A predictive machine learning model for providing a sun protection factor (SPF) value associated with an ultraviolet (UV) filter composition

Abstract

A computer-implemented method (100) for providing at least one predicted sun protection factor (SPF) value associated with an ultraviolet (UV) filter composition, the method (100) comprising: - a providing step (135) providing a set of original exemplary UV filter composition digital identifiers, the set of exemplary UV filter compositions comprising at least one UV filter composition digital identifier comprising: - at least one existing ingredient or at least one combination of existing ingredients, - at least one calculated UV spectrum, and - a value representative of an empirically measured SPF value, thereby forming a training set, - a training step (140) training a predictive machine learning model using the training set to predict at least one SPF value, - a defining step (145) defining at least one UV filter composition digital identifier, - a predicting step (150) predicting at least one SPF value using the trained predictive machine learning model, - a providing step (155) providing the predicted SPF value.

Description

Technical Field

[0001] This invention relates to a computer-implemented method for providing at least one predicted sun protection factor (SPF) value associated with a digital identifier of an ultraviolet (UV) filter composition, the UV filter composition digital identifier representing a materializable UV filter composition, the materializable UV filter composition being associated with at least one component digital identifier representing an existing component of the materializable UV filter composition; a computer-implemented method for training a predictive machine learning model to provide at least one SPF value associated with a digital identifier of an ultraviolet (UV) filter composition, the ultraviolet (UV) filter composition digital identifier representing a materializable UV filter composition; and corresponding computer program products, computer-readable storage media, and devices.

[0002] This invention is applicable to the field of UV filter products, such as cosmetics, skin care products, or sun protection products. Background Technology

[0003] Developing sunscreen products is costly and time-consuming because the body's sun protection factor (SPF) must be evaluated.

[0004] Today, sunscreen developers can use a variety of computational tools to predict the sun protection performance of ultraviolet (UV) filter compositions to some extent. These tools allow developers to quickly pre-evaluate and compare the performance of different UV filter compositions on a computer before conducting in vivo SPF studies, thereby screening out the most promising candidate ingredients. This saves development costs and time and helps in finding the optimal solution.

[0005] However, if these tools can determine SPF instead of performing a pre-assessment, they can save even more cost and time.

[0006] Therefore, a sufficiently accurate SPF prediction method allows sunscreen developers to focus their research on other important characteristics, such as sensory properties, environmental friendliness, or cost.

[0007] Therefore, using this method will ultimately lead to better sun protection products for consumers, namely any product labeled with an SPF value.

[0008] Unfortunately, because current prediction methods only consider UV filter compositions, their accuracy is still insufficient to replace in vivo studies.

[0009] However, it is well known that the distribution of UV filters on the skin and the sun protection factor (SPF) are strongly affected by other ingredients besides UV filters, such as emollients, thickeners, emulsifiers, and film-forming agents.

[0010] For example, it is obvious that high-viscosity sunscreens may distribute less evenly on the skin compared to low-viscosity products, so even if the UV filter composition of the two products is the same, the SPF value of the high-viscosity sunscreen will be lower.

[0011] Jiyong S. et al. Machine learning for the prediction of sunscreen sunprotection factor and protection grade of UVA, Experimental Dermatology, 2019, Volume 28, Issue 7, Pages: 872-874 and WO2022157011A1 both attempt to overcome this problem by using machine learning methods. These methods consider not only the concentration of UV filter substances but also various other factors that may affect the SPF or UVA protection factor.

[0012] The method described by Jiyong S. et al. uses four additional characteristics in addition to the concentration of UV filter substances, such as the presence of pigments, the amount of pigment-grade titanium dioxide, the formulation type, and the product type.

[0013] The method described in WO2022157011A1 distinguishes between UVB and UVA filters and adds seven additional coefficients to obtain the following feature set:

[0014] - The viscosity of the composition,

[0015] - High-polarity emollients, medium-polarity emollients, and low-polarity emollients,

[0016] -UVA filter, UVB filter, and the ratio of UVB filter to UVA filter,

[0017] - The ratio of UV filter in the oil phase to UV filter in the aqueous phase.

[0018] - The ratio of absorptive UV filters to scattering / reflecting UV filters.

[0019] Both methods use only a subset of features manually selected from the set of all possible UV filter composition features. Manually selected features almost always have lower predictive accuracy than automatically selected features.

[0020] Both methods neglect the influence of many other characteristics, such as the spreadability or chemical properties of the emulsifier, the type and amount of the emulsifier, and the presence, type, and amount of the film-forming agent. All of these factors can significantly affect the distribution of UV filters on the skin and SPF.

