Deep neural networks for biodegradability

Abstract

The present disclosure relates to the field of biodegradable materials or molecules, such as small molecules or macromolecules or formulations, using machine learning. Methods, apparatus, computer elements, biodegradable materials, or uses for generating biodegradable properties of one or more biodegradable materials are disclosed.

Description

Technology Field

[0001] The present disclosure relates to the field of biodegradable materials or molecules, such as small molecules or macromolecules or formulations, using machine learning. Methods, apparatus, computer elements, materials, or uses for generating biodegradable properties of one or more materials are disclosed. Background Technology

[0002] In general, functional chemical compounds, such as those referring to small molecules, are widely used in industrial and / or everyday products due to their extensive application characteristics. The uses of functional chemical compounds encompass, among others, coatings, personal care products, cleaning detergents, lubricants, packaging, and foams. However, these extensive applications, on the other hand, result in enormous amounts of waste containing the used functional chemical compounds. This becomes a problem when non-biodegradable waste is disposed of in undesignated environments. In particular, the accumulation of chemicals in the environment, such as phosphate buildup leading to algae blooms, is undesirable. Therefore, there is a need for biodegradable functional chemical compounds.

[0003] In one aspect, a method for generating at least one biodegradable property characterizing at least one material, comprising the following steps, is disclosed, in particular a computer-implemented method:

[0004] - A step of providing at least one substance-specific graph representation generated by providing the numeric graph representation(s) to at least one data-based graph representation model configured or trained to map the numeric graph representation(s) associated with at least one atom and bond of the at least one substance to at least one substance-specific graph representation associated with the at least one atom, bond, and correlations associated with the atom and / or bond;

[0005] - A step of generating at least one biodegradable property by providing the at least one material-specific graph representation to at least one data-based property model configured or trained to map the at least one material-specific graph representation to at least one biodegradable property;

[0006] - A step of providing the at least one biodegradable property generated to characterize the biodegradability of the at least one material.

[0007] In another aspect, a method is disclosed for an apparatus that produces at least one biodegradable property characterizing at least one material, comprising the following steps:

[0008] - An input interface configured to provide at least one material-specific graph representation generated by providing numeric graph representation(s) to at least one data-based graph representation model configured or trained to map numeric graph representation(s) associated with at least one atom and bond of the at least one material to at least one material-specific graph representation associated with at least one atom, bond, and correlations associated with the atom and / or bond;

[0009] - A feature generator configured to generate at least one biodegradable property by providing the at least one material-specific graph representation to at least one data-based feature model that is configured or trained to map the at least one material-specific graph representation to at least one biodegradable property;

[0010] - An output interface configured to provide the generated at least one biodegradable property to characterize the biodegradability of the at least one material.

[0011] In another aspect, a use of the selected biodegradable properties and associated material structure produced according to the method disclosed herein or by the apparatus disclosed herein for producing a biodegradable material having a selected material structure is disclosed.

[0012] In another aspect, a computer element, such as a computer-readable storage medium, a computer program, or a computer program product is disclosed, comprising an instruction that instructs the computing node or computing system to perform the method disclosed herein when executed by the computing node or computing system.

[0013] Any disclosures and embodiments described herein relate to methods, apparatus, surfactants, uses, and computer elements. Advantageously, the benefits provided by any embodiments and examples apply equally to all other embodiments and examples.

[0014] Mode of implementation

[0015] In the following, embodiments of the present disclosure will be outlined as embodiments and / or examples. It should be understood that the present disclosure is not limited to the embodiments and / or examples.

[0016] The method, method, apparatus, computer element, biodegradable material, or use disclosed herein enables more reliable and standardized production of the biodegradable properties of the biodegradable material.

[0017] The substance may be a functional chemical compound, such as a small molecule, a large molecule, such as a polymer or oligomer, a formulation, or a combination thereof, or may include such.

[0018] Functional chemical compounds may be any functional chemical compound or may include such compounds. Functional chemical compounds are generally chemical compounds that possess application properties for technical purposes, that is, chemical compounds that fulfill a specific function in a chemical product. For example, a chemical compound that provides UV protection in a UV blocking material is a functional chemical compound. Functional chemical compounds may include an active ingredient, that is, a component that provides the function of the functional chemical compound. Additionally, the active ingredient may provide the biological activity of the functional chemical compound. For example, the active ingredient may refer to antifungal agents, aromatic chemicals, UV absorbers, food additives, vitamins, nutrients, dyes, and surfactants. Functional chemical compounds are generally characterized by their chemical structure. However, different chemical structures may also exist within a single functional chemical compound. For example, a functional chemical compound may consist of a component that is a molecule undergoing tautomerization, protonation, or deprotonation. A functional chemical compound may consist of more than one stereoisomer. Accordingly, a functional chemical compound composed of a type of molecule may be associated with one or more chemical structures, for example, one protonated structure and one uncharged structure, or two stereoisomers. Thus, generally, a functional chemical compound may include all molecules associated with a single chemical formula through (de)protonation, isomerization, such as tautomerization and stereoisomerization. Additionally, a functional chemical compound may refer to any functional chemical compound that can be described by one or more chemical structures. In one embodiment, the chemical structures may be associated with a single chemical formula.

