discovered for a known material. This mechanism then allows the creation of a nomogram-based parameterization that already takes into account the computer-simulated time propagation. It is not possible to apply predictions to materials for which a mathematical model has not yet evolved from the mode of action.

[0054] In contrast, the present invention does not require knowledge and understanding of the links between material toxicity mechanisms, cellular responses, and individual early cellular events, and is thus a vast improvement over the methods of Kokot et al. Whereas the predictions achieved by the methods of Kokot et al. can only be used for materials that evolve toxicity through the same pre-discovered mechanisms and all interaction terms of all observables associated with a predefined in vitro model, the present invention only requires knowledge of which observables are relevant in vivo and which are in vitro observables. In the present invention, the time propagation of key events (in an in vitro system) for outcome prediction is implemented through a conventional system of differential equations, which are then automatically parameterized using the concept of an interaction matrix, a set of scenarios associated with each experiment, for inclusion / exclusion of each observable, and a special derivation algorithm capable of converting the time derivatives of the observables into power parameters for each interaction matrix element (interaction term). BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The pharmaceutical predictive test method of the present invention will be further disclosed in detail based on non-limiting exemplary embodiments, examples and drawings, which show:

[0056] Figure 1 Schematic diagram of the comparison between the alternative test method implemented by the present invention and the animal-based endpoint-oriented test.

[0057] Figure 2 It is an overview of the key steps a)-l) of the method of the present invention.

[0058] Figure 3 is a non-exclusive list of possible in vitro models used in step b).

[0059] Figure 4 is a non-exclusive list of possible methods for time-dependent data acquisition in step c).

[0060] Figure 5 is an example of an image quantification process, where 2 observables are related to cell properties and 3 observables are related to the surface of the agent (poison) used in step d).

[0061] Figure 6 are possible interaction functions and their parameterizations used in step f), all biologically relevant and numerically applicable interaction functions are grouped in the dashed area and two additional interaction functions used for the approximation of the quotient interaction function are grouped in the dotted area.

[0062] Figure 7 is a generalized interaction matrix with 9 blocks, in which typical interaction term expressions and interaction functions used in step f) are indicated.

[0063] Figure 8 is an exemplary fit of the evolution of events during the early 30 h corresponding to two selected materials exposed to the alveolar epithelium, with all observables indicated by the labels in step i).

[0064] Fig. 9 is the image pictogram used in the mechanism demonstration of step j).

[0065] Fig.10 : This is a parameterized mechanism-based display of the effect of the TiO2 nanotubes used in step j), with triply coded significance, insignificant effects are shown as almost transparent, gray and light, a color change from green to magenta indicates a change from a positive effect to a negative effect, the affected amount is in the inner circle, and the amount causing the effect is in the outer circle, and self-stimulatory and self-inhibitory effects are represented by semicircular arrows.

[0066] Fig.11 is a parameterized mechanism-based demonstration of the action of the multi-walled carbon nanotubes used in step j).

[0067] Fig.12It is the image-based dose biodistribution calculation used in step k), which is explained by taking dual-channel image analysis as an example to determine the dose from the poison surface to the membrane surface.

[0068] Fig.13 is an extended interaction matrix where the additional parts are not defined by the in vitro system but by the AOP knowledge and other available in vivo relevant data as used in step 1) represented by the dashed area.

[0069] Fig.14 is a representative example of a time-lapse three-channel (2 fluorescence and 1 backscatter) image of an in vitro lung model exposed to TiO2 actually acquired as part of the data acquisition in step c).

[0070] Fig.15 is an example of an interaction matrix and interaction functions implemented as part of step f) for the prediction of chronic inflammation in an in vitro lung model exposed to TiO2 nanotubes.

[0071] Fig.16 is a simplification of the interaction matrix in step f), where nulled interaction terms are assigned zero, excluding, for example, biodegradation / biosynthesis (in block D), adsorption / desorption from phenomenon-related structures (in block B), growth interference between cell lines (cross terms in the upper left block of block G), indirect toxicity (cross terms in the upper center block of block T), and transfer / direct exchange of toxicants between cell lines (in the center block R).

[0072] Fig.17 is an example of a simplified interaction matrix and interaction functions implemented as part of step f) for the prediction of chronic inflammation in an in vitro lung model exposed to TiO2 nanotubes.

[0073] Fig.18 is an example of a list of basis functions and associated parameters corresponding to a simplified version of the interaction matrix for the prediction of chronic inflammation in an in vitro lung model exposed to TiO2 nanotubes for each of the six scenarios presented in step g).

[0074] Fig.19 Parameterizations for predicting the generalized interaction matrix (A) and the simplified interaction matrix (B) for TiO2 nanotubes, obtained by SVD showing the effect of the additional constraints in the box (positive definiteness of the parameters in block R).

[0075] Fig. 20is an example of the final results (step 1), including the dose distribution histogram, the 800 h long-term spread of all observables indicated with annotations, and the long-term prediction of chronic inflammation-related observables (macrophages) associated with an in vitro lung model exposed to TiO2 nanotubes, where the left and right figures (bottom) correspond to the minimum and maximum local doses in the dose histogram (top), respectively.

[0076] Fig.21 Examples of two- and three-channel microscopy images used to predict adverse outcomes for 3 different agents: virus-like particle-based vaccines, micro- and nanoplastics, drugs, and food supplements, where the cellular compartment is always shown in the green channel and the agent (possible poison) channel is indicated by the color of the text on the image. DETAILED DESCRIPTION

[0077] The main objective of the present invention is to provide a means to move from the current concept of toxicity testing, which relies on testing for adverse outcomes in animals, to new tests that can rely on the observation of early events that occur immediately after the molecular initiation event (the first response to exposure) and that occur much earlier than the adverse outcomes. Thus, the present invention can significantly reduce the time required to complete the test.

[0078] Figure 1 A comparison of the time required for testing is shown between known (animal-based) in vivo testing methods, where long-term outcomes are only analyzed 14, 28, or 90 days after exposure, while the methods of the present invention can detect early molecular events indicative of long-term outcomes earlier, for example up to 48 hours after exposure.

[0079] The present invention is a predictive test method comprising 12 clearly defined steps, some of which are optional and used to improve reporting to test users. The implementation of the entire method is exemplarily shown in Examples 1 and 2 in a non-limiting manner. Various implementations in the individual steps are shown in Examples 3, 4 and 5.

[0080] Figure 2 A preferred embodiment of the predictive testing method is shown, wherein the following steps are performed:

[0081] a) identify relevant observables based on existing knowledge, AOPs, in vivo data, and previous results obtained with the method,

[0082] - In vitro experiments, including the following steps:

[0083] b) preparing a suitable in vitro model based on the observables and knowledge identified in step a),

[0084] c) performing scenario-based time-lapse acquisition of the events in the in vitro model prepared in step b) at at least three time points with any time delay exceeding 5 min in at least two scenarios, wherein a scenario is a set of measurable observables, which are structural, chemical, biological, mechanical, physiological, immunological and similar events,

[0085] d) quantifying observables to quantify their values ​​in the in vitro model, wherein at least one observable refers to a property of the unexposed in vitro model and at least one observable relates to a property associated with an exposure-related change in the in vitro model, all of which are obtained in step c),

[0086] e) generating a time series and a time derivative of the observable quantity quantized in step d) to obtain a time evolution of the observable quantity in the scene,

[0087] - Mechanism determination, including the following steps:

[0088] f) constructing the interaction terms of the interaction matrix through the interaction functions to define all possible interactions and couplings between the observables biophysically and biologically, thereby mathematically determining the time evolution of all observables by providing the most general description of the relationships between the time derivatives of the observables obtained in step e);

[0089] g) constructing orthogonal basis functions and super equations for the interaction matrix in step f) using all combinations of the scenario-related variables monitored in step c);

[0090] h) parameterizing the interaction terms using the definitions in step g) and the time-dependent values ​​and derivatives of the observables in step e), the obtained parameters defining the rate of change of the observables, said parameterization using the experimental data derived from the scenario, the set of orthogonal functions defined previously and the selected numerical method to determine the system of equations and to implement in a next step the numerical time propagation of the observables and to identify in a second step described below the most relevant interaction terms;

[0091] i) using the initial values ​​of the observables from steps d) and e) and the time evolution from step f), for each scenario used in step c), using the parameterization determined in step h), performing a scenario-dependent fit to the in vitro experiment;

[0092] j) optionally performing a mechanism demonstration using the parameterization derived in step h); and

[0093] - Perform outcome prediction based on the results obtained in the above steps, wherein the prediction may optionally further include the following steps:

[0094] k) performing dose biodistribution characterization on the image obtained in step c) using image-based dose biodistribution analysis,

[0095] l) performing a dose-dependent long-term propagation of the observables determined in step a), the interaction matrix parameterized in step h), and the evolution of the observables derived in step j) for each dose in the distribution histogram determined in step k).

[0096] The final user-friendly and understandable results of the pharmaceutical predictive test method of the present invention are most importantly achieved by step j), step k) and step l).

[0097] The results of step j) describing the mechanism (mode of action) can be described by something like Fig.10 The image material of the part is to be represented and may include the following statements:

[0098] - Drug T for cell type C i Severe toxicity,

[0099] - Drug T for cell type C i Non-toxic,

[0100] - Drug T for cell type C i Shows contact toxicity,

[0101] - If the drug T is taken by cell type C j Absorption, then for cell type C i Indirect toxicity.

[0102] - Local phenomenon Pl i Stimulate / inhibit cell type C j growth,

[0103] - Cell type C i Increase / decrease agent T in compartment C j accessibility,

[0104] - Cell type C i Inducing agent T in compartment C j Medium aggregation / biodegradation,

[0105] - Drug T cells are type C i Effective absorption,

[0106] - Drug T cells are type C i Cell type C j Effective absorption,

[0107] - The potion T is being quarantined.

[0108] -Phenomenon Pl j Increase / decrease compartment Cj The availability of Chinese medicine T,

[0109] -Phenomenon Pl j Inducing agent T in compartment C j Aggregation / biodegradation in

[0110] - When present in cell type C j During the period of drug T stimulation / inhibition phenomenon Pl i ,

[0111] - When isolated, the agent T stimulates / inhibits the phenomenon Pl i ,

[0112] -Phenomenon Pl i Stimulation / inhibition phenomenon Pl j ,

[0113] -Phenomenon Pl i Self-stimulation / self-inhibition its effects.

[0114] The results of step k) describing the biodistribution of the pharmaceutical agent can be described by analogy to Fig.12 The image material of the part may be represented and may include some statement of step j) and / or the following statement:

[0115] - Agent T was uniformly dispersed in different sizes in the in vitro model, indicating that the biological system favors the dispersed form of Agent T but not its aggregated state,

[0116] - Agent T was detected in large aggregates, indicating that the biological system favors the aggregation state of agent T,

[0117] - Agent T is effectively sequestered in cell type C i above / inside / side / outside,

[0118] - Agent T is not detected by cell type C i absorb,

[0119] - Agent T is dispersed in cell type C in the form of many submicron particles i middle.

[0120] The result of step 1) provides a prediction of the final adverse outcome associated with the drug, which can be used similar to Fig. 20 The image material of the part can be used to represent the evolution of the biological system, and can combine the statements in steps j) and k), add additional statements to further define the evolution of the biological system, and / or provide other types of statements. It may include the following statements:

[0121] - Drug T is safe - biological systems have not shown any abnormal evolutionary response

[0122] - Elixir T is safe - Elixir T will not cause any delayed response

[0123] - Agent T is safe at doses that are expected to be encountered under specific conditions

[0124] - Drug T is safe at doses below a certain value

[0125] - Agent T is safe at dose rates below a certain value

[0126] - Agent T triggers a weak / strong / very strong acute response (above a certain dose or dose rate), but it subsides within a certain time

[0127] - Agent T triggers an acute response (above a certain dose or dose rate), which then amplifies (above a certain different dose or dose rate) into a strong subacute / subchronic response, which may turn into a long-term chronic inflammation

[0128] - Agent T instantly causes severe cell damage, reflecting its great toxicity

[0129] - In terms of the phenomenon of sustained increase / immune response, the immune response induced by drug T is weak but lasts for a long time

[0130] - Agent T triggers a weak acute response, but then amplifies into an unresolved persistent inflammation (cytokine storm)

[0131] The steps to implement the predictive testing method are as follows:

[0132] a) Identification of Correlation of Related Observables

[0133] The choice of observables depends on the substance being tested and prior knowledge of the development of likely outcomes (symptoms) associated with the substance.