[0021] Therefore, neither of these methods is accurate enough to replace in vivo SPF studies of materialized compositions. Summary of the Invention

[0022] The present invention aims to overcome all or part of these shortcomings.

[0023] Therefore, according to the first aspect, the present invention aims to provide a computer-implemented method for providing at least one predicted sun protection factor (SPF) value associated with a digital identifier of an ultraviolet (UV) filter composition, the digital identifier representing a materializable UV filter composition, the materializable UV filter composition being associated with at least one component digital identifier representing an existing ingredient in the materializable UV filter composition, the computer-implemented method comprising:

[0024] - Providing steps that provide an original set of exemplary UV filter composition digital identifiers, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, including:

[0025] - At least one value representing an existing component, or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier.

[0026] - At least one calculated UV spectrum value associated with a digital identifier of the UV filter composition, and

[0027] - At least one value representing an empirically measured SPF value associated with a digital identifier of the UV filter composition.

[0028] This forms the training set.

[0029] - A training step that uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value.

[0030] - Define a step that defines at least one digital identifier for a UV filter composition.

[0031] - A prediction step that uses a trained predictive machine learning model to predict at least one SPF value;

[0032] - Provides steps that provide at least one predicted SPF value.

[0033] Thanks to these regulations, the predicted SPF values ​​are more accurate. The UV filter composition of sunscreen products and the spectra of each individual UV filter are the most important factors affecting SPF. By explicitly including the spectral information of the UV filter composition in the training set, prediction accuracy can be significantly improved.

[0034] Furthermore, encoding UV filter absorption capacity as UV spectra and ingredient properties as physicochemical properties allows for the definition of information-rich, dense input features that may not include an ingredient list. This improves the model's generalization performance for novel sunscreen compositions, even those that may include ingredients not present in the training set.

[0035] Furthermore, the method of the present invention has great flexibility and can achieve an accuracy sufficient to replace the in vivo SPF method as the amount of data increases.

[0036] In one particular implementation, in the training set, at least one value representing an existing component is the number of components.

[0037] Thanks to these regulations, the UV filter compositions in the training set have been defined more precisely.

[0038] In one particular implementation, in the training set, at least one value representing an existing component is a value representing the function of the existing component.

[0039] In one particular implementation, in the training set, at least one value representing an existing component or combination of existing components is a value representing the physicochemical properties of the existing component or combination of existing components.

[0040] In one particular embodiment, the training set includes at least one value representing the proportion of erythema energy absorbed in the oil phase, which is associated with at least one composition digital identifier; and / or at least one value representing the total amount of UV filter in the composition, which is associated with at least one composition digital identifier.

[0041] Thanks to these regulations, the accuracy of predictive machine learning models has become significantly more pronounced.

[0042] In one particular embodiment, the computer-implemented method of the present invention further includes:

[0043] - An assembly step that assembles at least one component into a UV filter composition.

[0044] - A calculation step, which calculates at least one UV spectrum of the assembled UV filter composition, or at least one UV spectrum of a combination of one or more components.

[0045] - An empirical measurement step, which empirically measures the SPF value of the assembled UV filter composition, and

[0046] - A storage step, which stores in a database at least one empirically measured SPF value and at least one calculated UV spectrum, the UV spectrum being associated with at least one digital identifier for the assembled UV filter composition.

[0047] The database is used during the step of providing the training set.

[0048] Thanks to these regulations, the database that makes up the training set is based on real-life samples and measurement data from the real world.

[0049] In one particular implementation, the method further includes a preprocessing step that preprocesses the training set, wherein the preprocessed training set is used in the provisioning step of providing the training set.

[0050] Thanks to these provisions, the number of values ​​in the training set can be reduced, thereby mitigating the risk of overfitting. All components can be replaced by their physicochemical properties and / or proportions within functional groups, thus generating the SPF value predictions for any given dataset using the method of this invention. Furthermore, the sparse component quantity matrix can be replaced by a dense component property matrix, allowing for more efficient training of machine learning tools and achieving better generalization performance.

[0051] In one particular implementation, the trained predictive machine learning model is a random forest regressor (RF).

[0052] Thanks to these rules, the predictions from the decision tree groups can be combined, thereby improving performance.

[0053] In one particular implementation, the trained predictive machine learning model is a gradient boosting regressor (GB, XGB).

[0054] Thanks to these regulations, large datasets can be used efficiently and with high accuracy. The ensemble method exhibits several useful characteristics that facilitate the implementation of the methods described in this invention, such as:

[0055] - It can achieve higher performance compared to a single machine learning model, especially for high-dimensional and noisy problems.

