Method for determining the composition of a reaction system

Abstract

The invention relates to a computer-implemented method comprising the automatic and rational (computer-aided) modification of a chemical formulation, in particular the design of the specific starting composition of a reaction system, such as an oligomeric or polymeric polyol system, using an optimization program intended to maximize similarity to the original target / reference system. Furthermore, the invention relates to a data processing device comprising means for carrying out the method; a computer program product containing instructions that cause a computer to execute the method when the computer executes the program; and a computer-readable storage medium containing instructions that, when executed by the computer, cause the computer to execute the method.

Description

[0001] The present invention relates to a computer-implemented method for designing a starting composition of a chemical reaction system to maximize the similarity with a reference system so as to replace individual components in the reference reaction system. The present invention also relates to a data processing device, comprising a device for executing the method; a computer program product containing instructions, when a computer executes the program, the instructions cause the computer to execute the method; and a computer-readable storage medium containing instructions, when a computer executes the instructions, the instructions cause the computer to execute the method.

[0002] Computer-aided modification of chemical information using optimization programs, in particular the design of specific starting material compositions for reaction systems, is a subject of current research. In order to further reduce the dependence on petrochemical-based raw materials and to comply with increasingly stringent regulatory requirements, it is of particular interest to use optimization programs in order to substitute these components in reaction systems while changing the material properties of the subsequent products as little as possible.

[0003] US2020 / 342960 A1 discloses a target-based drug screening method that uses inverse quantitative structure-(drug) performance relationship (QSPR) analysis and molecular dynamics simulation. The method includes modeling the molecular structure of a test compound group for a target molecule, obtaining a quantitative structure-(drug) performance relationship (QSPR) for the test compound group, obtaining the optimal pharmacophore of a new target-based drug by numerical inversion of the QSPR, and selecting candidate drugs from the test compound group that have a molecular structure similar to the optimal pharmacophore. This method is limited to single molecules and is not applicable to reaction mixtures or polymers. The optimization directly targets experimental properties that are highly uncertain or almost unavailable for computational prediction.

[0004] Maranas et al. solve the design of polymers with optimal macroscopic property levels by using topological indices. Specifically, two zero-order connectivity indices and two first-order connectivity indices are used as descriptors for structure-property correlations in optimization studies for the first time. Based on these descriptors, a new set of correlations for heat capacity, cohesive energy, glass transition temperature, refractive index, and dielectric constant are proposed. These correlations are integrated into the optimization framework. (Camarda, Kyle V. and Costas D. Maranas. "Optimization in polymer design using connectivity indices." Industrial & Engineering Chemistry Research 38.5 (1999): 1884-1892). This method does not take into account the inherent polydispersity of polymers, i.e., it is limited to ideal polymers modeled with infinite chains. Therefore, these methods cannot propose specific recipes. The optimization is directly targeted at experimental properties, which are highly uncertain or almost unavailable when calculated for prediction.

[0005] Faeder et al. describe the use of BioNetGen for rule-based modeling to study the dynamics of complex biochemical systems. Rule-based models can be analyzed by performing deterministic simulations or stochastic simulations. However, they are often difficult to calibrate due to many unknown parameters and limited experimental training data. They proposed a general parameter estimation framework to calibrate rule-based models. The experimental data as well as the qualitative properties of the system are encoded as canonical formulas in bounded linear temporal logic. Given a set of candidate parameter values, they apply a statistical model checking procedure to evaluate the quality of this set of candidate values. Based on the results, a new set of parameters is selected using a standard global search strategy. (Liu, Bing and James R. Faeder. "Parameter estimation of rule-based models using statistical model checking." 2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). IEEE, 2016). The method is limited to biological systems, i.e., it is not applicable to polymer formulations. In addition, it does not use atomic resolution of species and is therefore limited in calculations involving detailed characterizations.

[0006] Daoutidis et al. proposed a strategy that simultaneously determines (a) the most ideal biomass-derived products for applications of interest and (b) the corresponding synthesis routes. The strategy consists of: i) constructing an exhaustive reaction network consistent with a set of input chemical rules, and ii) using network information to formulate and solve an optimization problem that derives the best product distribution and the reaction sequence for synthesizing them. They use this strategy to determine the potential renewable oxygenates and hydrocarbons obtained from biomass heterogeneous catalysis, which can be blended with gasoline to meet ASTM specifications. Considered multiple targets (energy loss, catalyst requirements and absolute heat load), and found multiple alternative solutions in each case (Marvin, W.Alex, Srinivas Rangarajan and Prodromos Daoutidis. "Automated generation and optimal selection of biofuel-gasoline blends and their synthesis routes." Energy & Fuels 27.6 (2013): 3585-3594). The method is limited to single molecules and is not suitable for actual reaction mixtures or polymers. The optimization is directly aimed at experimental properties, which are highly uncertain or only available for small data sets when calculated and predicted. Mavrantzas et al. reported a method based on a multiscale model, which combines the computer-aided molecular design technology based on the group contribution model for predicting the properties of polymer repeating units with the first principle arrangement for providing repeating units and the atomic simulation of the physical properties for predicting the selected candidate polymer structure. The method has been developed and tested for designing polymers (Satyanarayana, Kavitha Chelakara et al., "Computer aided polymer design using multi-scale modelling." Brazilian Journal of Chemical Engineering 27 (2010): 369-380) with desired properties. The method is limited to polymers modeled as infinite chains of repeating units and is not suitable for actual reaction mixtures or polydisperse polymers.