[0134] For example, at least a portion of the observables may be selected based on the AOP, previous in vivo data, and / or scientific reports and articles. If the evolution of an adverse outcome is studied, at least one observable should preferentially match the earliest event observed in vivo or in a patient.

[0135] For example, when toxicity assessment involves inhalation and the lung is the organ of interest, since it is exposed first, observables associated with chronic inflammation are related to immune cells, such as:

[0136] - the number of monocytes entering the lungs (because they differentiate into macrophages, replacing dead resident macrophages),

[0137] - the number of polymorphonuclear cells (PMNs) entering the lungs (because they are most convenient to identify in vitro and they correlate well with the influx of monocytes),

[0138] - Number of leukocytes entering the lungs (because their time-dependent rate changes are most significant and also correlate well with PMN influx)

[0139] b) Preparation of suitable in vitro models

[0140] Therefore, the choice of in vitro model depends firstly on the observables needed to reproduce the relevant events in vivo, which are selected in step a). The latter may also depend on the test substance, e.g. inhaled particles are tested in an in vitro model relevant to the lung (bronchial or alveolar part, depending on the size of the substance), which in vitro model includes lung epithelial cells and / or lung macrophages and / or fibroblasts, and / or endothelial cells and / or other immune cells or other cell types.

[0141] Depending on the test substance, the in vitro model may comprise any combination of single cell cultures or various cell co-cultures or more complex cell cultures. Thus, the cells used for the in vitro model may 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, glial cells, ...), etc.

[0142] Those skilled in the art will be aware of appropriate relevant cell combinations for analyzing a particular substance.Growth media and growth conditions are optimized based on the selected in vitro model that mimics the relevant tissue and supports the growth of the selected cell combination.

[0143] Therefore, the identification of the observables (in step a)) directly influences the selection of the cell types that make up the chosen in vitro model. Therefore, the construction of the in vitro model includes the following requirements:

[0144] - the in vitro model selected is relevant to the specific drug being evaluated and simulates at least the relevant portion of the relevant target tissue and the relevant drug delivery pathway,

[0145] - Each cell type or combination used

[0146] ○ structurally or functionally mimic the AOP-related portion of the target tissue, and

[0147] ○ Ability to express one or more AOP-related key events identified from prior knowledge;

[0148] - All cell types can collectively respond to the relevant early part of the AOP

[0149] o precedes the later part of the AOP that is observed in vivo and develops into the observed deleterious effects, and

[0150] ○ Leading to at least one critical event that could predict an adverse outcome,

[0151] AOP refers to the broad concept of the pathway to an adverse outcome, including all causally related events from the molecular initiating event to the final adverse outcome, regardless of whether the event is physical, chemical, biological, biophysical, biochemical or other types.

[0152] In Example 3 and Figure 3 In FIG. 5 , a non-exclusive list of possible in vitro models relevant for various types of in vivo events (chronic inflammation, neurodegeneration, cancer and other slowly progressive diseases of various organs) for performing step b) is shown.

[0153] c) Scene-based time-lapse acquisition

[0154] After identifying observable factors (step a) and constructing appropriate in vitro models to enable simulation of molecular initiation events and their further early evolution (step b), methods for detecting such molecular events need to be identified and selected.

[0155] The method for acquiring the observables of interest in the in vitro model can be any suitable and widely used method, preferably in a time-lapse manner, with the temporal resolution required to capture the dynamics, typically on the order of minutes to hours, such as:

[0156] - microscopy, such as bright field, fluorescence wide field or confocal, scattering, Raman, ESEM, FTIR, etc., for the observation and quantification of the temporal evolution of said observables (related to structural or biochemical information, see below) and the determination of their temporal derivatives;

[0157] - Spectroscopy, such as Raman, FTIR, UV-VIS, fluorescence, specific staining, etc., similarly for the determination of the concentration / amount of compound classes of interest (e.g. proteins, lipids, aromatic substances, labelled substances, tagged substances, inorganic substances) in the whole sample, as well as the quantification of their time evolution and the determination of their time derivatives;

[0158] - Immunological methods, such as ELISA, antibody isolation and purification, ELISPOT, immunoblotting, immunohistochemistry, immunoprecipitation, immune cell isolation, etc., for the determination of the concentration / amount of a specific compound of interest (e.g. a selected cytokine) in the whole sample, as well as the quantification of its time evolution and the determination of its time derivative;

[0159] - Omics data, such as proteomics, lipidomics, transcriptomics, metabolomics, etc.

[0160] The observed early events (observables) can be any effect of the test substance on the cellular system, such as changes in quantifiable properties of molecules (e.g. lipids, proteins, RNA, DNA, etc.) or supramolecular (e.g. membranes, ribosomes, cytoskeleton, fibers, vesicles, etc.) or cells (cell surface, volume, shape, activity, etc.), where the events are typically selected from the group consisting of:

[0161] - changes in cell morphology, shape and mobility,

[0162] - cell aggregation and / or dissociation,

[0163] - Changes in gene expression,

[0164] - Increased or decreased lipid expression,

[0165] - Increased or decreased transport of molecules and other structures into and out of the extracellular space (vesicles, etc.),

[0166] - breaking the contacts between cells (tight junctions),

[0167] - Actin and tubulin rearrangement,

[0168] - chromatin condensation or decondensation and other nuclear phase changes,

[0169] - Cell lysis,

[0170] - apoptotic or necrotic events or events associated with other cell state changes,

[0171] - changes in the endoplasmic reticulum, endosomes, lysosomes, mitochondria and / or ribosomes,

[0172] - Cell death,

[0173] - sequestration, engulfment, or formation of other new structures on the cell surface or within the cell, such as the sequestration described by Kokot et al.,

[0174] - transcription, synthesis, expression, relocation and release of attractants in immune cells and other cell types,

[0175] - binding state, charge, chemical composition of interacting surfaces and aggregate structure (endogenous or exogenous) of molecules, molecular complexes and supramolecular structures,

[0176] - the type of interaction exhibited by any organelle or molecular complex, supramolecular structure or external / exogenous substance on the interaction surface,

[0177] - release and / or uptake of molecules, vesicles, etc. from one intercellular / intracellular compartment into another,

[0178] - the surface area of ​​the selected type of cells (which also corresponds to the number of cells or cell concentration, etc.),

[0179] - The surface area of ​​the agent (poison) in the selected compartment (also corresponds to mass, concentration, volume, etc.).

[0180] - the quantity of signaling or transport molecules or other structures (also corresponding to the mass, concentration, binding state, etc. of cytokines, chemokines, enzymes, receptors, various types of RNA, exosomes, endosomes, liposomes, etc.),

[0181] - Surface area of ​​the new or modified structure (also corresponds to mass, volume, density, lifetime, dimensionality, correlation length, shape descriptors, etc. of supramolecular structures (e.g. vesicular structures, plasma or inner membranes, fibers inside or outside cells, tight junctions between cells, intercellular spaces, mitochondria or mitochondrial networks, nuclei or parts / domains in nuclei, etc.)

[0182] Example 4 and Figure 4 A non-exclusive list of possible methods for performing the time-dependent data acquisition required for step c) is listed in.

[0183] d) Quantification of observable quantities

[0184] In in vitro models, the quantification of observables of control (no substance) and exposure (test substance applied at a selected dose) depends on the type of data acquired. If the latter is already in digital form (e.g. ELISA, transcriptomics, proteomics), it is directly passed to the next step of derivative generation, otherwise, for example when acquiring images, the analysis will perform a two-step approach:

[0185] - masking of image portions relevant to the event (observable), such as cell line type, labeling of cells or cell structures (when appropriate), sensitivity of the method, colocalization information, etc., and

[0186] - Quantify the corresponding observables, where the mask from the previous step is converted to a quantized observable q at each time t and each region of interest (ROI), where typically the surface area is obtained by summing the number of pixels within the mask and the quantity is obtained by summing the intensities of the appropriate image channels within the mask.

[0187] Optionally, additional histogram or intensity distribution analysis (e.g., FIDA, fluorescence intensity distribution analysis) can be used to further refine the quantification procedure. Quantification 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 (originating from local variability) in the analysis.

[0188] For illustration, 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, for which a binary mask of the corresponding cell / structure is required. In images with one cell type and a single expression event, such a mask can be easily derived by binarizing the image. In images of more complex cell cultures, such as co-cultures or cultures that can express multiple structures or biochemical events simultaneously, it is common to use multiple labeling (co-localized or non-localized) as well as morphological information (size, dimensions, shape, etc.) and information from z-stacks (analyzing localization above and below the image) to distinguish different events and obtain multiple masks from the same image. Preferably, when the quantity is related to a structure exhibiting certain geometrical or morphological properties, such as size, aspect ratio, density, volume or surface of the structure (bulk). o r su rfac e o fa s tructur e ), etc., the mask is segmented by one of the available image segmentation algorithms and each segmented object is analyzed according to the desired characteristics. Thus, the mask is reduced to contain only those objects (or parts thereof) that meet the desired criteria. For example, the mask of an isolated poison (e.g., a substance inaccessible to a cell sensing mechanism or a cell processing) is obtained by binarizing the intensity of the channel corresponding to the substance, then performing segmentation to find aggregates of a size greater than a threshold and finally removing the pixels of the mask corresponding to the surface of the segmented object.

[0189] In Example 5 and Figure 5 In this paper, five possible quantization procedures are proposed to illustrate, in a non-exclusive manner, how to process images and quantize observables.

[0190] e) Generation of time series and time derivatives

[0191] After quantifying the observables in all measured and analyzed time points, time series and temporal derivatives of the observables can be derived to provide time-dependent observables and time-dependent temporal derivatives of the observables. The latter is best performed in a noise-reduced manner, obtaining information from all available time points and ROIs to increase the certainty of the results due to limited field of view, cell migration during time-lapse monitoring, changes in labeling and labeling efficiency during monitoring, limited number of observed key events, noise in detection, uncertainty arising during mask processing and segmentation, etc.

[0192] The denoising method determined by the time derivative involves the following assumptions:

[0193] - Each observable quantity (key event) changes slowly over time, with at most one maximum or minimum value in the experimental monitoring time interval.

[0194] - During the experimental monitoring time interval, each observable can have at most one fast-changing segment, which means that the absolute value of the derivative of the observable can have at most one maximum or minimum,

[0195] - Any other rapid changes are due to experimental noise.

[0196] Taking this into account, at a given time interval {t i ..t max}, time t j Observable quantity q i The set of measured values ​​(i.e., vector, represented by {}) {q i (t j )} Available fitting value set {q i-f (t j )} (with the additional index -f indicating the fitted value), or by replacing {q i (t j )} is fitted to a predefined function such as the following for interpolation:

[0197] q i-f (t) = a + bt Exp[ct] [1]

[0198] Similarly, the time set of the derivative values ​​of the observable {dq i / dt(t j )} can be used at the same time point {t j The derivative value set {dq i-f / dt(t j )} to approximate.