[0019] Preferably, the functional chemical compound consists of small molecules. Preferably, the functional chemical compound has a molecular weight of less than 10,000 g / mol. More preferably, the chemical compound has a molecular weight of less than 800 g / mol, and even more preferably less than 400 g / mol. Additionally, it is desirable for the functional chemical compound to exist in the environment in a form that allows the molecule of the functional chemical compound to be fully described using a simple structural formula containing relevant information. A simple molecular structure refers to a molecule that can be clearly described by the covalent bonds between the atoms of the molecule. Examples that are not such are systems having a dynamic equilibrium between various forms, such as monomers and oligomers, as in the case of various inorganic acids, for example, or ionic species with highly localized charges that interact strongly with the solvent through hydrogen bonding, for example. Preferably, the functional chemical compound has at least one of the following characteristics: affecting living organisms, being suitable for affecting structure, or being suitable for affecting the function of living organisms.In one embodiment, the functional chemical compound comprises at least one of the following functional groups: ether group, hydroxyl group, peroxo group, hydroperoxide group, carboxyl group, carboxyl group derivative, carbonyl group, amine group, imine group, hydrazine group, urea group, urethane group, thiourethane group, nitrile group, azide group, azo group, cyanate group, isocyanate group, isocyanide group, pyridine group, alkane group, alkene group, alkyne group, phenyl group, ketone group, thioketone group, aldehyde group, thioaldehyde group, acetal group, ketal group, oxime group, hydrazone group, nitro group, nitroso group, thiol group, sulfide group, disulfide group, sulfonic acid derivative, sulfinic acid derivative, sulfate group, sulfate group derivative, sulfone group, sulfoxide group, sulfhydryl group, sulfide group, phosphoran A group, a phosphate group, a phosphate acid derivative, a phosphonate, a phosphine group, a silane group, a silazane group, a silicon group, a borate group, a borane group, a halogenide group, or a combination thereof. Preferably, the functional chemical compound corresponds to one of the following compound classes: a carboxyl group derivative, an ether group, an amine group, a hydroxyl group, a carbonyl group, an alkane group, an alkene group, a benzene derivative, a pyridine derivative, or a halogenide group.

[0020] Generally, macromolecules may be or include polymers and / or oligomers. A polymer or oligomer may include one or more subgroups, all of which together form a polymer or oligomer. For example, a subgroup may refer to a portion of a polymer or oligomer, which are connected together continuously along a chain or network to form a polymer. Preferably, a subgroup of a polymer refers to a repeating unit describing a portion of a polymer that, when repeated, creates a polymer chain. However, in some cases, a subgroup may refer to a single portion of a polymer or oligomer that is not repeating. A subgroup may include a repeating portion, for example, a subgroup of a polymer may include a repeating core present in other subgroups, and additional additional portions present in other subgroups that are not repeating. A subgroup may include at least one of a polymerized monomer or an oligomer fragment. A subgroup may include a polymerized monomer. In this regard, a polymerized monomer refers to a monomer after polymerization, which is sometimes also called a "mer unit" or "mer." In particular, the polymerized monomer does not refer to the monomer present in the reaction mixture before polymerization, i.e., the raw material, but rather to a repeating unit derived from the monomer that has been changed during or after polymerization.

[0021] The formulation may be any formulation comprising two or more chemical and / or biological components. The formulation comprises at least two components that may include any chemical and / or biological entities. For example, the components may include small molecules, polymers, etc. However, the components themselves may also be more complex chemical products.

[0022] Biodegradable properties may characterize at least one material in relation to its biodegradable behavior. Biodegradable properties may relate to the classification of biodegradable and / or non-biodegradable materials in relation to potentially one or more habitat type(s) and / or condition(s). Habitat types may relate to the environment or substrate where the biodegradation process or biodegradability is of interest. Habitat types may include soil, compost, sewage, aquatic systems, or other suitable environments or substrates of interest. Habitat conditions may relate to the conditions where the biodegradation process or biodegradability is of interest. This may include nutrient concentrations, temperature, pH, pO2, ionic conditions, toxicity, or other suitable conditions affecting the biodegradation process in the habitat.

[0023] Biodegradability properties may be related to quantities that characterize the time progression of the biodegradation process. Biodegradability properties may be related to measured quantities that characterize the time progression of the biodegradation process. Biodegradability properties may include reference quantities measured under reference measurement conditions. Reference quantities and reference measurement conditions may be provided by OECD, ASTM, ISO standards or other applicable standards, such as those cited in relation to FIG. 1.

[0024] In contrast to numerical graph representation(s), at least one material-specific graph representation is generated by at least one data-driven graph representation model based on at least one graph neural network architecture. Numerical graph representations may, in some cases, be generated based on predefined feature vectors encoding material structure and / or material composition. As a result, numerical graph representation(s) associated with at least one atom and bond of at least one material may be generated based on predefined feature vectors for at least one atom and bond of said material. Feature vectors may, in some cases, represent nodes and edges corresponding to material structure and / or material composition, and these feature vectors may form numerical graph representations. Meanwhile, material-specific graph representations are associated with at least one atom, bond, and correlations related to the atom and / or bond. In other words, the data-driven graph representation model may generate a representation of the material that considers correlations related to the atom and / or bond, and thus considers details related to the material structure and / or composition, including correlations between structural or compositional elements.

[0025] In one embodiment, at least one material specification associated with at least one material is provided, wherein the at least one specification is mapped to a corresponding numerical graph representation(s) associated with at least one atom and bond of at least one material.

[0026] In another embodiment, at least one data-based graph representation model is trained by generating numeric graph representation(s) associated with one or more augmented material(s) per material graph representation, wherein the augmented material includes at least one synthetic change in the numeric graph representation associated with at least an atom, bond, atomic-bond combination, or material component.

[0027] In another embodiment, at least one data-based graph representation model is trained using a contrast loss function that depends on the distance between numeric graph representation(s) of different materials and / or augmented materials.

[0028] In another embodiment, a data-based characteristic model is trained on a training dataset comprising at least one biodegradability characteristic measured for one or more material(s) and material-specific graph representation(s) generated by at least one data-based graph representation model.

[0029] In another embodiment, at least one material type related to a small substance, a large molecule, such as a polymer or oligomer and / or a formulation is provided to at least one data-based graph representation model and / or at least one data-based characteristic model and / or is provided to select at least one data-based graph representation model and / or at least one data-based characteristic model.

[0030] In another embodiment, at least one measurement type related to the measurement of biodegradability properties is provided to at least one data-based graph representation model and / or at least one data-based property model and / or is provided to select at least one data-based graph representation model and / or at least one data-based property model.

[0031] In another embodiment, based on a material-specific graph representation associated with at least one material having generated biodegradable properties, one or more similar material-specific graph representation(s) are generated based on a representation distance scale, wherein one or more similar material-specific graph representation(s) and corresponding material(s) are provided.