[0007] To date, a method for optimizing the starting composition of complex reaction systems, i.e., proposing specific alternative recipes that are optimized using the similarity of a set of descriptors derived from their components at monomeric and atomic resolutions, has not been described.

[0008] As mentioned above, computer-aided design methods in the art are typically based on computer-predicted properties that are used to define target values ​​or ranges that are desired for most well-defined single compounds or polymers modeled as monodisperse systems of infinite chain length. In practice, these methods are hampered by several problems. First, predicting the molecular properties of molecules, molecular mixtures, or polymers with chemical accuracy for many chemical systems remains an unresolved challenge, and generally any property prediction involves a certain inaccuracy when compared to experiment, which may degrade any subsequent optimization procedures. Second, the system is always assumed to have monodisperse characteristics, but for many systems, especially oligomer and polymer systems, polydisperse characteristics are important, for example, the effects of different molecular chain lengths and / or end groups are not negligible. In fact, most chemical systems are essentially mixtures with complex compositions, rather than pure well-defined systems.

[0009] The object of the present invention is to at least partially overcome the shortcomings in the art. Specifically, the object of the present invention is to efficiently and accurately design chemical reaction systems to replace certain raw materials, i.e. one or more components in a chemical reference reaction system (e.g. with bio-based materials), while largely maintaining the material properties of the products produced by the chemical reaction system.

[0010] This object is achieved by a method according to claim 1. A data processing device is the subject of claim 13, a computer program product is the subject of claim 14, and a computer-readable storage medium is the subject of claim 15. Preferred embodiments are the subject of the dependent claims. They can be freely combined unless the context clearly indicates otherwise.

[0011] Therefore, a computer-implemented method for designing a starting composition of a chemical reaction system to maximize similarity to a reference reaction system to replace individual components in the reference system comprises:

[0012] A) providing information about a reference reaction system, the information comprising components of the reference reaction system and their relative amounts in the reference reaction system;

[0013] B) providing information about the test reaction system, the information including components of the test reaction system and their relative amounts in the test reaction system;

[0014] C) providing at least one first rule for forming and / or breaking chemical bonds from components of a reference reaction system and / or species formed from components of a reference reaction system;

[0015] D) applying the at least one rule to a reference reaction system to obtain a modified reference reaction system;

[0016] E) providing at least one descriptor of at least a portion of the modified reference reaction system and at least a portion of the modified test reaction system as described in step H);

[0017] F) determining, from the at least one descriptor, at least a portion of the modified reference reaction system's descriptor value(s);

[0018] G) providing at least one second rule for forming and / or breaking chemical bonds from components of the test reaction system and / or species formed from components of the test reaction system;

[0019] H) applying the at least one second rule to the test reaction system to obtain a modified test reaction system;

[0020] I) determining at least part of (one or more) descriptor values ​​of the modified test reaction system from at least one descriptor as described in step E);

[0021] J) determining a distance and / or dissimilarity between the values ​​of at least one descriptor from steps F) and I) using a measure of distance in the descriptor space occupied by the descriptors;

[0022] K) determining whether to adjust the test reaction system according to the results of the distance and / or dissimilarity determined in step J);

[0023] L) If an adjustment has been decided in step K), the test reaction system is adjusted by changing the components and / or their relative amounts in the test reaction system to obtain an updated test reaction system, and then the method is repeated starting from step H) based on the updated test reaction system.

[0024] The method according to the example has several advantages. The method calculates the similarity based on descriptors between the first chemical reaction system (i.e., the reference reaction system) and the new (second reaction system) system (i.e., the test reaction system). The method does not just predict the target properties of the new system. This avoids some problems caused by inaccurate calculation of properties or insufficient models, which usually occur in complex chemical systems. In other words, the goal of the optimization problem is not to specify the reference range of properties, but to undergo a reference reaction system calculated with the same descriptor as the chemical system to be optimized (i.e., the test reaction system). Avoid directly predicting (target) properties, but use the concept of descriptor similarity to perform an overall optimization scheme. The basic concept is to utilize the benefits of error cancellation, because in both cases, the same approximation and assumptions are involved in calculating descriptors for the two systems.

[0025] Another advantage of the method is that, by using a kinetic model as the basis for any subsequent calculations, it explicitly takes into account the polydisperse nature of oligomeric or polymeric reaction systems, whereas only monodisperse pure and well-defined systems have been used so far. Furthermore, complete formulations (i.e. the composition of the reaction mixture) can be proposed by the present invention, whereas so far computer-aided design has been limited to pure and non-reactive single molecular structures.