[0199] In the subsequent mechanism determination steps, especially the formation and parameterization of the interaction matrix, the experimental data set {{t j},{q i (t j )},{dq i / dt(t j )}}, the time point, the value of the observable and its time derivative at the given time point are used as input

[0200]

[0201] f) Construction of the interaction matrix

[0202] Construct the interaction matrix, where the (long) time propagation of the observables qi is realized by using a system of differential equations, as shown below

[0203]

[0204] The interaction term defines the observable q i The rate of change, similar to the selected observable q i With other observables q j In essence, each interaction matrix element (the interaction term IM in Equation [3] ij ) is defined as the power parameter and q i and q j The product of the products of the interaction functions of the relevant observables, with the (i,j)-specific derivative o

[0205]

[0206] Definition, where the power parameter p i,j Define each interaction term on the observable q i As will be shown below, this product can generally involve more than two interaction functions, but they are always related to the indices of observables i and j.

[0207] In some cases, when the interaction function depends on the quotient of two observables q and r, such as in the case of concentration (amount of agent (poison) per unit volume of a compartment or surface area), it is desirable to transform this quotient interaction function into the product of two interaction functions, one of which depends on the first observable and the other on the second:

[0208]

[0209] As will be shown below, in practical cases the remainder O(q,r) represents less than 20% of the original value, and the above approximation will be considered satisfactory.

[0210] In order to generalize the interaction function IF(q) to a form that is independent of the (a priori) known mechanism, we assume that the interaction term has the following dependence on the observable q

[0211]

[0212] where a, b and c can have simple integer values ​​0, 1, 2, ... and can be transformed to o1 and o2, which have integer values ​​-1, 0, 1, 2, ..., and define the derivative dIF / dq at small and large values ​​of q, i.e. determine the sensitivity of the interaction term to small and / or large values ​​of the observable q. Because the sensitivity is now parameterized, it can either be fitted from in vitro data or predefined based on common sense.

[0213] The sets (a, b, c) and (o1, o2) in equation [6] are related by the following relation:

[0214]

[0215] The following six interaction functions (in Figure 6 The most biologically relevant interaction functions (influencing their rates) are represented by the dashed region in the figure and identified below by the number pairs (o1,o2):

[0216] -(1,1)-linear dependence on q,

[0217] -(1,0)-saturation dependence on q,

[0218] -(0,1)-threshold / delay dependence on q,

[0219] -(0,0)-saturation & threshold dependence on q,

[0220] -(0,2)-nonlinear / quadratic dependence on q,

[0221] -(-1,0)-terms that are independent of q,

[0222] The other two interaction functions (in Figure 6 The dotted area outside the dashed area in ) represents two functions used in the approximation of the interaction function with the quotient of the two observable quantities q and r mentioned in equation [5]:

[0223]

[0224] By functionalizing the interaction function using equation [6] and parameterizing its sensitivity using equation [4], the interaction term in equation [4] no longer needs to be defined in advance (as described by Kokot et al.), but can in principle be determined using time-dependent experimental data via an inverse problem solving procedure.

[0225] The scenarios mentioned above were used as a tool to include / exclude relevant / irrelevant observables in each experiment. This appropriately reduced the list of interaction terms in each equation that fit each scenario. Mathematically, this was achieved by rewriting the system of differential equations into matrix form,

[0226]

[0227] where a scenario is defined as a vector of zeros and ones, with dimension equal to the number of observables. Multiplying the interaction matrix by the scenario vector eliminates interaction terms in the differential equations that are not present in a particular experiment.

[0228] To illustrate, Figure 7 As shown, the observables q are subdivided into three categories: cell surface area (compartment), amount of agent (i.e., poison (compartment-related)), and frequency of occurrence of the selected phenomenon, which logically leads to the evolution of the interaction matrix, which subdivides the interaction matrix entries into a 3x3 block form, where each block of the interaction matrix is ​​associated with the following (see Figure 7 The block indicated in ):

[0229] (G)-Growth interference (T)-Toxic Regulated Growth (C)-phenomenon-regulated growth (D)-Cellular regulation of toxicant accessibility (R)-Toxic Transfer (B)-Phenomenal regulation of toxicant accessibility (M)-Phenomenon of cell regulation (S)-The phenomenon of toxicant regulation (2)-Phenomenon interference

[0230] In addition to the grouping of interaction terms, Figure 7 The most likely interaction function class associated with each interaction matrix group is also indicated using notation from equations [4] and [6]. However, the choice of interaction function class may vary from case to case and is optimized for the most general case of data. Note that some interaction terms are zero by definition; they are Figure 7 Appropriate positions in the general form of the interaction matrix shown in are indicated by zeros. The interaction terms that are left blank are mainly related to the algebraic definition of the total cell surface, the definition of inaccessible poisons, and the diagonal relocation block-related terms. Having more than one interaction term per block indicates that the terms between the block diagonal and / or upper triangle and / or lower triangle differ in form (also indicated by different fill colors later). The symbols associated with the interaction function involve the observable and 1 or 3 indices (in accordance with the definitions in equations [6] and [7]): the first index is always the index of the observable i or j, where i always represents the row index in the block row and j always represents the column index in the block column, the second index is the possible / expected derivative o1, and the third index is the possible / expected derivative o2; if the second and third indices are omitted or represented by x, they can take any possible integer values ​​and allow them to be determined throughout the parameterization process. Due to the transformation of the quotient-dependent interaction function explained above, multiple occurrences of the compartment size-dependent interaction function (F(Ci)) appear in the same interaction term in different forms (with different indices) (e.g., one occurrence acts according to the compartment-dependent normalization from a concentration-dependent class, another one acts according to other compartment-dependent interaction functions). In this case, the derivatives o1 and o2 are strictly defined, they correspond to the transformation rules and cannot be freely determined experimentally.

[0231] If the interactions between the observables (events) detected are not too complex, or the biological system being tested is not too complex, the interaction matrix can be further simplified by omitting or excluding additional parts or assigning them values ​​of zero (0). Fig.16 This situation was discussed further.

[0232] g) Formation of basis functions and super equations

[0233] The next step is to transform the system of differential equations into a set of terms, namely the linear product of the power parameters of the elements of the interaction matrix and the basis functions (the sum of the products of the interaction functions), in preparation for the automatic parameterization of the system (i.e., the determination of the power parameters). To achieve the latter, first the selected observable dq i Each derivative value of / dt (defined by the corresponding differential equation - the left side of equation [3]) is considered to be determined by the parameter p i,j (Combining equations [3] and [4]) A linear combination of the products of the weighted interaction functions.

[0234]

[0235] From the example of interaction matrix terms (see Figure 7 ), some blocks mix in interaction terms, and some parameters p i,j appears in more than one interaction term (see the set of functions provided in Example 1), making these terms linearly dependent. When determining the value of such a parameter with maximum sensitivity, equation

[11] must be rewritten as a linear combination of linearly independent expressions,

[0236]

[0237] Among them and p i,j Related B i,j (the sum of products of interaction functions) becomes the same as the observable {q b} Union-related basis functions, these observables appear in the interaction function of the black-written sum in equation

[12] .

[0238] In order to maximize sensitivity, the determination of the parameters by equation

[12] should be done in a subgroup of equations in which only the parameter subgroup appears, while the other parameters do not appear (they appear only in the complement of this particular subgroup of equations). This means that the other parameters can be determined independently by other equations (subgroups). Symbolically, each such subgroup of equations is called a superequation.

[0239] When solving such a super equation (parametric coupled algebraic equation system), it is also necessary to collect parameters and basis functions (mathematically, the union of the parameters associated with the equations coupled to a super equation and the union of the basis functions are replaced). Symbolically, this will be called the super parameter set and the super basis set.

[0240] Examples of hyperbasis function sets and corresponding parameters are as follows Fig.18 As shown in Example 1.

[0241] Note that the number of algebraic equations used to determine the set of hyperparameters (the set of parameters through a hyperequation) is defined by the product:

[0242] N aigebraic equations =N ROI x N time points x N scenaria x N cupled equations

[13]

[0243] h) Parameterization of the interaction term

[0244] Considering a typical experimental data set involving data from about 10 ROIs at at least 15 time points in each of 6 scenarios, and 1-4 coupled equations per superequation, the number of algebraic equations for each single task exceeds 100 or even 400 (e.g., in the case of parameters related to toxicant transfer). Since the number of parameters appearing in such algebraic systems never reaches such a high number, the algebraic systems are always overdetermined and need to be solved using maximum likelihood or least squares driven optimization procedures, singular value decomposition, or similar methods.

[0245] Here we show how to use singular value decomposition (SVD) to solve this system, due to its simplicity and computational efficiency. However, this does not exclude the implementation of other routines, such as linear or nonlinear optimization routines. While SVD is most computationally efficient, optimization routines can be implemented more directly by restricting the power parameter to a certain range, such as positive values ​​(see below).

[0246] Regardless of the method chosen, the optimization problem is first defined. To do this, the matrix M (not to be confused with IM) is constructed from the right side of equation

[12] using data on the values ​​of the observables and their derivatives, where:

[0247] - Each row represents a super equation with multiple terms, where {t,{q i},{dq i / dt}} are replaced by their fitted values ​​according to equation [2], which correspond to the specific combination of ROI, time point and scene, and

[0248] - Each column represents the equation

[12] and p i,j (i.e., the column) is associated with a basis function value, and the basis function is a member of the superbasis set.

[0249] The number of columns is equal to the number of parameters that can be defined in a super equation, and the number of rows is equal to the number of algebraic equations in a super equation defined by equation

[13] .

[0250] On the other hand, the left side of the set of equations

[12] , i.e., for each combination of ROI, time point, and scene, from {t, {q i},{dq i The derivatives of all observable quantities obtained from the equation are contained in the vector Y. According to equation

[12] , Y is equal to the product of the matrix M and the parameter vector P (equation

[14] ):

[0251] Y=MP

[14]

[0252]

[0253] The optimization problem can now be defined as minimizing by varying the element P (i.e. the value of the power parameter searched for)

[0254] ||MP-Y||2

[15] .

[0255] When using the SVD method, the matrix M (real, rectangular) can be expressed as a matrix product:

[0256] M=UWV T

[16]

[0257] in

[0258] -W is a diagonal matrix, whose value w ii are the singular values ​​of the matrix M

[0259] - U and V are column orthogonal matrices.

[0260] When SVD is used to minimize

[15] , the parameter vector P can be expressed as

[0261] P=V.(U T .Y) / w

[17]

[0262] where U and V are defined by equation

[16] , w is the vector of diagonal elements W, also defined by equation

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

[0263] To make the parameterization as reliable and meaningful as possible and to prevent ill-posed parameterization, two adjustments have been implemented.

[0264] The first one is used only in case of SVD, while the second one is also used when parameterizing with other optimizers:

[0265] When using SVD for minimization

[15] , the rank of the SVD (i.e., the dimensionality of W) is preferably minimized as follows: First, run the SVD at maximum rank to determine the vector w with all singular values. Then, reduce the rank of the SVD to the number of singular values ​​exceeding some low relative threshold, such as 5% of the largest singular value, and run the SVD again. This effectively reduces the rank of the SVD to the minimum required to satisfactorily describe the noisy experimental data.

[0266] On the other hand, regardless of the chosen minimization / optimization method (SVD or any other), ill-posed parameterizations can also be prevented by restricting the chosen parameters to positive values. According to the physical meaning of some equations, some parameter values ​​must be positive, such as the parameter describing the mass-preserving translocation of the toxicant (the central block of the central block row). Since other elements in the central block row of the interaction matrix allow in principle the reduction or production of the amount of the toxicant (due to biodegradation, aggregation, adsorption, inactivation, passivation or even biosynthesis), some interaction terms may compensate each other in the determination of the parameters, which may lead to a wrong parameterization and thus to oscillations in the time propagation of the final observable. Since such bad parameterizations usually stem from the wrong signs of the parameters, it is crucial to maintain their positive values. If a parameter defined as positive (such as the relocation parameter in the central block of the interaction matrix) turns out to be negative after the first run of adjustment 1), lock it to 0 and run the parameter determination using SVD again. In fact, even for extremely complex interaction matrices, the entire determination process becomes stable in less than 3 iterations.