[0032] In another embodiment, a material map comprising a plurality of material-specific graph representation(s) associated with a plurality of materials and any corresponding generated biodegradable properties is generated based on a representation distance scale. A material map comprising a map of materials for a representation distance scale and any corresponding generated biodegradable properties may be provided.

[0033] In another embodiment, one or more habitat condition(s) affecting one or more biodegradable properties are provided, wherein at least one data-based property model is trained on one or more associated biodegradable properties that depend on material-specific graph representation(s) and one or more habitat condition(s). One or more habitat condition(s) may be mapped to numerical habitat condition(s). One or more numerical habitat condition(s) may be fused with material-specific graph representation(s). The fused material-specific graph representation(s) may be provided to at least one data-based property model. In this regard, being fused may involve any vector or matrix operation, such as concatenation, addition, inner product, or other arithmetic vector or matrix transformation.

[0034] In another embodiment, one or more habitat condition(s) affecting one or more biodegradable properties are provided. At least one data-driven graph representation model can be trained on numeric graph representation(s) and one or more associated habitat condition(s). One or more habitat condition(s) can be mapped to numeric habitat condition(s). One or more numeric habitat condition(s) can be fused with numeric graph representation(s). The fused numeric graph representation(s) can be provided to at least one data-driven graph representation model.

[0035] In another embodiment, the biodegradability property relates to at least one material structure associated with a non-biodegradable or biodegradable material comprising at least one functional chemical compound, at least one macromolecule, at least one formulation, or a combination thereof, or at least one small molecule, at least one polymer, and / or at least one formulation. The biodegradability property may be associated with a biodegradable or non-biodegradable class. The biodegradability property may be associated with a biodegradability measurement related to a reference measurement and / or a reference habitat. The biodegradability property may be associated with a biodegradability measurement based on the chemical structure and / or composition of the material.

[0036] In another embodiment, the biodegradability is related to at least one application of the material and / or at least one application performance measurement of the material, based on the chemical structure and / or composition of the material. Brief explanation of the drawing

[0037] Hereinafter, the present disclosure is further described with reference to the accompanying drawings: Figure 1 illustrates an example of a biodegradable material. Figure 2 illustrates an exemplary method for generating a numerical material representation based on chemical structure specifications. FIG. 3 illustrates an exemplary system for generating at least one biodegradable property based on a numerical graph representation. Figure 4 illustrates an example of a training process for a data-driven graph representation model. Figure 5 illustrates an example of a training process for a data-driven feature model. Figures 6 and 7 show the output of the trained model. Figures 8 and 9 illustrate the comparison results from differently trained models. Specific details for implementing the invention

[0038] Figure 1 illustrates an example of a biodegradable material.

[0039] Biodegradable materials can be designed to decompose upon disposal through the action of living organisms. Biodegradability may be related to the environmental fate and / or behavior of the material. Biodegradability may be related to the extent to which the material can be degraded by microorganisms, such as bacteria, fungi, or algae. Biodegradability may depend on the material's chemical structure, chemical weight, physical factors, such as crosslinking density, branching, crystallinity, or solubility, and exposure conditions, such as habitats like soil, compost, or aquatic systems. In relation to exposure conditions, microorganisms, microbial populations, nutrient concentrations, temperature, pH, pO2, ionic conditions, or substrate characteristics, such as toxicity, affect biodegradability. Biodegradability can be measured based on measured mass loss (mg / hour), dissolved organic carbon (DOC, organic carbon concentration / hour), oxygen consumption (e.g., via pressure measurements, e.g., Pa / hour), or carbon dioxide generation over time (e.g., via pressure measurements, e.g., Pa / hour). Figure 1 illustrates some biodegradable and non-biodegradable materials, including small molecules and polymers. Despite having similar structures at first glance, very different biodegradable properties are exhibited depending on the material.

[0040] Quantifying biodegradability in terms of the measured properties of a material is difficult, and many measurement standards have been developed. Different measurement methods are defined to determine biodegradability under predefined laboratory conditions. For example, regarding wastewater, OECD Test No. 301: "Ready Biodegradability" (July 17, 1992) describes six methods for determining biodegradability. Additionally, for example, ASTM D5988-18 "Standard test method for determining aerobic biodegradation of plastic materials in soil" describes measuring the biodegradability of a reference material by measuring carbon dioxide produced by microorganisms as a function of exposure time. In addition, for example, ISO 17556:2019 “plastics-determination of the ultimate aerobic biodegradability of plastic materials in soil by monitoring the oxygen demand in a respirometer or the amount of carbon dioxide evolved” calculates the optimal biodegradation rate of plastic materials in test soil by controlling oxygen consumption or carbon dioxide generation.In addition, for example, ISO 14855-1:2012 "determination of the ultimate aerobic biodegradability of plastic materials under controlled composting conditions—method by analysis of evolved carbon dioxide—Part 1: General method" and ASTM D5338-15 "standard test method for determining aerobic biodegradation of plastic materials under controlled composting conditions, incorporating thermophilic temperatures" determine the ultimate aerobic biodegradability (means of microorganisms completely consuming chemical or organic material in the presence of oxygen) of plastics based on organic compounds under controlled composting conditions by measuring the percentage of carbon converted to carbon dioxide and the degree of degradation of the plastic at the end of the test. ASTM D6400-21 "Standard specification for labeling of plastics designed to be aerobically composted in municipal or industrial facilities" additionally includes elemental analysis, plant germination (phytotoxicity), and mesh filtration of the generated particles.ISO 17088:2021 “plastics-organic recycling-specifications for compostable plastics” includes an assessment of negative consequences for the composting process and facilities and negative impacts on the quality of the produced compost, including the presence of high levels of regulated metals and other hazardous components.