[0026] In summary, this method allows computer-aided rapid proposal of new chemical formulas for reaction systems that are most similar to a given reference system. This is particularly useful for effectively replacing raw materials (i.e., one or more components of a chemical formula that must be replaced, for example, due to regulatory or economic reasons) (e.g., with bio-based materials) without compromising the material properties of the overall product. Since the properties of the product produced from the proposed formula (i.e., the test reaction system) are already very close to the reference reaction system, far fewer experimental iterations are required to adjust the formula according to its changed components.

[0027] Computer-based methods for similarity searches of chemical reactions are well known in the art. For example, US2009 / 024575 A1 discloses a computer-based method for similarity searches of chemical reactions to study large reaction databases. Here, the query chemical reaction is characterized by a structural descriptor, and is compared by its similarity / distance with the reaction descriptors stored in the reaction database. In contrast to the present invention, the chemical reactions (including their rate constants) of the reference system and the test system have been defined at the beginning of the workflow, and are unrelated to the reactants or products (neither monodisperse nor polydisperse) of any specific reaction, nor to any kinetics (such as the form of chemical rate constants). Only the abstract chemical transformations between molecules themselves are searched for similarity. Therefore, this method cannot obtain reaction products, nor can any formula be proposed.

[0028] According to the present invention, information about a reference reaction system (step A)) and information about a test reaction system (step B)) are provided. Both the reference reaction system and the test reaction system may have a plurality of chemical components. According to a first embodiment of the present invention, the components of the reference reaction system as described in step A) and / or the components of the test reaction system as described in step B) include alkylene oxides, polyfunctional carboxylic acids, carboxylic anhydrides, cyclic ethers, carbon dioxide, cyclic carbonates, polyols, polyisocyanates, polyamines, aromatic hydrocarbons, olefins, (meth)acrylates, bisphenols, phosgene, dialkyl carbonates, aldehydes, lactams, glycolides, amino acids, hydroxy-substituted carboxylic acids or a combination of at least two of the foregoing components.

[0029] According to step C) of the computer-implemented method of the present invention, at least one first rule is provided for forming and / or breaking chemical bonds from components of the reference reaction system and / or species formed by components of the reference reaction system. In addition, according to step G), at least one second rule is provided for forming and / or breaking chemical bonds from components of the test reaction system and / or species formed by components of the test reaction system.

[0030] According to one embodiment of the present invention, at least one first rule as described in C) includes: which functional groups of the component exist, which functional groups of the component can be used to form covalent bonds or break covalent bonds, which functional groups of the component can react with other functional groups, the threshold criteria when the formation or breaking of covalent bonds occurs, or a combination of at least two of the aforementioned rules.

[0031] Preferably, at least one first rule of step C) is identical to at least one second rule of step G). This ensures that the reference reaction system and the test reaction system are modified according to similar chemical transformations.

[0032] In a preferred embodiment, the relative rate constant derived and / or measured from the kinetic experiment can be used as input data for setting at least one first rule of step C) and / or at least one second rule reaction rule of step G). Alternatively, these rules can be simulated. However, in some cases, the simulation may not be accurate enough.

[0033] According to step D) of the method of the invention, at least one first rule is applied to the reference reaction system, thereby obtaining a modified reference reaction system. According to a preferred embodiment, applying at least one first rule as described in step D) comprises running a kinetic Monte Carlo simulation, running a molecular Monte Carlo simulation, running a molecular dynamics simulation, running a Miller-Macosko type calculation. In this regard, in particular, kinetic Monte Carlo simulations have been shown to be very efficient and fast compared to other types of simulations.

[0034] Preferably, the modified reference reaction system obtained by applying at least one first rule to the reference reaction system includes a combination of discrete components and information about the discrete components of the modified reference reaction system and their relative amounts in the modified reference reaction system.

[0035] According to step E) of the method of the invention, at least one descriptor of at least part of the modified reference reaction system and at least part of the modified test reaction system as described in step H) is provided. In other words: according to the invention, it is ensured that the same descriptors are used for both the modified reference reaction system and the modified test reaction system. This in turn ensures that a valid distance / similarity measure according to step J) can be determined.

[0036] At least one descriptor can be selected from a very large number of descriptors. According to one embodiment of the present invention, at least one descriptor as described in step E) is a statistical descriptor / compositional descriptor, preferably number average molecular weight, weight average molecular weight, viscosity average molecular weight, polydispersity index, average functionality, OH value, acid value.

[0037] It is also possible that at least one descriptor is a topological descriptor, preferably the Wiener index, the Randic connectivity index, the Balaban-J index, the Ipc index, the Zagreb index, the Bertz index, the chi-molecule connectivity index, the Kier-Hall valence connectivity index, the seniority and priority indices, the kappa shape index, the BCUT index, the Estate index, the walk and path counts, the extended connectivity fingerprint.

[0038] It is also possible that at least one descriptor is a molecular descriptor, preferably Labute's approximate surface area, solvent accessible surface area, radius of gyration, 2D autocorrelation, eigenvalue-based descriptors, RDF descriptors, 3D-MoRSE descriptors, WHIM descriptors, GETAWAY descriptors, functional group count, number of rotatable bonds, number of hydrogen acceptors and donors, Gasteiger / Marsili partial charge, topological polar surface area, molar refractive index.