[0267] In Example 1, the parameters affected by the above method are restricted to positive values ​​to prevent bad parameterization. Fig.19 Marked with a box.

[0268] i) Scenario-dependent fitting of in vitro experiments

[0269] The parameterization of the interaction matrix corresponds to the determination of the derivatives of the observables. However, the values ​​of the observables are determined from their corresponding derivative values ​​using the corresponding initial conditions.

[0270] As the entire interaction matrix is ​​parameterized (all terms in the differential equations become numerically defined), the (early) evolution of the observables is defined for each scenario using initial conditions as follows: When mathematical models are used to describe the early evolution of an in vitro model, the initial conditions may be equal to the experimental initial conditions. On the other hand, when mathematical models are used in the final step to time-propagate the observables to simulate a real in vivo situation, the initial conditions may also differ from the experimental initial conditions.

[0271] The simplest way to determine initial values ​​for observables is to take the average of the earliest data points for each corresponding observable (or average ROI if applicable). However, initial conditions may also be taken from an in vivo system (if applicable or measurable) or from an available data source.

[0272] An exemplary fit of the early time evolution of all observables monitored in the in vitro model corresponding to Example 1 is as follows Figure 8Here we show the measured time evolution of all observables (points) and their computer fits (lines) monitored in an in vitro model of the lung (epithelial cells + lung macrophages) for both materials under all scenarios (3 controls + 3 exposure-related) to confirm the quality of the translation of the in vitro observed kinetics into computer simulated parameters.

[0273] j) Mechanism display

[0274] The determined parameterization defines the contribution of the elements of the interaction matrix (i.e. the time derivatives in the system of differential equations), from which the mechanism of action can be revealed if desired. Since the method of the invention determines how pairs of observables influence each other's values ​​over time, where the magnitude of the influence is parameterized by the power parameters in the interaction matrix, the generalized interaction matrix contains a large number of such parameters and is therefore difficult to examine quickly.

[0275] To rapidly examine parameterizations of interaction matrices and more easily identify key modes of action in individual material exposures, the present invention also optionally provides a mechanism unraveling approach.

[0276] Through this step, a concept of presentation of the key modes of action is provided to allow for effective comparison of mechanistic evaluations between materials or individual material exposures. Fig. 9 , 10 and 11) are defined as follows:

[0277] - Measure Fig. 9 The pictogram shown indicates that

[0278] - Effects are indicated by triple-coded arrows (see Fig.10 and 11 , which belongs to the exposure-determined action mode of Examples 1 and 2), larger effects are coded by darker saturated color opaque arrows, smaller effects are coded by lighter low-saturated color almost transparent arrows, and insignificant effects are almost invisible due to being almost pure transparent and gray; the positive and negative signs of the effects are coded in green and magenta, respectively,

[0279] - The affected quantities (to the left of the differential equation) are in the inner circle and their pictograms are filled in gray, while the quantities causing the impact are in the outer circle and their pictograms are filled in white

[0280] - Self-stimulatory and self-inhibitory effects are represented by semicircular arrows that originate from and point to the same pictogram (the amount to which the self-stimulatory or self-inhibitory effect refers).

[0281] Fig. 9A graphic symbol (pictogram) for illustrating a mechanism reporter element is shown. The particular situation shown corresponds to the co-culture of 2 cell lines (lung epithelial cells and lung macrophages) and 2 phenomena, namely, chemokines 12 (CXCL12) and 3 (CCL3) secreted by epithelial cells and macrophages, respectively.

[0282] Based on the scenario concept, it is clear that some blocks determine the function of the bioprocessor itself (biocontrol), while other blocks determine the response of the processor to exposure to agents (poison).

[0283] k) Dose biodistribution characterization

[0284] Because animal-based disease predictions are based on long-term observations of endpoints (symptoms, adverse outcomes, disease, etc.), it takes longer and is more expensive to determine dose-related information such as no observable adverse effect level (NOAEL) and least observable adverse effect level (LOAEL).

[0285] On the other hand, the present invention achieves disease prediction by monitoring early events in vitro combined with computer simulation of time propagation, which is obviously much faster than the original animal-based testing and therefore much cheaper. First, the early key events required for prediction are observed 10-30 times faster (compared to the time required to observe the endpoint required for prediction in animal-based experiments). Second, the local bio-relevant / biodistribution dose can be directly derived from the image of the exposed in vitro model in the form of a dose distribution histogram in a single exposure experiment (in contrast to animal-based experiments, in which each different dose requires its own exposure experiment and animal subgroup to be associated with disease induction efficiency).

[0286] Therefore, in the present invention, the dose biodistribution characterization (derived from the in vitro-computer simulation method) relies on the experimental fact that under real conditions, no test agent (toxicant / material / chemical / etc.) can be uniformly delivered to a biological system, regardless of the delivery method. Since early events at the cellular and subcellular levels are monitored and co-localized with the distribution of the test substance (toxicant), this method can directly analyze the dose response of the image-based dose distribution.

[0287] For this purpose, the dose biodistribution was assessed as follows:

[0288] - segmenting (by one of the available image segmentation algorithms) the images from the channel delivering or related to the poison information to create a list of objects related to the drug (poison),

[0289] - Optionally, the list of segmented objects may be additionally modified by excluding objects based on their size or other descriptors, wherein the latter may be calculated using one of the available algorithms taking into account intensity, size, aspect ratio, edge length / surface, skeleton information, mass or cross-sectional size, etc.,

[0290] - Optionally, if the model requires or assumes, the object can be further divided into the surface of the object (accessible agents) and the body of the object (inaccessible agents),

[0291] - For each segmented object in the list, its dose equivalent is calculated as follows:

[0292] o Dose expressed as mass or surface area of ​​the agent (if the intensity is not dependent on the labeling / identification process, then sum the pixel intensities within the segmented object - valid for Raman-based assays, for example)

[0293] ○ Model-corrected dose expressed as mass or surface area of ​​the agent (summing the number of pixels within the segmented object if the intensity does depend on the labeling / recognition process - e.g. valid for fluorescence-based or scatter-based assays; the model requires advanced calibration of information density, scatter intensity, etc.)

[0294] o Dose expressed as interaction dose (summing the intensity of only pixels in contact with the biological system of interest; could be just surface objects, or even just part of the surface, e.g. in contact / co-localized with membranes, proteins, RNA, etc.).

[0295] - From the list of subjects and their associated doses, create a dose histogram showing the number of subjects exhibiting a certain dose (bin, sub-range); where the number of dose bins is defined based on the total number of subjects in the analysis.

[0296] Since the effect propagation depends on the local dose, multiple ROI images are used to determine the histogram of the dose distribution.

[0297] Various combinations of dose concepts are used in prediction calculations:

[0298] - For barrier organs (with a large surface area to volume ratio, such as the lung or liver), the ratio of local dose to surface (of the material) to surface (of the cells) is used for materials, and the ratio of mass (of the compound) to surface (of the cells) is used for insoluble compounds.

[0299] - For large organs (thicker, with a small surface area to volume ratio, such as muscle or brain), use the ratio of surface (of the material) to mass (of the organ / body) for materials and the ratio of mass (of the compound) to mass (of the organ / body) for soluble compounds.

[0300] Using different dose concepts when submitting the final report (due to exposure-related issues):

[0301] - for barrier-like organs the ratio of mass to surface (of the lung) for insoluble compounds, with subsequent recalculation as the ratio of mass to time (of exposure) (since these organs may be exposed multiple times, which is usually related to the exposure time);

[0302] - For soluble compounds in bulk organs the ratio of mass (of compound) to mass (of organ / body) was always recalculated as the ratio of mass (of compound) to time (of exposure).

[0303] The above calculations of dose biodistribution are also shown in Fig.12 middle.

[0304] Alternatively, the observed dose (D) (for each bin in the provided histogram) can be related to a dose rate (dD / dt) that is frequently measured in a real environment and a critical exposure time t c It can be obtained as follows

[0305] t c =D / (dD / dt)

[18]

[0306] The dose D is the amount required to accumulate in the environment or organism at a measured rate dD / dt.

[0307] l) Long-term propagation of observable quantities - AO prediction

[0308] After fully deriving the in vitro -based parameterization, the final prediction of the development of an adverse outcome is calculated as the long-duration propagation of the early observable quantity following exposure to the material / chemical / drug / agent of interest.

[0309] Because in real organisms, long-term effects essentially involve systemic responses in addition to local tissue responses, prediction of adverse outcomes should involve the coupling between local tissue evolution (which is determined and parameterized by monitoring early evolution in in vitro models (even the most complex)) and systemic effects (which encompass the interaction of the organism's in vivo responses with local tissues).

[0310] This coupling ultimately transforms the evolution of early local events into responses that propagate the effects of the organism and make the final adverse outcome / symptom / disease predictions more relevant. This approach employs sophisticated knowledge from AOP or mode of action databases that collect events that rigorously define the evolution of adverse outcomes from the perspective of the complex responses of the organism.

[0311] In the present invention, this knowledge is achieved by extending the interaction matrix (see Fig.13 ). Thus, in addition to the already parameterized parts of the interaction matrix defined by the in vitro system, the extended interaction matrix implements additional information based on the AOPs, represented by local-to-system coupling (horizontal part at the bottom) and system-to-local coupling (vertical part on the right). The additional parts of the interaction matrix provide a way to couple local events to system 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, when a cytokine is released within the in vitro model, it is considered a local event. However, when it leaks from the local tissue into the bloodstream, it is considered a systemic event outside the in vitro model that can be used to simulate the alveolar epithelium. Since it is related to a local event, the present invention models it under the local-to-system coupling area of ​​the interaction matrix. Note that if the in vitro model also includes the vascular endothelial barrier, this leakage will continue to remain a local event. On the other hand, signaling molecules released by the system may recruit new immune cells into the alveolar tissue (for example, monocytes or PMN cells will enter the lungs from the blood in response to the release of local immune cell attractants). This event is considered systemic, but it directly affects local observables (the number of immune cells in the local tissue). Therefore, in the present invention, it is covered by the system-to-local coupling. Logically, the expansion of the interaction matrix relies in part on the relevant in vivo observable variables related to the prediction of adverse outcomes, which are defined at the beginning of the present invention.

[0312] The interaction terms of the extended part of the interaction matrix can have arbitrary forms and can depend on any combination of observables internal to the in vitro model and observables external to the in vitro model.

[0313] The method of the present invention was validated by comparing the results obtained with previously published data.

[0314] The present invention provides a solution for the safety and / or toxicity prediction of possible poisons such as materials, chemicals, drugs, vaccines and similar substances and medicaments. It can predict harmful outcomes in vivo in advance based on the evolution of harmful outcomes in vivo in animal-based tests. Therefore, it is (more) rapid, more cost-effective, and easy to apply to various experimental settings.

[0315] Example

[0316] The following examples are provided to illustrate various embodiments and uses of the present invention, but should not be considered to limit the scope of the present invention.

[0317] Example 1 - Comparison of the prediction method and results of TiO2 nanotube chronic inflammation published by Kokot et al. (2020)

[0318] The first example is intended to illustrate that the present invention is able to independently determine the early mode of action and convert the latter into a safety assessment monitored by an in vitro model alone. This example is related to the prediction of long-term toxicity of a material (TiO2 nanotubes) for which mechanistic studies have been published in Kokot et al. Adv. Mater. 2020 (so the mechanism disclosed herein can be directly compared with the published mechanism) and in vivo data can be used for validation (so the prediction can be directly compared with real in vivo data).

[0319] This example illustrates all the steps of the method according to the preferred embodiment of the present invention ( Figure 2 ).