[0056] - It is less prone to overfitting, especially when the number of relevant variables is small compared to the total number of variables.

[0057] - Provides natural indicators to determine the importance of features for prediction.

[0058] In one particular implementation, the trained predictive machine learning model is a neural network (NN), such as a feedforward neural network.

[0059] Thanks to these regulations, predictive machine learning is highly scalable.

[0060] In one particular embodiment, the computer-implemented method of the present invention further includes a sending step of sending a digital command representing an instruction for materializing at least one UV filter composition, the at least one UV filter composition corresponding to a provided digital identifier for at least one UV filter composition.

[0061] Thanks to these provisions, the computer-implemented method of the present invention can be connected to a device that executes digital commands, thereby materializing the UV filter composition.

[0062] In one particular embodiment, the computer-implemented method of the present invention further includes a materialization step, which materializes at least one UV filter composition corresponding to a provided digital identifier of at least one UV filter composition.

[0063] Thanks to these regulations, UV filter compositions were materialized.

[0064] According to a second aspect, the present invention aims to provide a computer-implemented method for training a predictive machine learning model to provide at least one value associated with a digital identifier of an ultraviolet (UV) filter composition, the UV filter composition digital identifier representing a materializable UV filter composition, the method comprising:

[0065] - Providing steps that provide an original set of exemplary UV filter composition digital identifiers, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, including:

[0066] - At least one value representing an existing component, or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier.

[0067] - At least one calculated UV spectrum value associated with a digital identifier of the UV filter composition, and

[0068] - At least one value representing an empirically measured SPF value associated with a digital identifier of the UV filter composition.

[0069] This forms the training set, and

[0070] - Training step, which uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value.

[0071] According to the third aspect, the present invention aims to provide a computer program product comprising instructions that, when executed by a computer, cause the computer to perform the method of the present invention.

[0072] According to the fourth aspect, the present invention aims to provide a computer-readable storage medium storing programming instructions that, when executed by a computer, cause the computer to perform the method of the present invention.

[0073] According to a fifth aspect, the present invention aims to provide an apparatus for providing at least one predicted sun protection factor (SPF) value associated with a digital identifier of an ultraviolet (UV) filter composition, the digital identifier representing a materializable UV filter composition, the materializable UV filter composition being associated with at least one ingredient digital identifier representing an existing component of the materializable UV filter composition, the apparatus comprising:

[0074] - A provider that provides an original set of exemplary UV filter composition digital identifiers, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, comprising:

[0075] - At least one value representing an existing component, or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier.

[0076] - At least one calculated UV spectrum value associated with a digital identifier of the UV filter composition, and

[0077] - At least one value representing an empirically measured SPF value associated with a digital identifier of the UV filter composition.

[0078] This forms the training set, and

[0079] - A training facility that uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value.

[0080] - Define an organization that defines at least one digital identifier for a UV filter composition.

[0081] - A prediction agency that uses a trained predictive machine learning model to predict at least one SPF value;

[0082] - Provide an agency that provides at least one predicted SPF value.

[0083] The second to fifth embodiments of the present invention have the same advantages as the related first embodiment. Attached Figure Description

[0084] Other advantages, objects, and specific features of the invention will become apparent from the following non-exhaustive description of at least one particular embodiment of the invention with reference to the accompanying drawings:

[0085] Figure 1 The illustration schematically depicts the first sequential steps of a particular embodiment of the method of the present invention.

[0086] Figure 2 The computer system shown is schematically illustrated as an embodiment that can be used to carry out the method of the present invention.

[0087] Figure 3 The diagram illustrates the correlation between predicted and calculated SPF values. A first dataset, labeled "S1," is used to train and evaluate the machine learning model XGB implemented in the method of this invention.

[0088] Figure 4 The diagram illustrates the correlation between predicted and calculated SPF values. A second dataset, labeled "S2," is used to train and evaluate the machine learning model implemented in the method of this invention.

[0089] Figure 5 The illustration schematically depicts the first sequential steps of a specific implementation of the training set preprocessing method of the present invention.

[0090] Figure 6 This represents a list of features categorized according to their importance in accurately predicting the SPF value of component compositions. Detailed Implementation

[0091] This description is not exhaustive, as each feature of one embodiment can be advantageously combined with any other feature of any other embodiment. Furthermore, various inventive concepts can be embodied as one or more methods, an example of which has been provided herein. Operations performed as part of a method can be ordered in any suitable manner. Thus, embodiments can be constructed in which the order of operation execution differs from that illustrated, including performing certain operations simultaneously, even if they are shown as sequential operations in the exemplary embodiment.