[0039] It is also possible that at least one descriptor is a force field based descriptor, preferably intramolecular potential energy, radial distribution function, bond angle distribution function, dihedral angle distribution function; and / or a quantum chemical descriptor, preferably atomic charge, HOMO and LUMO energy, molecular hardness, molecular polarizability, dipole moment, total energy, molecular quantum number descriptor.

[0040] It is also possible that at least one descriptor is a thermodynamic descriptor, preferably octanol-water partition coefficient, sigma-profile, sigma moment, polar moment vector, activity coefficient, chemical potential, free energy, octanol-water partition coefficient, solubility.

[0041] It is also possible that at least one descriptor is a health, sustainability or environment related descriptor, preferably toxicity, CO2 equivalent, carbon footprint, energy consumption, product recycling rate, sustainability index, supply chain mileage, water footprint, price.

[0042] Preferably, a combination of at least two of the aforementioned descriptors is selected.

[0043] Using the at least one descriptor thus provided, one or more descriptor values ​​of at least a portion of the modified reference reaction system can be determined from the at least one descriptor as taught in step F) of the method according to the invention.

[0044] If the modified reference reaction system comprises a set of discrete components and information about the discrete components of the modified first reaction system and their relative amounts in the modified reference reaction system (as mentioned above), it is preferred to use atomic resolution information for each discrete component to carry out the determination of at least part of the (one or more) descriptor values ​​of the modified reference reaction system by at least one descriptor in step F). In the case where at least one descriptor is a thermodynamic descriptor (as mentioned above), these descriptors preferably contain information about polarity and solubility and are preferably obtained from the liquid phase thermodynamic method COSMO-RS. Here, the previously obtained atomic resolution information is used to construct the molecular input (σ-distribution) for COSMO-RS calculations and descriptor generation.

[0045] According to step G) of the method of the invention, at least one second rule is provided for forming and / or breaking chemical bonds from components of the test reaction system and / or species formed from components of the test reaction system.

[0046] According to a preferred embodiment of the present invention, at least one second rule (as in the case of at least one first rule described in step C) comprises: which functional groups of the component exist, which functional groups of the component can be used to form covalent bonds or break covalent bonds, which functional groups of the component can react with other functional groups, the threshold criteria when the formation or breaking of covalent bonds occurs, or a combination of at least two of the aforementioned rules.

[0047] As mentioned above, at least one second rule may be the same as at least one first rule.

[0048] According to step H) of the method of the present invention, at least one second rule is applied to the test reaction system to obtain a modified test reaction system. Preferably, at least one second rule also includes running a kinetic Monte Carlo simulation, running a molecular Monte Carlo simulation, running a molecular dynamics simulation, and running a Miller-Macosko type calculation.

[0049] Preferably, the modified test reaction system obtained by applying at least one second rule to the test reaction system (i.e., the system to be optimized) - as in the case of a modified reference reaction system - comprises a set of discrete components and information about the discrete components of the modified test reaction system and their relative amounts in the modified test reaction system.

[0050] In step I) of the method according to the invention, one or more descriptor values ​​of at least part of the modified test reaction system are determined from at least one descriptor as described in step E).

[0051] Similarly, if the modified test reaction system includes a set of discrete components and information about the discrete components of the modified test reaction system and their relative quantities in the modified test reaction system (as mentioned above), it is preferred to use atomic resolution information of each discrete component to perform step I) of determining (one or more) descriptor values ​​of at least a portion of the modified test reaction system by at least one descriptor.

[0052] Using the (one or more) descriptor values ​​for both the modified reference reaction system and the modified test reaction system determined according to steps F) and I), the distance and / or dissimilarity between the values ​​of at least one descriptor from steps F) and I) can be determined (step J)) using a measurement of the distance in the descriptor space occupied by the descriptors.

[0053] In a preferred embodiment, the distance and / or dissimilarity described in step J) comprises Euclidean / L2 distance, Manhattan / L1 / city block distance, Canberra distance, Chebyshev distance, Mahalanobis distance, Minkowski distance, Rogers-Tanimoto dissimilarity, Russell-Rao dissimilarity, Sokal-Michener dissimilarity, Sokal-Sneath, mean absolute percentage error, Yule dissimilarity, cosine dissimilarity, dice dissimilarity or a combination of at least one of the foregoing distances and / or dissimilarity.

[0054] Based on the results of the distance and / or dissimilarity determined in step J), it is decided whether to adjust the test reaction system. Many different decision criteria can be applied. According to a preferred embodiment, the decision as described in step K) includes one or more of the following decision criteria: a threshold value for the distance and / or dissimilarity in step J); a threshold value for the change in the distance and / or dissimilarity relative to one or more previously performed steps J); a threshold value for the execution time of the method; a threshold value for the number of repetitions of the method; a threshold value for at least one of the descriptors. Here, it is preferred to apply a threshold value for the change in the distance and / or dissimilarity relative to one or more previously performed steps J).