[0320] a) Correlation identification of relevant observables:

[0321] Based on published in vivo data related to inhaled TiO2 nanotubes (H2020 project SmartNanoTox reports and publications such as Adv Mater.2020doi:10.1002 / adma.202003913, Toxicol.Appl.Pharmaco.2020doi:10.1016 / j.taap.2019.114830), relevant in vivo observables for chronic inflammation prediction include:

[0322] - The surface of the epithelial cells that make up the alveoli are the first candidates to be affected during exposure,

[0323] - The number of macrophages, because after lung exposure, monocytes enter the lungs and differentiate into macrophages, replacing dead resident macrophages, while polymorphonuclear cells (PMNs) enter the lungs and their influx is closely related to the influx of monocytes,

[0324] - The surface of the isolation material, since the isolation process 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).

[0325] In addition, relevant in vivo observables in this case (confirmed by transcriptomic studies of 50,000 gene expression in response to several substance exposures) were the following cytokines:

[0326] - CCL3 - macrophage inflammatory protein (MIP)-1a / CCL3 - an inflammatory chemokine produced by cells during infection or inflammation, detected by mouse CCL3 ELISA kit (Proteintech, KE10023),

[0327] - Cxcl 12 / CXCL12 - a constitutive chemokine involved in inflammation of the lung, brain, and joints, detected by mouse CXCL12 ELISA kit (Proteintech, KE10049),

[0328] - CCL4 / CCL4, also known as macrophage inflammatory protein-1β (MIP-1β) - a CC chemokine specific for the CCR5 receptor, which is a chemoattractant for natural killer cells, monocytes and a variety of other immune cells, detected by mouse CCL4 ELISA kit (Proteintech, KE10030).

[0329] b) Preparation of in vitro models

[0330] To replicate alveolar tissue exposed to sub-3 micron sized particles / matter (TiO2 nanotubes with a diameter of 10 nm and a length of 100-200 nm) that are not cleared by the mucociliary escalator, the in vitro model of the simulated lung consists of lung epithelial cells (whose surface is represented by observable compartment 1 (Epi)-C1) and lung macrophages (whose surface is represented by observable compartment 2 (Imu)-C2).

[0331] In vitro models that simulate the lung include:

[0332]

[0333] Cell lines were checked weekly for morphological changes, viability, and metabolic activity. Only cells within quality control limits were used for experiments.

[0334] c) Scene-based time-lapse acquisition

[0335] Methods for obtaining high-throughput in vitro time-lapse data:

[0336] - Fluorescence confocal microscopy using partition-based labeling and cell transfection-based labeling, with excitation / detection bands of: 488 / 500-550+561 / 570-630+640 / 650-710nm,

[0337] - Confocal backscatter (dark field) microscopy, excitation / detection wavelengths 488 / 488 nm,

[0338] -ELISA is used to detect CCl3, CCl4 and Cxcl12.

[0339] Use additional supporting measurements to identify observables:

[0340] - Helium ion microscopy (HIM),

[0341] - Scanning electron microscopy (SEM),

[0342] - Raman microscopy,

[0343] - Transcriptomics,

[0344] - Proteomics of coronary tissue (coronome).

[0345] Time points: 14 for microscopy, 5 for ELISA

[0346] Number of ROIs (regions of interest) per scene: 10

[0347] Number of slices per z-stack: 4 (height difference 3 microns)

[0348] Total collection time: 30h

[0349] Observables used in this forecast:

[0350]

[0351] Number of scenes: 6

[0352] Scenario description:

[0353]

[0354] Where applicable (exposure scenarios: EpiTox, ImuTox, EpiImuTox), exposure-related data are as follows:

[0355]

[0356] 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 must not exceed 10% of the volume of the culture medium. The total exposure time was 30 h, and all tests were performed at different time points within 30 h. Different interaction doses were calculated based on different regions of interest (ROI).

[0357] Representative examples of images (fluorescence and backscatter microscopy) acquired in this experiment are shown in Fig.14 Here we present partial images of time-lapse 3-channel (two fluorescence and one backscatter) monitoring of an in vitro lung model exposed to TiO2. As can be seen, these images and the choice of acquisition methods presented allow the identification of objects and compartments, their segmentation, further quantification and colocalization of events.

[0358] The fluorescent probes used in this particular experiment were:

[0359] -CellMask Orange marks all cells,

[0360] - Cell lines transfected with GFP label only epithelial cells.

[0361] Tracking of nanomaterials by confocal backscatter microscopy.

[0362] d) Quantification of observable quantities through image analysis

[0363] The epithelial cell mask (C1) was generated directly from the transfected cell line channel image by binarization (at locations without cells, the threshold was 3 times higher than the average background noise). To exclude errors caused by cell overgrowth, masks from different z slices have been down-projected (the final mask pixel value is equal to the union of the pixel values ​​of each mask).

[0364] The immune cell mask (C2) was generated by binarization (thresholding) of the difference between the mask corresponding to the CellMask Orange channel and the mask corresponding to the transfected cell line channel. To exclude errors caused by cell overgrowth, the masks of different z slices were down-projected before subtraction (the final mask pixel value was equal to the union of the individual mask pixel values).

[0365] The particle mask has been derived from the backscatter image of the lower z slice. The image of the upper z slice has been used to derive a mask for extracellular material (ToCO). The particle mask has been segmented to identify objects larger than 3 pixels in diameter. The object list has been filtered based on object size, excluding objects smaller than 1 micron. A new material mask has then been constructed, considering only the masks of larger objects and excluding the 3 pixel borders. This is labeled the ToCQ mask (isolated).

[0366] After subtracting the ToCQ mask from the original material mask, the remaining mask (ToC) is used to derive material masks within epithelial cells (ToC1) and immune cells (ToC2) by multiplying the ToC mask with the cell masks C1 and C2 corresponding to the previous steps.

[0367] By summing the intensities corresponding to the pixels in masks ToC0, ToC1, ToC2, ToCQ, the intensity of the material within a specific compartment is obtained.

[0368] The process used is as follows Figure 5 shown.

[0369] e) Generation of time series and time derivatives

[0370] By using Equation 1, {q i (t j )} fit to {q i-f (t j )}, will be replaced by {{t j},{qi (t j )},{dq i / dt(t j )}} The entire experimental data set is converted to {{t j},{q i-f (t j )},{dq i-f / dt(t j )}}. The original set of experimental data is as follows Figure 5 As shown in the points.

[0371] f) Construction of the interaction matrix

[0372] To define the time propagation of the observable vector above (the definition of the observable in step c), the definition of the time propagation in Eq. 10) is obtained by a separate interaction function ( Figure 6 and 7 , constructing the interaction matrix elements (in Equations 4 and Figure 7 The entire automatically constructed interaction matrix for predicting TiO2 nanotubes is shown in Fig.15 (Due to the large size, the elements are wrapped and grid lines are added; the power parameter p x,ij The blocks have been named based on their function and location within the 3x3 block of the interaction matrix).

[0373] Since TiO2 nanotubes are virtually insoluble, the interaction matrix can be further simplified - mainly for demonstration purposes (see Fig.17 ) - but can still be used to predict TiO2 nanotubes. In this case, blocks D and B are left nulled, and some other minor simplifications are implemented (see Fig.16 ).

[0374] g) Construction of basis functions and superequations

[0375] The basis functions and super equations are created using Equation 12 described above.

[0376] Here, an example of a list of basis functions sorted by equation (by row) (along with the corresponding observables and the time derivatives of the power parameters) is shown in Fig.18 Note that this image shows a list of basis functions for a simplified version of the interaction matrix - the basis functions for the generalized interaction matrix are too large to be shown here.

[0377] h) Parameterization of the interaction term

[0378] Additional interventions made by the algorithm in restricting the parameterization of the interaction matrix to positive defined values ​​(if stated) are indicated by boxes. Other blanked terms correspond to those blanked in the definition of the interaction matrix.

[0379] When simplifying the interaction matrix (e.g. Fig.16 and 17 As shown in the figure, several terms are left blank), the parameterization result of the generalized interaction matrix is ​​as follows Fig.19 This result can also be used to predict TiO2 nanotubes, as shown in B. The additional interventions made by the algorithm when restricting the parameterization of the interaction matrix to positive defined values ​​(where stated) are again indicated by boxes.

[0380] i) Scenario-dependent fitting of in vitro experiments

[0381] With the parameterized interaction matrix, it is now possible to plot the fitted values ​​of the early evolution of observables in the in vitro monitoring of the alveolar model exposed to TiO2 nanotubes versus time for all six scenarios. Figure 8 As shown (top and middle rows correspond to relevant scenarios exposed to TiO2 nanotubes).

[0382] j) Demonstration of interaction mechanism

[0383] By using Fig. 9 The graphical pictograms defined in (each pictogram refers to a selected observable) and the mechanisms of interaction between observables in the alveolar model exposed to TiO2 nanotubes are presented in Fig.10 Described in schematic form.

[0384] The presentation of the mechanism is based on the parameterized interaction matrix (see the results in step h), using the block organization and the concepts explained before, which are briefly summarized here:

[0385] - Effects are indicated by triple-coded (transparency, color and intensity) arrows - those with almost transparent, gray and light colors have less significant effects

[0386] - The quantities affected are in the inner circle, while the quantities causing the effects are in the outer circle

[0387] - Self-stimulation and self-inhibition are represented by semicircular arrows

[0388] k) Dose biodistribution characterization

[0389] In this example, toxicant biodistribution was determined by image analysis as local dose (here in units of toxicant area, using ToCO, ToC1 and ToC2 observables, and C1, C2 per cell area).

[0390] Ten ROI images from the lower z-slices were used to determine the histogram of the dose distribution.

[0391] Local 3D stacks (on one ROI) were used to calibrate the conversion from surface to bulk concentration.

[0392] In this example, the calculations were performed using surface to surface dose, which is understood by the definition of all compartments (Ci - cell surface, ToCi - poison surface).

[0393] In this example, the dose to the mass (lung) is shown.

[0394] In this example, the cell-associated image channel delivers the intensity (density) of the labeled membrane. Because it is label-dependent, it must be decoupled from labeling efficiency and all experimental factors to deliver the desired information at the local membrane surface. The following points were considered during the conversion:

[0395] - The required membrane surface per pixel dS / dNpx can be rewritten as the product of the labeling efficiency dl / dS and the inverse of the actual measured intensity histogram dNpx / dl

[0396] dS / dNpx=1 / (dl / dS dNpx / dl)

[0397] - Experiments show that the labeling efficiency function changes very slowly with intensity due to the redistribution of labels in the system (they always try to dilute them locally), so the function may be approximately constant; nevertheless, this can be calibrated experimentally;

[0398] - The inverse of the intensity histogram used in the above equation represents the increase in surface with increasing intensity and mainly defines the transition from intensity to surface; it reflects the fact that at low intensities the surface is practically not intensity dependent due to insufficient labeling, while on the other hand the surface increases strongly at higher intensities; since over-labeling is always avoided in the experiments, the latter only reflects dense packing of local membranes (e.g. in the case of liposomes and localized endoplasmic reticulum stacks);

[0399] - The above considerations were used to empirically fit the inverse intensity histogram by a translated potential function with exponent greater than 2 (cutoff at lower intensities to avoid using noisy pixels), and this fit was used to convert the intensities to surfaces by the above equation, with a calibrated slope in the average distance between packed membranes in the cell obtained throughout the experiment.

[0400] The material-related dose is calculated from the poison-related image channel. Since it is a backscatter image, the intensity here can be approximately proportional to the surface of the poison (material).

[0401] The local dose distribution is finally calculated as the ratio between the local surface of the material and the local film surface pixel-wise.

[0402] For example, the dose histogram calculation method is as follows Fig.12 shown.