[0041] Regarding aerobic biodegradation, ISO 18830:2016 "plastics-determination of aerobic biodegradation of non-floating plastic materials in a seawater / sandy sediment interface-method by measuring the oxygen demand in closed respirometer" and ISO 19679:2020 "plastics-determination of aerobic biodegradation of non-floating plastic materials in a sea-water / sediment interface-method by analysis of evolved carbon dioxide" have been developed. Biodegradation is assessed by oxygen demand or CO2 emissions.Additional standards include, for example, ISO 14853:2016 "plastics - determination of the ultimate anaerobic bio-degradation of plastic materials in an aqueous system - method by measurement of biogas production", ISO 23977-1:2020 "plastics - determination of the aerobic biodegradation of plastic materials exposed to seawater - Part 1: method by analysis of evolved carbon dioxide", and ISO 23977-2:2020 "plastics - determination of the aerobic biodegradation of plastic materials exposed to seawater - Part 2: method by measuring the oxygen demand in a closed breathing system" Includes "oxygen demand in closed respirometer".

[0042] The quantified biodegradability of a material may vary depending on the measurement methods and conditions used, the measurement environment, and measurements related to the degradation process, such as mass loss over time, DOC, oxygen consumption, or carbon dioxide generation. The measurement methods and measured characteristics may be provided as metadata for measurement points related to biodegradability.

[0043] Figure 2 illustrates an exemplary method for generating a numerical material representation based on chemical structure specifications.

[0044] The chemical structural specifications of a biodegradable material can be mapped to a numerical graph representation. Such mapping may involve the determination of feature vectors based on predefined feature specifications. Predefined feature specifications for the chemical structure may include, but are not limited to, atomic types, atomic arrangements, e.g., rings or chains, hybridization, number of bonds, bond types, bond arrangements, e.g., rings or chains, conjugation, stereochemistry, etc. Features may be implemented as one-hot encoding, or in other words, based on predefined feature specifications. The predefined feature specifications may relate to atomic features corresponding to atoms in the chemical structure. Atomic features may be represented as nodes or vertices in the numerical graph representation. The predefined feature specifications may relate to bonding features corresponding to bonds in the chemical structure. Bonding features may be represented as edges or arcs in the numerical graph representation. That is, the graph representation may include nodes corresponding to atoms and vertices corresponding to bonds between two atoms. Feature vectors can be assigned per node and vertex representing atomic type, e.g., C atoms or orbital hybridization, and bond type, e.g., double bonds or ring structures. Each node can encode atomic information such as atomic type, aromaticity, hybridization, the number of bonds connecting atoms, and the number of bonded hydrogen atoms to implicitly handle hydrogen. In one embodiment, atomic type can be one-hot encoded into a predefined number of categorical features based on a predefined list of chemical elements. Edge features (e.g., bond type) can be explicitly included through the bond type, where the bond is part of the ring, bonding, and stereochemistry. This type of graph data representation can generate a predefined number of categorical features per atom and per bond.

[0045] Based on feature vectors, chemical structures can be represented as matrices as numerical representations. For example, feature matrices and / or adjacency matrices can be generated from feature vectors. In this way, a substance can be represented as a chemical graph where nodes correspond to atoms and edges correspond to bonds between two atoms. By assigning feature vectors containing information regarding atomic types and bond types to each node and each edge, the chemical structure can be mapped from a graph representation, for example via SMILES, to a numerical representation, for example via feature matrices and / or adjacency matrices. Feature matrices and / or adjacency matrices can represent the chemical structural specifications of a surfactant in a numerical graph representation that can be processed by a graph neural network (GNN).

[0046] For small substances, the numerical graph representation may include categorical features per atom and per bond. For macromolecules such as polymers, the numerical graph representation may include categorical features per atom, per bond, and per macromolecular arrangement.

[0047] In the case of a polymer, the monomer structure, including polymerized monomers and / or raw monomers in a non-polymerized state, may be represented, for example, as a chemical fingerprint encoded as a graph representation or binary vector as described above. The chain architecture of the polymer may be encoded based on monomer representation(s). The stoichiometry or polymer architecture may be represented by taking the sum of monomer representation(s) weighted according to their respective ratios. The stoichiometry or polymer architecture may be represented as an architecture vector of integer values ​​that captures the frequencies of different monomer patterns to reflect the stoichiometry of the monomers, for example, including polymerized monomers and / or raw monomers in a non-polymerized state. The graph chemical representation may further include edges to describe the average structure of repeating units weighted according to the probability of occurrence in the polymer. This can reflect (i) the repeating nature of the polymer's repeating units, (ii) different topologies and isomers of the polymer chains, and (iii) various monomer compositions and stoichiometry. A polymer graph representation may include one or more edge(s) associated with weights that reflect the representation of atoms and bonds per repeating unit and / or the probability or frequency of the existence of bonds per repeating unit. For example, by connecting individual monomers, including polymerized monomers and / or raw monomers in an unpolymerized state, with weighted edges, not only the repeating nature of the polymer chains but also an ensemble of possible chain architectures can be represented. One possible implementation is described, for example, in the literature [Matteo Aldeghi and Connor W. Coley, "A graph representation of chemical ensembles for polymer property prediction", Chem. Sci., 2022, 13, 10486-10498].

[0048] In the case of a formulation, the formulation components or constituents, their mass ratios, and interactions between the formulation components can be represented by a chemical graph representation. For example, the formulation components can be represented by the small molecule or large molecule graph representations described above. The formulation composition can be represented by edges having weights representing relative concentrations, interactions between formulation components, and / or probabilities associated with concentrations or interactions. In this way, the chemical structure can be represented by numeric vectors and / or matrices that can function as input representations for a graph neural network to generate material-specific graph representations.

[0049] FIG. 3 illustrates an exemplary system for generating at least one biodegradable property based on a numerical graph representation.

[0050] The system includes at least one data-driven graph representation model configured to map numerical graph representations associated with atoms and bonds of at least one material(s) to material-specific graph representations associated with atoms, bonds, and correlations related to atoms and bonds of the material(s), and at least one data-driven property model configured to map material-specific graph representations to at least one biodegradable property. A training process for the data-driven model is described in more detail below. Essentially, this process comprises two steps: 1) the data-driven graph representation model generates material-specific graph representations for one or more materials, and 2) the material-specific graph representations can be used in the second step to generate at least one biodegradable property. Thus, the architecture enables the generation of material-specific graph representations for multiple materials. These material-specific graph representations may be stored in relation to material specifications. Next, using material-specific graph representation(s), biodegradable properties can be generated according to measurement type, such as habitat type, measurement method, measurement conditions, property type, application type, etc. This enables a more efficient setup for generating biodegradable properties, as material-specific graph representation(s) can be generated once, and different data-driven property models can be trained and used based on this material-specific graph representation space provided by the GNN.