[0055] If it has been decided in step K) to adjust the test reaction system (i.e. the system to be optimized), then in step L) of the method the test reaction system is adjusted by changing the components and / or their relative amounts in the test reaction system to obtain an updated test reaction system.

[0056] Preferably, the adjustment of the test reaction system as described in step L) is performed by executing an algorithm selected from the following: random search, grid search, Bayesian optimization, simplex optimization, evolutionary optimization, genetic algorithm, particle swarm optimization, Metropolis-Hastings Markov chain Monte Carlo, adaptive Markov chain Monte Carlo, simulated annealing, parallel tempering, hybrid linear and nonlinear programming or a combination of at least two of the aforementioned algorithms.

[0057] Using the adjusted test reaction system, the method according to the invention is repeated starting from step H), i.e., by subsequently repeating the method starting from step H) based on the updated test reaction system. This means that at least one second rule is also applied to the updated test reaction system, thereby obtaining a modified (updated) test reaction system again. Subsequently, one or more descriptor values ​​of at least part of the modified (updated) test reaction system are determined again using at least one descriptor as provided in step E) of the method according to the invention.

[0058] Such modified (updated) test reaction system's descriptor value(s) are then used to again determine the distance and / or dissimilarity between this / these value(s) and the value(s) obtained for the modified reference reaction system in step F).

[0059] If the adjustment of the test reaction system in step L) has been carried out in a suitable manner, the distance / dissimilarity determined again in step J) should decrease. Such iterations of steps H) to L) can be performed using an updated test reaction system until the distance / dissimilarity drops below a value satisfactory to the user.

[0060] If, usually after a certain number of iterations, the distance and / or dissimilarity between the (one or more) descriptor values ​​of the modified reference reaction system and the (one or more) descriptor values ​​drops below a certain threshold, then in step K), it is decided not to adjust the test reaction system, i.e., the optimization of the test reaction system is completed. In this case, an optimized or final test reaction system is proposed and preferably communicated to the user. It can be communicated to the user by all suitable means, such as by a personal computer or a mobile device. Then, the user can use the optimized or final test system as the starting composition of a chemical reaction to produce certain chemicals, especially in the green transformation process of the chemical industry to replace certain raw materials (traditionally based on petrochemicals) with bio-based materials.

[0061] Another aspect of the invention relates to a data processing device comprising means for performing the method of any of claims 1 to 12.

[0062] Yet another aspect of the present invention relates to a computer program product comprising instructions which, when a computer executes the program, cause the computer to perform the method of any one of claims 1 to 12.

[0063] Even a further aspect of the present invention relates to a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 12.

[0064] The invention is further described with reference to the following drawings, but is not intended to be limited thereto.

[0065] Figure 1 A flow chart of a method according to the invention is shown.

[0066] Combination Figure 1 The working process of the embodiment is described in detail.

[0067] The following example workflow was performed as follows with the goal of replacing a raw material in a polymer formulation (test reaction system) (e.g. due to regulatory or economic reasons) without compromising its properties in subsequent applications. The workflow is fully automated and coded in Python.

[0068] Example 1

[0069] Example The workflow was performed as follows with the goal of replacing the starter molecules of a polyol formulation (test reaction system) without compromising its properties in subsequent applications (e.g. in polyurethane reactions). The workflow was fully automated and coded in Python.

[0070] Reference reaction system 10 (first reaction system) is defined as consisting of the following compounds, i.e., the formula: ethylene oxide (0.6), glycerol (0.2) and ethylene glycol (0.2), with the composition in parentheses being the mass fraction. Glycerol is defined as having three alcohol groups, ethylene glycol is defined as having two alcohol groups, and ethylene oxide is defined as having one reactive epoxy group.

[0071] A kinetic Monte Carlo simulation (such as, for example, Suderman, Ryan et al., "Generalizing Gillespie's direct method to enable network-free simulations." Bulletin of mathematical biology 81.8 (2019): 2822-2848.) is set up, wherein a plurality of kinetic rules 20 are defined to allow alcohol groups to react with epoxy groups at a relative rate constant of 1, i.e., the functional groups of the identified components have the same reactivity. The kinetic rules 20 are applied to the first reaction system 10 (step 30) by using 100.000 initial seed molecules for simulation. After all reactive groups have reacted, a discrete representation of the polymer is obtained (modified first reaction system 40).

[0072] Subsequently, a plurality of descriptors 35 are selected and corresponding descriptor values ​​60 are calculated from this modified reference reaction system 40 (step 50), as shown in Table 1: The selected descriptors 35 are: polydispersity (PD), OH value (OHZ), average number of OH groups per molecule (fn(OH)), topological polar surface area (TPSA) and Wiener index obtained from the molecular graph. An explanation of the Wiener index can be found in Todeschini, Roberto and Viviana Consonni. "Handbook of molecular descriptors", John Wiley & Sons, 2008. The descriptor values ​​60 are saved for later use.