[0092] Unless otherwise expressly stated, the indefinite articles “a” and “an” used in this specification and claims shall be understood as “at least one (type)”.

[0093] The phrase “and / or” as used in this specification and claims should be understood as “one of both” or “both,” meaning that they exist side-by-side in some cases and separately in others. Multiple elements listed using “and / or” should be interpreted in the same way, i.e., “one or more” side-by-side elements. Other elements may be selectively included in addition to those expressly specified in the “and / or” clause, regardless of whether these elements are associated with the expressly specified elements. Thus, for example, when “A and / or B” is used in conjunction with open-ended language such as “comprising,” in one embodiment, it may refer only to A (optionally including elements other than B); in another embodiment, it may refer only to B (optionally including elements other than A); in yet another embodiment, it may refer to both A and B (optionally including other elements); and so on.

[0094] In this specification and claims, "or" should be understood to have the same meaning as "and / or" as defined above. For example, when separating items in a list, "or" or "and / or" should be interpreted as inclusive, meaning it includes at least one element, as well as multiple elements, and may selectively include other elements not listed. Only terms that explicitly indicate the opposite meaning, such as "only one" or "exactly one," or "consisting of" as used in the claims, refer to an element among multiple elements. Generally, the term "or" as used in this specification should only be interpreted as indicating a mutually exclusive choice (i.e., "either one or the other") when preceded by an exclusive term (e.g., "any," "one," "only one," or "exactly one"). The phrase "mainly composed of" when used in the claims should have the usual meaning used in the field of patent law.

[0095] As used in this specification and claims, the term "at least one" when referring to a list of one or more elements should be understood as selecting at least one element from one or more elements in the list, but does not necessarily include every element specifically listed in the list, nor exclude any combination of elements in the list. This definition also allows for the presence of elements other than those specifically identified in the list referred to by "at least one," regardless of whether such elements are associated with any element specifically identified in the list. Thus, as a non-limiting example, "at least one of A and B" (or equivalently, "at least one of A or B," or equivalently, "at least one of A and / or B") in one embodiment may refer to at least one A, which may optionally include multiple A's and exclude B (and may optionally include other elements besides B). In another embodiment, at least one (optionally multiple) B is included, where A is absent (and may optionally include elements other than A); in yet another embodiment, at least one (optionally multiple) A and at least one (optionally multiple) B are included (and may optionally include other elements); and so on.

[0096] In the aforementioned claims and specification, all transitional phrases such as "comprising," "including," "carrying," "having," "containing," "involving," "holding," and "constituting from" should be understood as open-ended phrases, meaning "including but not limited to." Only the transitional phrases "composed of" and "mainly composed of" should be understood as closed or semi-closed transitional phrases, respectively.

[0097] It should be noted that these figures are not drawn to scale.

[0098] In this invention, a "UV filter composition" (also commonly referred to as "sunscreen" or "SPF product") refers to ultraviolet light that includes at least one component that absorbs ultraviolet light, specifically light with a wavelength range of 280 to 400 nm. Such components are generally referred to as "UV filters" and are typically categorized as UV-B filters (i.e., components with a wavelength maximum (λmax) in the range of 280–320 nm), UV-A filters (i.e., components with a wavelength maximum (λmax) in the range of 320–400 nm), or broadband filters (components that absorb ultraviolet light in both the UV-B and UV-A ranges).

[0099] The properties listed thereafter, in the form of "E" followed by a number, refer to the UV spectral components associated with the wavelength corresponding to that number. These properties indicate that a certain component can block light of that wavelength.

[0100] Subsequently, the features marked with numbers refer only to specific components that can be found by those skilled in the art through the feature importance determination algorithm (described below).

[0101] UV filters used in skincare products (or sunscreens) help prevent sunburn, premature skin aging, and / or skin cancer by absorbing, scattering, or reflecting ultraviolet radiation.

[0102] UV filters can be physical (inorganic), chemical (organic), or a combination of both; they can block or reduce the penetration of ultraviolet rays into the skin through different mechanisms.

[0103] In addition to one or more UV filters, UV filter compositions typically include a cosmetically acceptable carrier, which is usually composed of common cosmetic auxiliaries and additives such as preservatives / antioxidants, fatty substances / oils, water, organic solvents, silicones, thickeners, softeners, emulsifiers, defoamers, humectants, aesthetic ingredients such as fragrance ingredients (aromatics), surfactants, fillers, chelating agents, anionic, cationic, nonionic or amphoteric polymers or mixtures thereof, propellants, acidifiers or alkalizers, dyes, colorants / colorants, abrasives, absorbents, essential oils, skin-feeling agents, astringents, defoamers, pigments or nanopigments, or any other ingredients typically formulated into UV filter compositions.