[0051] A computing system for generating at least one biodegradable property may include a user interface configured to provide requests or instructions related to a material for which at least one biodegradable property is to be generated. For example, at least one material specification may be provided in a format such as a SMILES string, a graphical representation, and / or a numerical representation associated with the material.

[0052] Biodegradability properties may relate to at least one material structure associated with a non-biodegradable or biodegradable material, for example, including at least one small molecule, at least one polymer, and / or at least one formulation. Biodegradability properties may relate to biodegradable or non-biodegradable classes. Biodegradability properties may relate to biodegradability measurements associated with reference measurements and / or reference habitats, such as mass loss (mg / hour), dissolved organic carbon (DOC, organic carbon concentration / hour), oxygen consumption (e.g. via pressure measurements, e.g., Pa / hour), or carbon dioxide generation over time (e.g. via pressure measurements, e.g., Pa / hour). Biodegradability properties may relate to biodegradability measurements based on the chemical structure and / or composition of the material. Biodegradability properties may relate to at least one application of the material. Biodegradability properties may relate to at least one application performance measurement of the material based on the chemical structure and / or composition of the material. Examples of applications and / or application performance measurements are diverse and may include polymers or formulations used in cosmetics or personal care, such as surfactants, and related performance measurements, such as foaming properties, foaming rate, etc.

[0053] A request or instruction may be processed by a mapping agent configured to map one or more material specification(s) to corresponding material-specific graph representation(s). The mapping agent may be configured to provide material-specific graph representation(s) based on a request containing one or more material specification(s). The mapping agent may be configured to provide material-specific graph representation(s) generated by providing numeric graph representation(s) to at least one data-based graph representation model configured to map numeric graph representation(s) associated with atoms and bonds of chemical structure(s) to material-specific graph representation(s) associated with atoms, bonds, and correlations related to atoms and bonds of chemical structure(s). The material-specific graph representation(s) may be stored in a structure store containing the material-specific graph representation(s) and the associated material specification(s). The material-specific graph representation(s) may be pre-generated and stored by at least one data-based graph representation model. If material-specific graph representation(s) for the requested material specification(s) are stored in the structure store or are pre-generated, the mapping agent may be configured to provide material-specific graph representation(s) corresponding to one or more material specification(s). If material-specific graph representation(s) for the requested material specification(s) are not stored in the structure store or are not pre-generated, the mapping agent may be configured to request model execution by the model execution engine. The mapping agent may be configured to provide one or more material specification(s) to the model execution engine. The model execution engine may be configured to access, for example, a trained data-based graph representation model stored in the model store and generate material-specific graph representation(s) based on this access.

[0054] The request may include at least one material type related to small substances, macromolecules, such as polymers or oligomers and / or formulations. At least one material type may be provided with or derived from one or more material specification(s). At least one material type may be provided to at least one data-driven graph representation model. At least one material type may be provided to select at least one data-driven graph representation model. The model store may include one or more data-driven graph representation models dependent on material types. The model store may include one or more data-driven graph representation models trained on training data related to material types. The model store may include one or more data-driven graph representation models per material type dependent on material subclasses. For example, in the case of small molecules as material types, the models may vary depending on the material class related to the type of molecule, the functionality of the molecule, the application of the molecule, etc. For example, in the case of polymers as a material type, the model may vary depending on the material class related to the application, such as the type of monomer including polymerized monomers and / or raw monomers in an unpolymerized state, the number of monomers per repeating unit, e.g., copolymers, terpolymers, quaternary copolymers, etc. For example, in the case of formulations as a material type, the model may vary depending on the material class related to the type of component, solution, number of component, etc.

[0055] The request may include at least one type of measurement related to the measurement of biodegradability properties. At least one type of measurement may be provided with or derived from one or more material specification(s). At least one type of measurement may relate to one or more measured quantities / amounts, e.g., but not limited to, mass loss (mg / hour), dissolved organic carbon (DOC, organic carbon concentration / hour), oxygen consumption (e.g. via pressure measurement, e.g. Pa / hour), or carbon dioxide generation over time (e.g. via pressure measurement, e.g. Pa / hour). At least one type of measurement may relate to one or more measurement methods, e.g., but not limited to, measurement methods based on reference measurement setups such as those provided in the exemplary standards cited in connection with FIG. 1. At least one type of measurement may relate to one or more measurement conditions, e.g., but not limited to, reference measurement conditions such as those provided in the exemplary standards cited in connection with FIG. 1. At least one measurement type may relate to one or more habitat type(s) and / or conditions, e.g., non-limitingly, reference measurement habitat types and / or conditions as provided in exemplary standards cited in connection with FIG. 1, e.g., FIG. 1. At least one measurement type may be provided to at least one data-driven graph representation model. At least one measurement type may be provided to select at least one data-driven graph representation model. The model store may include one or more data-driven graph representation models dependent on the measurement type. The model store may include one or more data-driven graph representation models trained on training data related to the measurement type.

[0056] One or more habitat type(s) and / or condition(s) affecting one or more biodegradable properties may be provided. At least one data-driven graph representation model may be trained on numerical graph representation(s) and the associated one or more habitat type(s) and / or condition(s). One or more habitat type(s) and / or condition(s) may be mapped to numerical habitat condition(s) through predefined feature vectors, for example, as described in relation to FIG. 2. One or more numerical habitat type(s) and / or condition(s) may be fused with numerical graph representation(s) associated with the material and habitat. The fusion may represent the material and habitat, such as the formulation representation described in relation to FIG. 2. Substrate components, material components, and their interrelationships may be represented by predefined feature vectors, and a graph representation matrix may be generated. The fused numerical graph representation(s) may be provided to at least one data-driven graph representation model. The data-driven graph representation model may therefore vary according to one or more habitat type(s) and / or condition(s) and may be selected based on one or more habitat type(s) and / or condition(s).