[0073] A new test system (test reaction system 70) is then defined using starting materials different from the reference reaction system, in this case ethylene oxide, trimethylolpropane and 1,2-propylene glycol, now using the same set of kinetic rules 20 as defined for reaction system 10. This test system is the system to be optimized in the present method. Trimethylolpropane is defined as having three alcohol groups, 1,2-propylene glycol is defined as having two alcohol groups, and ethylene oxide is defined as having one reactive epoxy group. The composition by mass fraction begins as a random guess.

[0074] To apply the kinetic rules 20 to the test reaction system 70 (step 80), a new kinetic Monte Carlo simulation is performed and a discrete representation of the polymer is obtained again after all reactive groups have reacted (modified test reaction system 90). The same set of descriptors 35 are selected for the modified test reaction system 90 and the corresponding descriptor values ​​are calculated 110 (step 100, see Table 1).

[0075] The distance 130 in descriptor space between the two modified reaction systems 40, 90 is then calculated by using the mean absolute percentage error between the descriptor vector 60 of the modified reference reaction system 40 and the descriptor vector 110 of the experimental system (modified test reaction system 90) (step 120).

[0076] The distance 130 is then handed over to a Bayesian optimizer (as implemented in a Python library), including the actual composition of the test reaction system 70. Depending on the value of the distance 130, the Bayesian optimizer now decides to adjust the test reaction system 70 (step 140) or not to adjust the test reaction system 70 and proposes a recipe (step 160). The decision criterion for the Bayesian optimizer to make a decision is to stop after a maximum of 100 iterations, when no significant changes in the distance metric are recorded. If the Bayesian optimizer decides to adjust the test reaction system 70, an updated test reaction system 180 is obtained by adjusting the test reaction system for the next optimization step (step 150). The adjustment can be achieved by an optimizer that proposes new starting materials / mass fractions based on the optimization history.

[0077] The process 70-140 is iterated for about 100 iterations / cycles until no change in the distance 130 is observed or the change is below a defined threshold. The final set of descriptor values ​​110 is obtained by the minimum distance between the descriptor values ​​60, 110 of the reference system / optimized system (i.e., the modified reference reaction system and the test reaction system 40, 90):

[0078]

[0079] Table 1

[0080] As a result, a formulation of a test reaction system 70 is proposed (step 160) having the following chemical composition: ethylene oxide (0.54), trimethylolpropane (0.22) and 1,2-propylene glycol (0.24), which is final test reaction system information 170. This proposed test reaction system 70 is then communicated to the user.

[0081] Example 2

[0082] Reference reaction system 10 (first reaction system) is defined as consisting of the following compounds, i.e., the formula: trimethylolpropanol (TMP) (0.5), 1,2-propylene glycol (0.1) and isophthalic acid (0.4), with the composition in mass fractions in brackets. Since TMP has been labeled as reproductively toxic according to REACH and the polyester reference formula contains residual amounts of monomeric TMP, the purpose of this example is to replace TMP, while the change in the overall polyester properties (given by the descriptor similarity) should be kept to a minimum.

[0083] TMP is defined as having three primary alcohol groups, propylene glycol is defined as having a primary alcohol group and a secondary alcohol group, and isophthalic acid is defined as having two carboxylic acid groups. Set up kinetic Monte Carlo simulation (such as Suderman, Ryan et al., "Generalizing Gillespie's direct method to enable network-free simulations." Bulletin of mathematical biology 81.8 (2019): 2822-2848. Kinetic Monte Carlo simulation), wherein multiple kinetic rules 20 are defined, to allow primary alcohol groups and carboxylic acid groups to react with a relative rate constant of 1, and secondary alcohol groups react with a relative rate of 0.1. Simulation is performed using 100.000 initial seed molecules, and kinetic rules 20 are applied to reference reaction system 10 (step 30). After all reactive groups have reacted, a discrete representation of polymer is obtained (modified reference reaction system 40).

[0084] Subsequently, a plurality of descriptors 35 are selected and corresponding descriptor values ​​60 are calculated from this modified reference reaction system 40 (step 50), as shown in Table 1: The selected descriptors 35 are: OH value (OHZ), average number of OH groups per molecule (fn(OH)), logarithm of octanol-water partition coefficient (logP) and Wiener index obtained from molecular graph. An explanation of Wiener index can be found in Todeschini, Roberto and Viviana Consonni. "Handbook of molecular descriptors", John Wiley & Sons, 2008. The descriptor values ​​60 are saved for later use.

[0085] A new test reaction system 70 (second reaction system) is then defined using different starting materials than the first reaction system (i.e., the reference system), in this case glycerol, ditrimethylolpropanol, diethylene glycol, ethylene glycol, 1,2-propylene glycol, and isophthalic acid, using the same set of kinetic rules 20 as defined for the reference reaction system 10. This test system is the system to be optimized in the present method. Glycerol is defined as having two primary OH groups and one secondary OH group, ditrimethylolpropanol is defined as having four primary OH groups, ethylene glycol and diethylene glycol are defined as having two primary OH groups, and 1,2-propylene glycol is defined as having one primary OH group and one secondary group. The composition by mass fraction begins with a random guess.