[0403] l) Dose-dependent long-term propagation of observable quantities

[0404] For the prediction of chronic inflammation associated with inhaled TiO2 nanotubes, as shown in this example, the interaction matrix was finally expanded by introducing the following:

[0405] - A local-to-system coupling (a new row, Ps1) represents the leakage of chemokine 3 from the in vitro model (Pl2) into the bloodstream (there are already three local terms related to chemokines 3, 12 and their perturbations in the in vitro model); because leakage depends on the formation of gaps between or within epithelial cells, which in turn are related to toxic areas within epithelial cells (ToC1), the final form of the local-to-system coupling term is

[0406]

[0407] - a system-to-local coupling (a new term in the C2 differential equation) that describes the PS1-induced migration of monocytes from the blood to the alveoli and the associated macrophage differentiation (increase in local macrophage surface, i.e., C2) (in vivo observations suggest that the systemic response is already initiated within 5-15 min); the additional term looks like

[0408]

[0409] Finally the entire extended interaction matrix (in vitro parameterized part + in vivo predetermined local-to-system and system-to-local couplings) is assembled. The values ​​of the additional parameters (here, pL2S1 and pS2L1 are taken from in vivo microscopy and in vivo data of mice exposed to TiO2 nanotubes for validation purposes). With the fully extended interaction matrix, the observables are time-propagated over a larger time scale (here 800 h - about 1 month) compared to in vitro monitoring (here 30 h - about 1 day) and plotted ( Fig. 20 The in vivo associated / predicted observable (in this case, the number of macrophages - Fig. 20 , bottom - blue line C2) of the maximum time spread (here 800h) ( Fig. 20 , top - blue dots, secondary axis).

[0410] This example (exposure to test TiO2 nanotubes) is used

[0411] - Demonstrate that the invention can independently reproduce the same transfer mechanism (block R of the interaction matrix, Fig.10 ), representing the same nanomaterial cycle published by H. Kokot et al.; and

[0412] - Validation of correct values ​​for local-to-systemic and system-to-local coupling from in vivo data (PMN influx measurements at day 1 and day 28 and monocyte influx measured during the first 4 hours).

[0413] Example 2 - Illustrate the possibility of exploring and explaining the mechanistic differences between metal oxide nanotubes (whose mechanism is known) and carbon nanotubes (whose mechanism is unknown but known in vivo data can be used to verify) in inducing chronic inflammation

[0414] After parameterizing the interaction matrix of the two materials - metal oxide nanotubes (TiO2) and carbon nanotubes (MWCNT) in step h), the concept is demonstrated using the mechanism in step j). Fig.10 and 11 The mechanisms leading to the harmful consequences of TiO2 and MWCNT are shown in FIG.

[0415] Using the same in vitro system and relevant in vivo observables as in Example 1, and ignoring blocks G and M that directly describe the in vitro system, the user first explains the direct toxic effects:

[0416] -Toxins regulate growth - Block T,

[0417] - Poison transfer - Block R,

[0418] -Phenomenon of poison regulation - Block S.

[0419] Block T represents different types of toxicity of the toxicant on all cell types used in the in vitro model, in this case the effects of two nanomaterials with different sizes, chemical compositions, surface properties, etc. on the alveolar epithelium, including direct toxicity (the toxicant is inside / taken up by a specific cell type), indirect toxicity (the toxicant is taken up by one cell type or compartment and affects other cell types or compartments), and contact toxicity (the toxicant is outside the cell and affects a specific cell type). A closer look Fig.10 and 11 It can be seen that metal oxide nanotubes are mildly toxic to both cell types when taken up into macrophages (a direct toxic effect and an indirect toxic effect). MWCNTs, on the other hand, have no such toxic effects. They stimulate the growth of certain cells without harming them. Block R describes the transfer of the poison between cell types and compartments. A closer look Fig.10 and 11It can be seen that metal oxide nanotubes accumulate slowly in epithelial cells, with the majority accumulating in the sequestered state. On the other hand, MWCNTs accumulate rather weakly in the sequestered state, despite a balanced mix between accessible and sequestered (inaccessible) forms of these nanotubes.

[0420] In our case, panel S shows the release of two early cytokines, CCl3 (Pl2) and Cxcl12 (Pl1). Normally, they are secreted by macrophages (C2) and epithelial cells (C1), as shown in panel M. A closer look Fig.10 and 11 It can be seen that upon exposure, free metal oxide nanotubes (indirectly) as well as nanotubes internalized into epithelial cells (directly) stimulate the secretion of Cxcl12, which is normally not secreted. In contrast, MWCNTs inhibit the secretion of Cxcl12. For CCl3, the two situations are even more similar - note that this cytokine is always secreted under normal conditions.

[0421] In the second step, secondary impacts are considered, as follows:

[0422] -Phenomenon-regulated growth-block C

[0423] - Phenomenon Interference-Block 2

[0424] Block C resembles the effects of cytokines on the growth of in vitro models. Under normal conditions, they remain stable in culture. A closer look Fig.10 and 11 As can be seen, CCl3 appears to inhibit the growth of both cell types when exposed to metal oxide nanotubes, while Cxcl12 stimulates the growth of both cell types. For MWCNTs, the inhibitory effect based on CCl3 disappears.

[0425] Block 2 represents the second-order effects of cytokines on cytokine release. Fig.10 and Fig.11 As can be seen, in our case there is no difference between the two exposures. Finally, the most subtle effects can be shown if the interaction matrix is ​​allowed to contain non-zero blocks B and D. In this case, these effects are:

[0426] - Phenotypic regulation of toxicant accessibility - Block B and

[0427] - Cellular regulation of toxicant accessibility - Block D

[0428] Block B describes changes in the accessibility of toxicants due to actions of the cell itself, such as biodegradability and bioinduced aggregation. A closer look Fig.10 and Fig.11As can be seen, in our case, exposure to metal oxides does not lead to any significant changes in accessibility, whereas exposure to MWCNTs causes a slight change in accessibility (biological effect in the form) between the free and isolated forms.

[0429] Block D conveys changes in the accessibility of the toxicant due to the action of the phenomenon (in our case, due to the action of cytokines). Fig.10 and 11 It can be seen that in our case the effects of MWCNTs are mixed but stronger, meaning that the latter can interfere with biosignaling (adsorbed to or released from the nanotube surface).

[0430] Example 3 - Illustrative of the possibility of using various in vitro models in step b) -

[0431] Therefore, the choice of in vitro model depends first on the adverse outcome that the drug may induce and the observable quantity that is related to the events observed in vivo that lead to the identified adverse outcome. The choice of in vitro model may also depend on the test substance, such as testing inhaled particles on an in vitro model relevant to the lung.

[0432] The most straightforward approach is to construct in vitro models from cells that constitute the organ affected or associated with the adverse outcome or the organ primarily exposed to the agent of choice. However, there are also in vitro models that need to address secondary exposures or more complex situations where these models are considered mechanistically plausible.

[0433] For example, inhalation-based exposures would logically lead to in vitro models that resemble lung function. However, these may include a variety of in vitro models, such as simulations of the alveolar epithelium (the air-liquid interface, which is exposed to smaller particles that enter the alveoli- Figure 3 - lungs, Epi cells), the immune system in the alveoli ( Figure 3 -lung fraction, Imu cells), bronchial epithelium (exposed to larger particles that do not reach the alveoli), any combination of these in vitro models ( Figure 3 - lung part, Epi+Imu) and its further complex with fibroblasts ( Figure 3 - Lung fraction, Epi+Imu+Fibro), which introduces numerous other characteristics to the tissue (ability to produce excess extracellular matrix).

[0434] Inhalation-based exposures can also affect neural tissues, such as the olfactory barrier, which is suspected to transport substances directly to the central nervous system, leading to diseases such as neurodegenerative diseases with a very high socioeconomic impact ( Figure 3 -brain parts, Epi), and these diseases can logically be transmitted through the immune cells of the nervous tissue - glial cells ( Figure 3-Brain part, Epi+Imu) for expansion.

[0435] The immune system can be further enriched by adding a variety of other immune cells to any in vitro model ( Figure 3 - part of the immune system, dendritic cells or T cells).

[0436] To address secondary exposure, even if inhalation is the primary route of administration, the most direct in vitro model to address secondary exposure is the vascular endothelial cells with additional immune system components (monocytes, etc.). Of course, more distant organs can also be affected, resulting in an in vitro model that simulates the renal epithelium ( Figure 3 -kidneys). Because some agents can accumulate in organs, they can present serious long-term health complications, such as cancer, and some can even impose a huge burden on society, leading to the need for in vitro models to address early changes in these systems ( Figure 3 - breasts, umbilical cord, parts of the uterus).

[0437] Example 4 - Illustrative of various possibilities for the acquisition process in step c)

[0438] The choice of method to be used in the acquisition process depends on the type of toxicant and the type of observables identified in step a). Any combination of methods selected to perform step c) must be able to detect all desired specific events in a time-dependent manner in the in vitro model selected in step b).

[0439] Figure 4 A (non-exclusive) list of examples of such possible methods is shown (described below). These methods are always combined according to the type of observables used in the acquisition process.

[0440] For observables related to cell properties (identity, shape, surface, dimensions, structure, function, etc.), the following methods are preferably used:

[0441] - Fluorescence microscopy (confocal or widefield) uses specific fluorescent markers to locate specific cell types or specific targets within cells;

[0442] - Fluorescence microscopy (confocal or widefield version) uses compartmentalization-based fluorescent labeling to locate all organelles or compartments with the same physicochemical properties, such as the plasma membrane, all membranes, cytoplasm, actin cytoskeleton, mitochondria, etc.

[0443] - Bright-field (transmitted) optical microscopy, when lower resolution is sufficient,

[0444] - Nonlinear microscopy (second or third harmonic generation) as a non-fluorescent label-free imaging method

[0445] - Super-resolution fluorescence techniques, such as stimulated emission depletion (STED) microscopy, when higher resolution is required;

[0446] - Electron microscopy (SEM, TEM) and ion microscopy (HIM), when localization with higher resolution is required (to resolve details finer than 10-30nm), more advanced sample holder or chamber condition control concepts need to be used to allow time-lapse acquisition or imaging under physiological conditions.

[0447] When the observables are related to a specific local environment, the following approaches can be used:

[0448] - Fluorescence Lifetime Imaging (FLIM)

[0449] - Hyperspectral imaging (microspectroscopy)

[0450] But both require advanced image decomposition technology.

[0451] When the observables are related to the dynamics (diffusion, migration, migration, etc.) of local biological components or toxicants, fast time-lapse techniques are required, for example:

[0452] - Fluorescence correlation spectroscopy (FCS),

[0453] - Fluorescence cross correlation spectroscopy FCCS,

[0454] - AI-based tracking and recognition time-lapse imaging, etc.

[0455] But both require advanced image / signal analysis.

[0456] Observables related to tracking poisons or localized cellular compartments / molecules / supramolecular complexes identified by specific chemical properties, as well as observables related to their concentration, amount, mass, surface, charge, etc., can be preferentially applied using various imaging techniques:

[0457] - Fluorescence microscopy for any toxicant that naturally fluoresces or can be fluorescently labeled prior to in vitro exposure (certain drugs and supplements, organic and inorganic particles, supramolecular complexes such as delivery vehicles, vaccine vectors, viruses, aged plastics, etc. Fig.21 Four such systems are shown - virus-like particle-based vaccines, micro- and nanoplastics, drugs and food supplements),

[0458] - Backscattered light microscopy, for toxic objects (inorganic particles, some nanoplastics, etc.) that are larger than 1 / 10 of the wavelength of light and are partially crystalline or exhibit a large electron density or periodic electron density distribution,

[0459] 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 spectroscopy (PIXE), and similar microspectroscopy and hyperspectral microscopy techniques for detecting poisons that cannot be fluorescently labeled (most small molecules) or that do not naturally fluoresce (many drugs);

[0460] - Electron microscopy (SEM, TEM) or ion microscopy (HIM) for all objects with dimensions of at least a few nanometers, but where the 3D localization is performed by additional edge or cutting techniques (e.g. microtome TEM, cryo-TEM, FIB SEM, etc.).