[0057] A mapping agent may be configured to provide one or more material specification(s) to a model execution engine. The model execution engine may be configured to access a trained data-based feature model, for example, stored in a model store, and to generate at least one biodegradable feature based on this access. Based on the provided material-specific graph representation(s), the model execution agent may be configured to generate at least one biodegradable feature by providing the material-specific graph representation(s) to at least one data-based feature model configured to map the material-specific graph representation(s) to at least one biodegradable feature.

[0058] At least one material type may be provided with or derived from one or more material specification(s). At least one material type may be provided to at least one data-based characteristic model. At least one material type may be provided to select at least one data-based characteristic model. The model store may include one or more data-based characteristic models dependent on material types. The model store may include one or more data-based characteristic models trained on training data related to material types. The model store may include one or more data-based characteristic models per material type dependent on material subclasses.

[0059] At least one type of measurement may be provided with or derived from one or more material specification(s). At least one type of measurement may relate to one or more measured quantities / quantities, e.g., but not limited to, mass loss (mg / hour), dissolved organic carbon (DOC, organic carbon concentration / hour), oxygen consumption (e.g. via pressure measurement, e.g. Pa / hour), or carbon dioxide generation over time (e.g. via pressure measurement, e.g. Pa / hour). At least one type of measurement may relate to one or more measurement methods, e.g., but not limited to, measurement methods based on reference measurement setups such as those provided in the exemplary standards cited in connection with FIG. 1. At least one type of measurement may relate to one or more measurement conditions, e.g., but not limited to, reference measurement conditions such as those provided in the exemplary standards cited in connection with FIG. 1. At least one measurement type may relate to one or more habitat type(s) and / or conditions, for example, without limitation, as provided in the exemplary standards cited in connection with FIG. 1, for example. Habitat types may relate to soil, compost, aquatic systems, sewage, etc. Habitat conditions may relate to nutrient content, toxicity, or other characteristics of the habitat that affect biodegradation processes. At least one measurement type may be provided to at least one data-based characteristic model. At least one measurement type may be provided to select at least one data-based characteristic model. The model store may include one or more data-based characteristic models dependent on the measurement type. The model store may include one or more data-based characteristic models trained on training data related to the measurement type.

[0060] One or more habitat type(s) and / or condition(s) affecting one or more biodegradable properties may be provided, for example, upon request. At least one data-driven property model may be trained on one or more associated biodegradable properties dependent on material-specific graph representation(s) and one or more habitat type(s) and / or condition(s). One or more habitat condition(s) may be mapped to numeric habitat condition(s). One or more habitat type(s) and / or condition(s) may be mapped to numeric habitat condition(s) through predefined feature vectors, for example, as described in relation to FIG. 2. One or more numeric habitat condition(s) may be fused with material-specific graph representation(s). Fusion of numeric representations may include one or more operations such as concatenation, summation, inner product, etc. The fused material-specific graph representation(s) may be provided to at least one property model. Therefore, the characteristic model may vary according to one or more habitat type(s) and / or condition(s) and may be selected based on one or more habitat type(s) and / or condition(s).

[0061] A mapping agent and / or a model execution agent may be configured to provide at least one generated biodegradability property characterizing the biodegradability of the associated material(s) for one or more material specification(s). From material-specific graph representation(s) associated with material structure(s) having at least one generated biodegradability property, one or more similar material-specific graph representation(s) may be generated based on a representation distance scale. One or more similar material-specific graph representation(s) and corresponding biodegradable material structure(s) may be provided. For easy exploration and overview, one or more similar material-specific graph representation(s) and corresponding biodegradable material structure(s) may be provided in the form of a diffusion map. In this way, a material map containing multiple material-specific graph representation(s) associated with biodegradable material structures having the generated biodegradability property may be generated and displayed based on a representation distance scale.

[0062] At least one data-driven graph representation model and / or at least one data-driven characteristic model may be trained based on a training dataset relating to material specification(s) and corresponding numerical graph representation(s) for training the data-driven graph representation model, and material-specific graph representation(s) and corresponding biodegradable characteristic(s) for training the data-driven characteristic model. Examples of model architectures, training processes, and model features will be described in more detail in relation to the figures below. This should not be considered limiting, as numerous implementations exist and the examples are provided merely as exemplary examples.

[0063] Figure 4 illustrates an example of a training process for a data-driven graph representation model.

[0064] At least one data-driven graph representation model can be trained to map numeric graph representation(s) associated with at least atoms and bonds of the substance(s) to material-specific graph representation(s) associated with at least atoms, bonds, and correlations related to atoms and bonds of the substance(s). The data-driven graph representation model can be trained to generate numeric graph representation(s) for a single type of substance, e.g., small molecules, macromolecules, e.g., polymers or oligomers, formulations, or combinations of substance habitats.

[0065] For training purposes, material specification(s) may be provided and mapped to a numerical graph representation, for example, as described in relation to FIG. 2. The numerical graph representation per material may be augmented to represent a number of augmented material structures. The augmentation may include at least one synthetic change among atoms, bonds, atomic-bond combinations, components of macromolecules, components of small molecules, or combinations of components.

[0066] Multiple graph neural networks may be individually provided with augmented numeric graph representations per material. Graph neural networks may include convolutional graph neural networks or isomorphic graph neural networks. As an example, a graph convolutional network may be defined by the following:

[0067]

[0068] As another example, a graph isomorphic network can be defined by the following:

[0069]

[0070] Therefore, each graph representing a material can have a D-dimensional representation in D-dimensional space. For each layer of the graph neural network, node and / or bond states can be updated using neighbor states, such as the nearest neighbor atom, the next nearest neighbor atom, and so on. The update can also be referred to as message passing. An update function for each layer can be provided, for example, by using the notation described above:

[0071] ,

[0072] Here, the MLP is an example of a regression model, such as a multilayer perceptron (MLP). An MLP may include input and output layers and multiple hidden layers between them. An activation function may be used in each of the computed layers. Thus, the input can be forward-propagated through the MLP by taking the inner product of the input and the weight existing between the input layer and the first hidden layer. This inner product can yield a value in the hidden layer. Subsequently, the output computed in the current hidden layer can be transformed through one or more activation function(s), such as the Rectifying Linear Unit (ReLU), the sigmoid function, or tanh. Once the output computed in the hidden layer is propagated through the activation function, it can be propagated to the next layer within the MLP by taking the inner product with the corresponding weight. These steps can be repeated until the output layer is reached. At the output layer, the computation will be used for a backpropagation algorithm corresponding to the activation function selected for the MLP (in the case of training).