[0086] To apply the kinetic rules 20 to the test reaction system 70 (step 80), a new kinetic Monte Carlo simulation is performed and a discrete representation of the polymer is obtained again after all reactive groups have reacted (modified second reaction system or test reaction system 90). The same set of descriptors 35 is selected for the modified test reaction system 90 and the corresponding descriptor values ​​are calculated 110 (step 100, see Table 1).

[0087] The distance 130 in descriptor space between the two modified reaction systems 40, 90 is then calculated by using the mean absolute percentage error between the descriptor vector 60 of the modified reference system (modified reference reaction system 40) and the descriptor vector 110 of the test system (modified second reaction system or test reaction system 90) (step 120).

[0088] This distance 130 is then handed over to a Bayesian optimizer (as implemented in a Python library), including the actual composition of the second reaction system 70. Depending on the value of the distance 130, the Bayesian optimizer now decides (step 140) to adjust the test reaction system 70 (step 150) or not to adjust the test reaction system 70 and propose a recipe (step 160). The decision criterion for the Bayesian optimizer to make a decision is to stop after a maximum of 1000 iterations, when no significant changes in the distance metric are recorded. If the Bayesian optimizer decides to adjust the test reaction system 70, an updated test reaction system 180 is obtained by adjusting the test reaction system for the next optimization step (step 150). The adjustment can be achieved by an optimizer that proposes new starting materials / mass fractions based on the optimization history.

[0089] The process 70-140 is iterated for about 1000 iterations / cycles until no change in the distance 130 is observed or the change is below a defined threshold. The final set of descriptor values ​​110 is obtained by the minimum distance between the descriptor values ​​60, 110 of the reference test system / optimized test system 40, 90, i.e., the modified first reaction system and the second reaction system (reference reaction system and test reaction system):

[0090]

[0091] Table 1

[0092] As a result, a formulation of a second reaction system (test reaction system) is proposed (step 160) having the following chemical composition: ditrimethylolpropane (0.48), 1,2-propylene glycol (0.14), ethylene glycol (0.04) and isophthalic acid (0.33), which is the final test reaction system information 170. The weight fraction of glycerol is optimized to such a low value that it is considered unnecessary for the proposed formulation. This proposed test reaction system 170 is then communicated to the user for further laboratory testing.

Claims

1. A computer-implemented method for designing a starting composition of a chemical reaction system to maximize similarity to a reference reaction system to replace individual components in the reference system, the method comprising: A) providing information about a reference reaction system (10), the information comprising components of the reference reaction system (10) and their relative amounts in the reference reaction system (10); B) providing information about the test reaction system (70), the information including components of the test reaction system (70) and their relative amounts in the test reaction system (70); C) providing at least one first rule (20) for forming and / or breaking chemical bonds from components of the reference reaction system (10) and / or species formed from components of the reference reaction system (10); D) applying (30) the at least one first rule (20) to a reference reaction system, thereby obtaining a modified reference reaction system (40); E) providing at least one descriptor (35) of at least a portion of the modified reference reaction system (40) and at least a portion of the modified test reaction system as described in step H); F) determining (50) at least a portion of the descriptor value(s) (60) of the modified reference reaction system (40) from the at least one descriptor (35); G) providing at least one second rule (20) for forming and / or breaking chemical bonds from components of the test reaction system (70) and / or species formed from components of the test reaction system (70); H) applying (80) the at least one second rule (20) to the test reaction system (70), thereby obtaining a modified test reaction system (90); I) determining (100) at least a portion of the descriptor value(s) (110) of the modified test reaction system (90) from at least one descriptor (35) as described in step E); J) determining (120) a distance and / or dissimilarity (130) between the values ​​(60, 110) of at least one descriptor (35) from steps F) and I) using a measure of distance in the descriptor space occupied by the descriptors (35); K) deciding (140) whether to adjust the test reaction system (70) based on the distance and / or dissimilarity results determined in step J); L) If an adjustment has been decided in step K), the test reaction system (70) is adjusted (150) by changing the components and / or their relative amounts in the test reaction system (70) to obtain an updated test reaction system (180), and then the method is repeated starting from step H) based on the updated test reaction system (180).

2. The method according to claim 1, wherein the components of the reference reaction system (10) as described in step A) and / or the components of the test reaction system (70) as described in step B) include alkylene oxides, multifunctional carboxylic acids, carboxylic anhydrides, cyclic ethers, carbon dioxide, cyclic carbonates, polyols, polyisocyanates, polyamines, aromatic hydrocarbons, olefins, (meth)acrylates, bisphenols, phosgene, dialkyl carbonates, aldehydes, lactams, glycolides, amino acids, hydroxy-substituted carboxylic acids, or a combination of at least two of the foregoing components.

3. The method according to claim 1 or 2, wherein the at least one first rule (20) mentioned in C) and / or the at least one second rule (20) mentioned in G) comprises: Which functional groups of the components are present, which functional groups of the components can be used to form or break covalent bonds, which functional groups of the components can react with other functional groups, threshold criteria when formation or breaking of covalent bonds occurs, or a combination of at least two of the aforementioned rules (20).