[0461] Observables associated with specific (DNA) transcripts or (protein) expression, using various omics approaches and other specific detection technologies:

[0462] - Transcriptomics,

[0463] - Proteomics,

[0464] - Lipidomics,

[0465] -ELISA.

[0466] Example 5 - Illustrative of various possibilities in the quantification process of step d)

[0467] Figure 5 Five possible examples of quantification procedures are shown, illustrating how to process images and quantify observables. In this case, two observables are related to cell properties (cell surface C1 and C2 for epithelial and immune cells, respectively), and three are related to toxicant surface concentrations in different regions (cell type 1 - toxicant surface in epithelial cells - ToC1, cell type 2 - toxicant surface in immune cells - ToC2, and toxicant surface that is sequestered, i.e., inaccessible - ToCQ).

[0468] The illustrated embodiment illustrates a process for quantifying an observable

[0469] - Image derived masks

[0470] - z-stack derived masks

[0471] - A mask obtained by multiplication, addition, or subtraction of several masks

[0472] - Masks obtained by object segmentation and filtering

[0473] - The mask obtained by mathematically processing all the above masks

[0474] Here, the first step is to convert the z-stack of 3D cell positions-fluorescence images into appropriate masks for epithelial cells (C1) and immune cells (C2). Two probes are used to distinguish the two cell types - the first probe P1 recognizes all cells (Cs), while the other probe P2 depicts only epithelial cells. Therefore, the first mask C1 is derived as the maximum projection of the P1 z-image stack and directly assigned to the C1 mask. The second mask C2 results from the maximum projection of the P2 z-image stack, from which the P1 z-image stack is subtracted.

[0475] In the second step, the backscatter image BS of the substance (poison) is first thresholded (threshold delivering) to obtain a total substance mask (To). The To mask is segmented into a list of larger objects (aggregates) where the poison can be identified. Then, the interface areas (surfaces) of these aggregates are excluded to obtain a mask for the isolated poison ToCQ. Since the masks of all non-aggregated materials are provided as the complement of the isolated material ToCQ mask to the total material mask To, the masks for the amount of internalized poison in the corresponding cell types ToC1 and ToC2 are obtained by multiplying the non-aggregated material mask with the specific cell type masks (C1 and C2), respectively.

[0476] Once all masks are created, simply multiply the poison intensity image with the corresponding mask to get the poison content.

Claims

1. A method for predictive testing of a pharmaceutical agent based on the time propagation of at least two observables using a system of differential equations, wherein An observable is any property that changes in a set of in vitro experiments, and each observable is experimentally measurable during at least one in vitro experiment in the set of in vitro experiments, and · at least one observable is causally related to a known adverse effect that can be measured in vivo or in vitro following exposure, and At least one in vitro experiment involves exposing the in vitro model to the selected agent to look for the predicted outcome.

2. A method for predictive testing of a pharmaceutical agent according to claim 1, wherein prediction of a pharmaceutical agent-related adverse outcome is determined by computer simulation-based time propagation of an in vitro detectable observable, the evolution of which can be transformed into a parameterizable set of predefined interaction terms that define a system of differential equations applicable to the desired time propagation, wherein the observable (q i ) is at a specific time point {t m } for concentration, surface, amount, size, and similar properties measured experimentally in vitro, and wherein the interaction term defines the effect of the interaction on the concentration, surface, amount, size, and similar properties due to at least one other observable (q j ) causes the change in the observable quantity (qi) due to the interaction.

3. A method for predictive testing of a pharmaceutical agent according to claim 1 or 2, wherein the interaction terms can be denoted as IMij and are generally integrated into an interaction matrix IM(t) .

4. A method for predictive testing of pharmaceutical agents according to any one of the preceding claims, wherein the interaction matrix IM(t) With scene vector Multiply, using the scene vector The concept of is to select the interaction terms relevant to a specific experiment and adapt the system of differential equations to describe the evolution of the entire in vitro system or only a part of it, where the scenario vector is defined as where u i is unit value 0 or 1, and where u i Non-zero values ​​of are used to select detectable observables {q i }, the experimental setup is a combination of in vitro (sub)systems and exposures.

5. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein at least two in vitro detectable and in vivo relevant observable quantities q i The computer simulation predictions are made by using the matrix equation Define the system of differential equations, starting from their initial values The matrix equation is implemented by the scenario Selected Interaction Matrix IM(t) is defined by the interaction term.

6. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein q i Refers to an amount detectable in vitro and corresponding to or correlable to an amount detectable in vivo / ex vivo that can be correlated to a critical event that inevitably leads to the next event in the sequence of events from exposure to adverse outcome (AO).

7. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the interaction matrix term IM ij By the power parameter p ij and the product of at least two interaction functions IF, which define the interaction matrix terms IM ij For the observable quantity q k where each interaction matrix element is defined as The power parameter p i,j Each interaction term is defined as i The impact of the rate of change.

8. Method for predictive testing of pharmaceutical agents according to any of the preceding claims, wherein for parameterization of the interaction matrix, the values ​​of time points, observables and their time derivatives at a given time point are used as input:

9. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the interaction function IF(q) is generalized to take a form independent of the (pre)known mechanism, assuming that the dependence of the interaction term on the observable q is as follows where a, b and c can have simple integer values ​​0, 1, 2, ... and can be converted to o1 and o2, which have integer values ​​-1, 0, 1, 2, ..., and define the derivative dIF / dq at smaller and larger values ​​of q, i.e. determine the sensitivity of the interaction term to smaller and / or larger values ​​of the observable q.

10. The method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the sets (a, b, c) and (o1, o2) can be related by an interaction function representing the most biologically relevant interaction function: -(1,1)-linear dependence on q, -(1,0)-saturation dependence on q, -(0,1)-threshold / delay dependence on q, -(0,0)-saturation & threshold dependence on q, -(0,2)-nonlinear / quadratic dependence on q, -(-1,0)-Term that is independent of q.

11. The method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein for determining the parameter a linear combination of linearly independent expressions is used. Among them and p i,j Related B i,j (the sum of the products of the interaction functions) becomes dependent on the observable {q b }, the observable {q b } appears in the interaction function within the inner sum of equation [12].

12. A method for predictive testing of a pharmaceutical agent according to the preceding claim, wherein the parameterization using equation 12 is performed in the presence of only the parameter p i,j The other parameters appear only in the complement of this particular subgroup of equations.

13. A method for predictive testing of a pharmaceutical agent according to any one of claims 8 to 12, wherein the system of equations in claim 11 is solved using any suitable optimization approach, such as a linear or nonlinear optimization approach, preferably using a singular value decomposition (SVD), wherein the time point t m , observable quantity q i (t m ) and their corresponding derivatives are replaced by their measured values ​​or the values ​​in claim 8.

14. Method for predictive testing of a pharmaceutical agent according to the preceding claim, wherein SVD is used and the determined parameters are restricted to positive values.

15. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the observable is analysed, measured and / or detected in a time-lapse manner, ie at least three time points.

16. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein monitoring of the in vitro test is performed for up to 1 week, preferably up to 3 days, most preferably up to 50 hours.

17. A method for predictive testing of a pharmaceutical agent according to the preceding claim, wherein observables are acquired at at least three different time points with arbitrary delays, and wherein the minimum delay is 5 minutes.

18. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the method comprises the following steps: a) a preparation phase to identify relevant observables in a relational manner, b) preparing an in vitro model for a living system associated with a critical event and associated with a deleterious effect, preferably selected based on the expected function, mode of action and / or entry point of the living system, wherein the in vitro model comprises at least one cell type, c) scenario-based time-lapse acquisition to detect the early time evolution of an observable in said in vitro model at at least three time points with any time delay exceeding 5 min in at least two scenarios, wherein at least one scenario refers to an unexposed in vitro model and at least one scenario refers to an exposed in vitro model, wherein the test substance is applied at a dose capable of inducing at least a minimal effect in said in vitro model, d) quantifying observables to quantify their values ​​in the in vitro model, wherein at least one observable refers to a property of the unexposed in vitro model and at least one observable relates to a property that is associated with exposure-related changes in the in vitro model, e) generating time series and time derivatives of said observables to obtain a vector of observables for said scenario, which vector defines the time evolution of the in vitro model and constituent events, f) constructing the interaction terms of the interaction matrix through interaction functions to define all possible interactions and couplings between observables biophysically and biologically, thereby mathematically determining the time evolution of all observables, g) constructing basis functions and super equations to transform the set of interaction functions of the previous step into a set of mathematically orthogonal functions that can be used to numerically parameterize the evolution of observables in the next step, h) parameterizing the interaction terms defining the rates of change of the observables, said parameterization using the experimental data derived from the scenario, the set of orthogonal functions defined previously and the chosen numerical method to determine the system of equations and to implement in a next step the numerical time propagation of the observables and to identify in a second step described below the most relevant interaction terms, i) perform scenario-dependent fitting of the in vitro experiments to derive the value of each observable under each scenario, using the parameterization of the interaction matrix from the previous step and starting from the given initial conditions for each observable, j) optionally present the identified mechanisms to identify the most relevant interaction terms and describe their contribution to the observed early events, i.e. the observable quantities of the test drug and their evolution to potential adverse outcomes, k) optional dose biodistribution characterization, and / or l) Long-term propagation of the dose-dependence of observables to ultimately determine AO ​​predictions using a parameterized interaction matrix over long-term propagated AO-related observables.

19. A method for predictive testing of a pharmaceutical agent according to the preceding claim, wherein the interaction function IF(q i ,o1,o2) respectively defines the interaction matrix term IM through the derivatives o1 and o2 ij For the observable quantity q i The dependence of and interaction function IF(q i ) for q i The small value of q i The sensitivity to large values ​​of , where the interaction matrix term IM ij can be defined as the power parameter p ij With at least two interaction functions IF(q k ), where k can be the index of the i row or j column, depending on the interaction matrix term IM ij An array of observables / positions.

20. The method for predictive testing of a pharmaceutical agent according to claim 18 or 19, wherein the basis function Bij is the interaction function IF(q k ) and the power parameter p ij Related, creating a model that can describe each observable q i A basis of linearly independent functions of the time derivatives of .

21. The method for predictive testing of a pharmaceutical agent according to claims 18 to 19, wherein the super equation refers to a subgroup of differential equations, i.e. the group of rows of the interaction matrix, in which only the power parameter p appears. ij subgroup of , without any other power parameters.

22. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the choice of the observable depends on the substance being tested and previous knowledge about the development of outcomes or symptoms that may be associated with the substance, e.g. based on AOPs, previous in vivo data and / or scientific reports and articles.

23. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the observable quantity is a change in a quantifiable property of a molecule such as a lipid, protein, RNA, DNA, etc., or a supramolecular such as a membrane, ribosome, cytoskeleton, fiber, vesicle, etc., or a cell such as a cell surface, volume, shape, activity, cell compartment surface, volume, shape, etc.

24. A method for predictive testing of a pharmaceutical agent according to the preceding claim, wherein the event is typically 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 into and out of the extracellular space (vesicles, etc.), - breaking the contacts between cells (tight junctions), - Actin and tubulin rearrangement, - chromatin condensation or decondensation and other nuclear phase changes, - Cell lysis, - apoptotic or necrotic events or events associated with other cell state changes, - changes in ER, endosomes, lysosomes, mitochondria and / or ribosomes, - Cell death, - sequestration, engulfment, or formation of other new structures on the cell surface or within the cell, such as the sequestration described by Kokot et al., - transcription, synthesis, expression, transfer and release of attractants in immune cells and other cell types, - binding state, charge, chemical composition of interacting surfaces and aggregate structure (endogenous or exogenous) of molecules, molecular complexes and supramolecular structures, - the type of interaction exhibited by any organelle or molecular complex, supramolecular structure or external / exogenous substance on the interaction surface, - release and / or uptake of molecules, vesicles, etc. from one intercellular / intracellular compartment into another, - the surface area of ​​the selected type of cells (which also corresponds to the number of cells or cell concentration, etc.), - the surface area of ​​the toxicant in the selected compartment (which also corresponds to mass, concentration, volume, etc.), - the quantity of signaling or transport molecules or other structures (also corresponding to the mass, concentration, binding state, etc. of cytokines, chemokines, enzymes, receptors, various types of RNA, exosomes, endosomes, liposomes, etc.), - Surface area of ​​the new or modified structure (also corresponds to mass, volume, density, lifetime, dimension, correlation length, shape descriptors, etc. of supramolecular structures (e.g. vesicular structures, plasma or inner membranes, fibers inside or outside cells, tight junctions between cells, intercellular spaces, mitochondria or mitochondrial networks, nuclei or parts / domains in nuclei, etc.).