[0073] A graph neural network can provide multiple representations per augmented material. The results per augmented material can be pooled by one or more pooling functions, such as sum, mean, maximum, set2set, etc., to generate material-specific graph representations per augmented material.

[0074] Pooled representations per augmented material can be provided to the loss function for contrastive learning. The loss function may relate to a similarity measure that measures the distance between the calculated representations per augmented material for each material. An exemplary loss function may relate to a cosine similarity measure using, for example, the known Tanimoto similarity. For two graphs i, j from the same group or per material, the loss function may be defined by the following:

[0075] ,

[0076] Here, z i is the R of the i-th molecule N It is an expression.

[0077] In this way, at least one data-driven graph representation model can be trained using a contrast loss function that may depend on a distance measure between material-specific graph representation(s) of different materials and / or augmented materials. By training a graph neural network based on material representations and their augmentations, material-specific graph representations can be learned by learning the relationships and correlations of different components of material representations. In this manner, similar materials can be mapped into a D-dimensional embedding space, and contrast learning enables the learning of a feature space that can combine or group related points together and push out unrelated points. The contrast loss function essentially seeks to minimize the distance between similar material-specific graph representations and maximize the distance between unrelated material-specific graph representations. This can also be referred to as an unsupervised learning approach for generating material-specific graph representations. Further details of this approach can be found in, for example, the literature [Wang, Y., Wang, J., Cao, Z. et al. Molecular contrastive learning of representations via graph neural networks]. Nat Mach Intell 4, 279-287 (2022). https: / / doi.org / 10.1038 / s42256-022-00447-x] or [Improving Molecular Contrastive Learning via Faulty Negative Mitigation and Decomposed Fragment Contrast, Yuyang Wang, Rishikesh Magar, Chen Liang, and Amir Barati Farimani, Journal of Chemical Information and Modeling 2022 62 (11), 2713-2725, DOI: 10.1021 / acs.jcim.2c00495].

[0078] Figure 5 illustrates an example of a training process for a data-driven feature model.

[0079] For example, a material-specific graph representation, such as one generated by a trained GNN as described in relation to FIG. 4, may be stored in relation to its material specifications. For a data-driven feature model, a training dataset may be provided that includes at least one biodegradable property as measured for a material associated with a material specification or a corresponding material-specific graph representation. The data-driven feature model may be trained on a training dataset that includes at least one biodegradable property measured for one or more material(s) and at least one material-specific graph representation(s) generated by a data-driven graph representation model. The data-driven model architecture may include any suitable architecture for supervised learning that generates at least one biodegradable property based on the material-specific graph representation(s). An example of a suitable regression model may be a random forest regression model that maps feature vectors of the material-specific graph representation(s) to at least one biodegradable property according to a tree structure. Other possible architectures can range from simple classification models that identify biodegradable and non-biodegradable through gradient boosting methods to more sophisticated models, such as transformer-based models, that generate at least one biodegradable property.

[0080] Figure 5 illustrates the training process. On the input layer of the model, material-specific graph representation(s) potentially fused with additional feature vectors, such as habitat type and / or condition, measurement type and / or condition, etc., as mentioned in relation to Figure 3, may be provided. Model weights and / or structures may be trained to generate at least one biodegradable property. A loss function may be defined to minimize the difference between at least one biodegradable property measured from the training dataset and at least one generated biodegradable property. Multiple models may be trained depending on the property type, material type, and / or measurement type. Here, transfer learning, ensemble learning, or other conventional techniques may be used for training.

[0081] When inferring or using, a data-driven characteristic prediction model trained in this way can map a given material-specific graph representation and potential additional representations to at least one biodegradable characteristic.

[0082] Figures 6 and 7 show the output of the trained model.

[0083] To illustrate the results of generating biodegradable properties, FIGS. 6 and 7 illustrate a user interface displaying such results. In FIG. 6, a molecular structure for a small molecule is used as an exemplary basis. A molecular structure provided upon request and on which biodegradable properties are generated is displayed along with a molecular structure similar to the requested molecular structure. Following the structure and biodegradable properties, in this case, a similarity measure calculated based on, for example, mass loss after 10 days, confidence intervals in terms of the standard deviation of the model output, and a numerical distance measure between each material-specific graph representation(s).

[0084] In Fig. 7, representations of molecular structures are used to generate 2-d similarity or diffusion maps. Each point on the map represents a single substance, e.g., a small molecule. The distance between points represents a distance scale between each substance-specific graph representation(s).

[0085] Figures 8 and 9 illustrate the comparison results from differently trained models.

[0086] In the table in Figure 8, a GNN model including a feature prediction GNN1 is compared with a GNN model using a two-stage approach that combines the GNN model with a separate feature prediction model, GNN2. Additionally, traditional fingerprint models, such as the Morgan fingerprint, which is not based on the GNN-generated fingerprint MF, were compared. More details regarding the Morgan fingerprint can be found in the literature [The Generation of a Unique Machine Description for Chemical Structures-A Technique Developed at Chemical Abstracts Service. HL Morgan Journal of Chemical Documentation 1965 5 (2), 107-113 DOI: 10.1021 / c160017a018].

[0087] Figure 9 illustrates various pooling and head options and compares standard error or confidence measures.

[0088] The present disclosure is also described together with preferred embodiments and examples. However, from the drawings, the present disclosure and the claims, other variations may be understood and practiced by those skilled in the art practicing the claimed invention.