4. The method according to any of the preceding claims, wherein at least one first rule (20) of step C) is identical to at least one second rule (20) of step G).

5. A method according to any of the preceding claims, wherein applying (30) at least one first rule (20) as described in step D) and / or applying at least one second rule (20) as described in step H) comprises running a kinetic Monte Carlo simulation, running a molecular Monte Carlo simulation, running a molecular dynamics simulation, running a Miller-Macosko type calculation.

6. A method according to any one of the preceding claims, wherein the modified reference reaction system (40) comprises a set of discrete components and information about the discrete components of the modified reference reaction system (40) and their relative amounts in the modified reference reaction system (40), wherein the step F) of determining (50) at least part of (one or more) descriptor values ​​(60) of the modified reference reaction system (40) by at least one descriptor (35) is preferably performed using atomic resolution information of each discrete component.

7. A method according to any one of the preceding claims, wherein the modified test reaction system (90) comprises a set of discrete components and information about the discrete components of the modified test reaction system (90) and their relative quantities in the modified test reaction system (90), wherein the determining (100) of at least part of (one or more) descriptor values ​​(110) of the modified test reaction system (90) by at least one descriptor (35) in step I) is preferably performed using atomic resolution information of each discrete component.

8. The method according to any one of the preceding claims, wherein the at least one descriptor (35) as described in step E) is a statistical / compositional descriptor, preferably number average molecular weight, weight average molecular weight, viscosity average molecular weight, polydispersity index, average functionality, OH value, acid value; and / or Topological descriptors, preferably Wiener index, Randic connectivity index, Balaban-J index, Ipc index, Zagreb index, Bertz index, chi molecular connectivity index, Kier-Hall valence connectivity index, seniority and priority index, Kappa shape index, BCUT index, Estate index, walk and path counts, extended connectivity fingerprint; and / or Molecular descriptors, preferably Labute approximate surface area, solvent accessible surface area, radius of gyration, 2D autocorrelation, eigenvalue-based descriptors, RDF descriptors, 3D-MoRSE descriptors, WHIM descriptors, GETAWAY descriptors, functional group count, number of rotatable bonds, number of hydrogen acceptors and donors, Gasteiger / Marsili partial charges, topological polar surface area, molar refractive index; and / or Force field based descriptors, preferably intramolecular potential energy, radial distribution function, bond angle distribution function, dihedral angle distribution function; and / or Quantum chemical descriptors, preferably atomic charge, HOMO and LUMO energies, molecular hardness, molecular polarizability, dipole moment, total energy, molecular quantum number descriptors; and / or Thermodynamic descriptors, preferably octanol-water partition coefficient, sigma distribution, sigma moment, polar moment vector, activity coefficient, chemical potential, free energy, octanol-water partition coefficient, solubility; and / or Health, sustainability or environment related descriptors, preferably toxicity, CO2 equivalent, carbon footprint, energy consumption, product recycling rate, sustainability index, supply chain mileage, water footprint, price; Or a combination of at least two of the foregoing descriptors.

9. The method according to any of the preceding claims, wherein the distances and / or dissimilarities (130) as described in step J) comprise Euclidean / L2 distance, Manhattan / L1 / city block distance, Canberra distance, Chebyshev distance, Mahalanobis distance, Minkowski distance, Rogers-Tanimoto dissimilarity, Russell-Rao dissimilarity, Sokal-Michener dissimilarity, Sokal-Sneath, mean absolute percentage error, Yule dissimilarity, cosine dissimilarity, Dyss dissimilarity or a combination of at least one of the aforementioned distances and / or dissimilarities.

10. The method according to any of the preceding claims, wherein the decision as described in step K) comprises one or more of the following decision criteria: The distance and / or dissimilarity thresholds in step J), a threshold value for a change in distance and / or dissimilarity relative to one or more previously performed steps J), a threshold for the execution time of the method, a threshold value for the number of repetitions of the method, A threshold value for at least one of the descriptors.

11. A method according to any of the preceding claims, wherein adjusting (150) the test reaction system (70) by changing the components of the test reaction system (70) and / or their relative amounts in the test reaction system (70) as described in step L) includes executing an algorithm selected from the following: random search, grid search, Bayesian optimization, simplex optimization, evolutionary optimization, genetic algorithm, particle swarm optimization, Metropolis-Hastings Markov chain Monte Carlo, adaptive Markov chain Monte Carlo, simulated annealing, parallel tempering, hybrid linear and nonlinear programming, or a combination of at least two of the preceding algorithms.

12. The method according to claim 1, further comprising communicating the final test reaction system (170) to the user if it has been decided in step K) not to adjust the test reaction system in step K).

13. A data processing apparatus comprising means for performing the method of any one of claims 1 to 12.

14. A computer program product comprising instructions which, when a computer executes said program, cause the computer to perform the method of any one of claims 1 to 12.

15. A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 12.

Images