25. The method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the method for obtaining an observable quantity may be any suitable and widely used method, preferably: - microscopy, such as bright field, fluorescence wide field or confocal, scattering, Raman, ESEM, FTIR, etc., for the observation and quantification of the temporal evolution of said observables (related to structural or biochemical information, see below) and the determination of their temporal derivatives; - Spectroscopy, such as Raman, FTIR, UV-VIS, fluorescence, specific staining, etc., for the determination of the concentration / amount of the compound classes of interest (e.g. proteins, lipids, aromatic substances, markers, labelled substances, inorganic substances) throughout the sample, as well as 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., for the determination of the concentration / amount of a specific compound of interest (e.g. a selected cytokine) in the whole sample, as well as the quantification of its time evolution and the determination of its time derivative; - Omics data, such as proteomics, lipidomics, transcriptomics, metabolomics, etc.

26. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein at least a portion of the observables are selectable based on AOPs, previous in vivo data and / or scientific reports and articles.

27. A method for predictive testing of a pharmaceutical agent according to any of the preceding claims, wherein if the evolution of an adverse outcome is studied, at least one observable should preferably match the earliest event observed in vivo or in a patient.

28. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the observable quantity q is subdivided into 3 categories: cellular compartments, quantities of the pharmaceutical agent, i.e., the poison, associated with the compartment and quantities associated with the phenomenon.

29. A method for predictive testing of pharmaceutical agents according to the preceding claims, wherein in the case where the toxicity assessment involves inhalation and the lung is the organ of interest, e.g. because it is first exposed, the observables associated with chronic inflammation are associated with immune cells, e.g.: - the number of monocytes entering the lungs (because they differentiate into macrophages, replacing dead resident macrophages), - the number of polymorphonuclear (PMN) cells that enter the lungs (because they are most convenient to identify ex vivo and they correlate well with the influx of monocytes), - The number of leukocytes entering the lungs (because their time-dependent rate changes are the most significant and they also correlate well with PMN influx).

30. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein - the in vitro model selected is relevant to the specific agent being evaluated, mimics at least a relevant portion of the relevant target tissue and the relevant agent delivery pathway, - Each cell type or combination used o mimics the AOP-related portion of the target tissue in structure or function, and ο Ability to express one or more AOP-related key events identified from prior knowledge; - All cell types can respond together to the relevant early part of the AOP o precedes the later part of the AOP that is observed in vivo and develops into the observed deleterious effects, and o results in at least one critical event that predicts an adverse outcome, AOP refers to the broad concept of an adverse outcome pathway, which includes all causally related events from the molecular initiating event and all subsequent events to the final adverse outcome, whether the event is physical, chemical, biological, biophysical, biochemical or any other type.

31. A method for predictive testing of an agent according to any one of the preceding claims, wherein the in vitro model may comprise any combination of single cell cultures or various co-cultures of cells or more complex cell cultures, depending on the properties of the test agent.

32. A method for predictive testing of a pharmaceutical agent according to the preceding claims, wherein the cells used for the in vitro model can thus be selected from the group consisting of at least epithelial lung cells, epithelial skin cells, neurons, endothelial cells, muscle cells, intestinal epithelial cells, mucus cells, parietal cells, chief cells, endocrine cells, immune cells such as macrophages, glial cells, etc.

33. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the quantification of observable quantities in control (no substance) and exposed (test substance applied at a selected dose) in vitro models depends on the type of data acquired: - If the acquired data is already in numerical form, such as ELISA, transcriptomics and / or proteomics, it can be directly used for derivative generation, - If the acquired data is not in numerical form, such as when acquiring images, the analysis will perform a two-step approach: o masking portions of the image that are relevant to event observables, such as cell line type, labeling of cells or cell structures when appropriate, sensitivity of the method, co-localization information, etc., and ο Quantify the corresponding observables, where the mask from the previous step is converted into a quantized observable q for each time t and each region of interest (ROI), where typically the surface area is obtained by summing the number of pixels within the mask and the quantity is obtained by summing the intensities of the appropriate image channels within the mask.

34. A method for predictive testing of a pharmaceutical agent according to the preceding claim, wherein an additional histogram or intensity distribution analysis (such as FIDA, Fluorescence Intensity Distribution Analysis) is used to further optimize the quantification procedure.

35. A method for predictive testing of a pharmaceutical agent according to the preceding claim, wherein quantification results from a plurality of ROIs can be summed.

36. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the noise reduction approach to the time derivative determination comprises the following assumptions: - Each observable quantity (key event) changes slowly over time, with at most one maximum or minimum value in the experimental monitoring time interval. - During the experimental monitoring time interval, each observable can have at most one fast-changing segment, which means that the absolute value of the derivative of the observable can have at most one maximum or minimum, - Any other rapid changes are due to experimental noise, - and at a given time interval {t i ..t max }, time t j Observable quantity q i The set of measured values ​​(i.e., vector, represented by {}) {q i (t j )} Available fitting value set {q i-f (t j )} (with the additional index -f indicating the fitted value), or by replacing {q i (t j )} is fitted to a predefined function such as the following for interpolation: q i-f (t)=a+bt Exp[ct] - and similarly, the time set of the observable derivative values ​​{dq i / dt(t j )}Use the same time point {t j The derivative value set {dq i-f / dt(t j )} to approximate.

37. A method for predictive testing of an agent according to any one of the preceding claims, wherein the test agent is delivered in any manner, preferably uniformly, into the biological system.

38. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the dose biodistribution is assessed as follows: - segmenting (by one of the available image segmentation algorithms) the images from the channel delivering or related to the poison information to create a list of objects related to the drug (poison), - optionally, the list of segmented objects can be additionally modified by excluding objects based on their size or other descriptors, wherein the latter can be calculated by one of the available algorithms taking into account intensity, size, aspect ratio, edge length / surface, skeleton information, mass or cross-sectional size, etc., - Optionally, if the model requires or assumes, the object can be further divided into the surface of the object (accessible agents) and the body of the object (inaccessible agents), - For each segmented object in the list, its dose equivalent is calculated as follows: o Dose expressed as mass or surface area of ​​the agent (if the intensity is not dependent on the labeling / identification process, then sum the pixel intensities within the segmented object - valid for Raman-based assays, for example) o Model-corrected dose expressed as mass or surface area of ​​the agent (summing the number of pixels within the segmented object if the intensity does depend on the labeling / recognition process - e.g. valid for fluorescence-based or scatter-based assays; the model requires advanced calibration of information density, scatter intensity, etc.) o Dose expressed as interaction dose (summing up the intensity of only pixels in contact with the biological system of interest; this could be just the surface of an object, or even just a portion of a surface, e.g. one in contact / co-localized with a membrane, protein, RNA, etc.) - From the list of subjects and their associated doses, create a dose histogram showing the number of subjects exhibiting a certain dose (bin, sub-range); where the number of dose bins is defined based on the total number of subjects in the analysis.

39. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein a histogram of dose distribution is determined using a multi-ROI image.

40. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein various combinations of dosage concepts can be used in the calculation and / or presentation of the prediction, such as: - For barrier organs with a large surface area to volume ratio, such as the lung or liver, the ratio of the local dose of the material surface to the cell surface is used for materials, and the ratio of the mass of the compound to the cell surface is used for insoluble compounds, - For thicker and bulky organs with a low surface area to volume ratio, such as muscle or brain, use the ratio of the material surface to the organ / body mass for materials and the ratio of the compound mass to the organ / body mass for soluble compounds; - for insoluble compounds in barrier-like organs the ratio of the mass of compound to the surface of the lung, subsequently recalculated as the ratio of the mass of compound to the exposure time, - For large organs the ratio of the mass of compound to the mass of the organ / body when used for soluble compounds is always recalculated as the ratio of the mass of compound to the exposure time.

41. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein: The parameterized portion of the interaction matrix is ​​extended to enable additional biological response information at the tissue level based on AOPs or other in vivo data sources represented by local-to-system coupling and system-to-local coupling, where the additional portion of the interaction matrix provides a means to couple local events to system events and vice versa.

42. A method for predictive testing of a pharmaceutical agent according to any one of the preceding claims, wherein the end result describing the mechanism (mode of action) may include any of the following statements: - Drug T for cell type C i Severe toxicity, - Drug T for cell type C i Non-toxic, - Drug T for cell type C i Shows contact toxicity, - If the drug T is taken by cell type C j Absorption, then for cell type C i Indirect toxicity. - Local phenomenon Pl i Stimulate / inhibit cell type C j growth, - Cell type C i Increase / decrease agent T in compartment C j accessibility, - Cell type C i Inducing agent T in compartment C j Medium aggregation / biodegradation, - Drug T cells are type C i Effective absorption, - Drug T cells are type C i Cell type C j Effective absorption, - The potion T is being quarantined. -Phenomenon Pl j Increase / decrease compartment C j The availability of Chinese medicine T, -Phenomenon Pl j Inducing agent T in compartment C j Aggregation / biodegradation in - When present in cell type C j During the period of drug T stimulation / inhibition phenomenon Pl i , - When isolated, the agent T stimulates / inhibits the phenomenon Pl i , -Phenomenon Pl i Stimulation / inhibition phenomenon Pl j , -Phenomenon Pl i Self-stimulation / self-inhibition its effects.

43. A method for predictive testing of a pharmaceutical agent according to the preceding claims, wherein the final result describing the biodistribution of the pharmaceutical agent may contain additional statements such as: - Agent T was uniformly dispersed in different sizes in the in vitro model, indicating that the biological system favors the dispersed form of Agent T but not its aggregated state, - Agent T was detected in large aggregates, indicating that the biological system favors the aggregation state of agent T, - Agent T is effectively sequestered in cell type C i above / inside / side / outside, - Agent T is not detected by cell type C i absorb, - Agent T is dispersed in cell type C in the form of many submicron particles i middle.

44. A method for predictive testing of a pharmaceutical agent according to the preceding claims, wherein the final result provides a prediction of a final adverse outcome associated with the pharmaceutical agent, and the statements in claims 43 and 44 can be combined with additional statements to further define the evolution of the biological system, and / or to provide other types of statements, such as the following statements: - Drug T is safe - biological systems have not shown any abnormal evolutionary responses, - Elixir T is safe - Elixir T will not cause any delayed response, - Agent T is safe at doses that are expected to be encountered under specific conditions, - Drug T is safe at doses below a certain value, - Agent T is safe at dose rates below a certain value, - Agent T will cause a weak / strong / super-strong acute response when exceeding a certain dose or dose rate, but it will subside within a specific time. - Agent T will cause an acute response when it exceeds a certain dose or dose rate, and then it will be amplified into a strong subacute / subchronic response when it exceeds a certain different dose or dose rate, and it may turn into a long-term chronic inflammation. - Agent T instantly causes severe cell damage, showing its great toxicity. - In terms of the phenomenon of sustained increase / immune response, the immune response induced by drug T is weak but lasts for a long time. -Agent T triggers a weak acute response, but then amplifies into an unresolved persistent amplified inflammation (cytokine storm).

Images