[0089] Any of the steps presented herein may be performed in any order. The method disclosed herein is not limited to a specific order of these steps. It is also not required that different steps be performed at a specific location or at a specific computing node of a distributed system; that is, each of the steps may be performed at different computing nodes using different equipment / data processing.

[0090] As used herein, “determining” also includes “initiating or inducing a determination,” “generating” also includes “initiating and / or inducing a generation,” and “providing” also includes “initiating or inducing a determination, generation, selection, transmission and / or reception.” “Initiating or causing an action” includes any processing signal that triggers a computing node or device to perform each respective action.

[0091] In the claims as well as in the description, the words "comprising" or "including" or similar words do not exclude other elements or steps and should not be interpreted as a limitation on the listed elements or steps. Singular expressions (“a” or “an”) do not exclude the plural. A single element or other unit may perform the function of several entities or items mentioned in the claims. The mere fact that specific means are described in mutually distinct dependent claims does not indicate that a combination of these means cannot be used in a favorable implementation or that additional elements may be included.

[0092] Within the scope of the present disclosure, the provided may include any interface configured to provide data. This may include application programming interfaces, human-machine interfaces, such as display and / or software module interfaces. The provided may include the communication of data, or the submission of data to an interface, in particular the use of data by a display or receiving entity to a user.

[0093] Any disclosures and embodiments described herein relate to the methods, systems, apparatuses, devices, chemicals, materials, services, uses, and computer program elements listed above, and vice versa. Advantageously, the benefits provided by any embodiments and examples apply equally to all other embodiments and examples, and vice versa.

[0094] All terms and definitions used herein are broadly understood and have their general meanings.

Claims

Claim 1 A method for generating at least one biodegradable property that characterizes at least one substance, comprising: providing at least one substance-specific graph representation generated by providing the numeric graph representation(s) associated with at least one atom and bond of the at least one substance to at least one substance-specific graph representation model trained to map the numeric graph representation(s) associated with at least one atom, bond, and correlations associated with the atom and / or bond; generating at least one biodegradable property by providing the at least one substance-specific graph representation to at least one data-based property model trained to map the at least one substance-specific graph representation to at least one biodegradable property; and providing the generated at least one biodegradable property to characterize the biodegradability of the at least one substance. Claim 2 A method according to claim 1, wherein at least one material specification associated with at least one material is provided, wherein the at least one specification is mapped to a corresponding numerical graph representation(s) associated with at least one atom and bond of the at least one material. Claim 3 A method according to claim 1 or 2, wherein the at least one data-based graph representation model is trained by generating numeric graph representation(s) associated with one or more augmented material(s) per material graph representation, wherein the augmented material comprises at least one synthetic change in the numeric graph representation associated with an atom, bond, atomic-bond combination, or material component. Claim 4 A method according to any one of claims 1 to 3, wherein the at least one data-based graph representation model is trained using a contrast loss function that depends on the distance between numeric graph representation(s) of different materials and / or augmented materials. Claim 5 A method according to any one of claims 1 to 4, wherein the data-based characteristic model is trained on a training dataset comprising at least one biodegradable characteristic measured for one or more material(s) and material-specific graph representation(s) generated by at least one data-based graph representation model. Claim 6 A method according to any one of claims 1 to 5, wherein at least one type of substance related to a small substance, a large molecule, e.g., a polymer or oligomer and / or a formulation is provided to the at least one data-based graph representation model and / or the at least one data-based characteristic model, or is provided to select the at least one data-based graph representation model and / or the at least one data-based characteristic model. Claim 7 A method according to any one of claims 1 to 6, wherein at least one measurement type related to the measurement of the biodegradable property is provided to the at least one data-based graph representation model and / or the at least one data-based property model and / or is provided to select the at least one data-based graph representation model and / or the at least one data-based property model. Claim 8 A method according to any one of claims 1 to 7, wherein one or more similar material-specific graph representation(s) are generated based on a representation distance scale based on a material-specific graph representation associated with at least one material having the generated biodegradable properties, and wherein one or more similar material-specific graph representation(s) and corresponding material(s) are provided. Claim 9 A method according to any one of claims 1 to 8, wherein a material map comprising a plurality of material-specific graph representation(s) associated with a plurality of materials and optionally corresponding generated biodegradable properties is generated based on a representation distance scale, wherein a material map comprising a map of materials and optionally corresponding generated biodegradable properties is provided in relation to said representation distance scale. Claim 10 A method according to any one of claims 1 to 9, wherein one or more habitat conditions affecting one or more biodegradable properties are provided, wherein the at least one data-based property model is trained on one or more related biodegradable properties dependent on material-specific graph representations and one or more habitat conditions, wherein the one or more habitat conditions are mapped to numerical habitat conditions, wherein the one or more numerical habitat conditions are fused with the material-specific graph representations, and wherein the fused material-specific graph representations are provided to the at least one data-based property model. Claim 11 A method according to any one of claims 1 to 10, wherein one or more habitat conditions affecting one or more biodegradable properties are provided, wherein the at least one data-based graph representation model is trained on numeric graph representations and one or more associated habitat conditions, wherein the one or more habitat conditions are mapped to numeric habitat conditions, wherein the one or more numeric habitat conditions are fused with the numeric graph representations, and wherein the fused numeric graph representations are provided to the at least one data-based graph representation model. Claim 12 A method according to any one of claims 1 to 11, wherein the biodegradable property relates to at least one material structure associated with a non-biodegradable or biodegradable material comprising at least one functional chemical compound, at least one macromolecule, at least one formulation, or a combination thereof, wherein the biodegradable property relates to a biodegradable or non-biodegradable class, wherein the biodegradable property relates to a biodegradable measurement related to a reference measurement and / or a reference habitat, and wherein the biodegradable property relates to a biodegradable measurement based on the structure and / or composition of the material. Claim 13 A method according to any one of claims 1 to 12, wherein the biodegradable property is related to at least one application of the material and / or at least one application performance measurement of the material based on the chemical structure and / or composition of the material. Claim 14 Use of the selected biodegradable properties and associated material structure produced according to any one of claims 1 to 13 for producing a biodegradable material having a selected material structure.

Images