[0104] The ingredients (including UV filters) applicable to the UV filter compositions of this invention can be found, for example, in the International Dictionary and Handbook of Cosmetic Ingredients published by the Personal Care Products Association (http: / / www.personalcarecouncil.org / ) (accessible via an online database (http: / / online.personalcarecouncil.org / jsp / Home.jsp)), but are not limited thereto. Therefore, in this invention, "ingredient" means any component of the UV filter composition, including but not limited to UV filters.

[0105] Particularly preferred UV filters in UV filter compositions include:

[0106] -Physical (inorganic) UV filters, such as:

[0107] Zinc oxide: A mineral UV filter that provides broad-spectrum UVA and UVB protection. It adheres to the skin surface and reflects ultraviolet radiation; this zinc oxide is often surface-coated, for example, by coating with triethoxyoctylsilane. A particularly suitable surface-coated zinc oxide product is PARSOL® ZX manufactured by dsm-firmenich.

[0108] Titanium dioxide: Another mineral UV filter, its working principle is similar to zinc oxide, working by reflecting ultraviolet radiation, effective against both UVA and UVB rays. This type of titanium dioxide is typically coated with one or more of silica, polydimethylsiloxane, and triethoxyoctylsilane. For example, PARSOL®TX manufactured by dsm-firmenich is a particularly suitable surface-coated titanium dioxide.

[0109] - Chemical (organic) UV filters, such as:

[0110] - Avobenzone: A commonly used organic UV filter that absorbs UVA rays, for example, sold by dsm-firmenich under the trade name PARSOL® 1789.

[0111] - Diethylaminohydroxybenzoylhexyl benzoate: A commonly used organic UV filter that absorbs UVA rays, for example, sold by dsm-firmenich under the trade name PARSOL® DHHB.

[0112] Cinnamyl esters, such as octyl methoxycinnamate (trade name PARSOL® MCX, manufactured by dsm-firmenich) and isoamyl p-methoxycinnamate, can absorb UVB rays.

[0113] - Octocrylene: Absorbs UVB rays and stabilizes other UV filters, such as avobenzone, for example, sold by dsm-firmenich under the trade name PARSOL® 340.

[0114] - Benzophenone: can absorb UVA and UVB rays.

[0115] Triazine derivatives, such as bis(ethylhexyloxyphenol)methoxyphenyltriazine, can absorb both UVA and UVB rays (broad-spectrum filters), and are marketed by dsm-firmenich under the trade name PARSOL® SHIELD. Triazine compounds that absorb UVB rays, such as those marketed by dsm-firmenich under the trade name PARSOL® EHS, as well as terphenyltriazine and diethylhexylbutyramide triazine (DBT), can also absorb UVB rays.

[0116] - Benzimidazole derivatives, such as phenylbenzimidazole sulfonic acid, can absorb UVB rays, such as those sold by dsm-firmenich under the trade name PARSOL® HS, and disodium phenyldibenzimidazole tetrasulfonate.

[0117] - Salicylates, such as homosalate and octyl salicylate, can absorb UVB rays. For example, they are available at dsm-firmenich in the form of PARSOL® HMS and PARSOL® EHS, respectively.

[0118] -Methylenebisbenzotriazolyltetramethylbutylphenol is an organic pigment that absorbs UVA and UVB rays (broad-spectrum filter), such as the commercially available product PARSOL® MAX.

[0119] - Camphor derivatives, such as 4-methylbenzyl camphor (e.g., sold by dsm-firmenich under the trade name PARSOL® 5000), 3-benzyl camphor, and terephthalic acid benzyl camphor. Dicamphor sulfonic acid.

[0120] - Cresoltrazol trisiloxane is a broad-spectrum ultraviolet absorber with two absorption peaks, one at 303 nm (UVB) and the other at 344 nm (UVA).

[0121] -Polysiloxane-15 is a UV-B filter, marketed by dsm-firmenich under the trade name PARSOL® SLX.

[0122] -Methoxypropylaminocyclohexeneylethoxyethylcyanoacetate can absorb UV-A light, especially in the 360 ​​to 400 nm range;

[0123] - Menthol anthranilate.

[0124] Typical auxiliaries and additives commonly used in UV filter compositions include:

[0125] - Antioxidants: These are typically added to UV filter compositions to help neutralize free radicals generated by ultraviolet radiation and provide additional sun protection, such as vitamin E or vitamin C and their derivatives.

[0126] - Stabilizers: Added to the UV filter composition to enhance the stability and long-term effectiveness of the UV filter, for example by reducing and / or inhibiting chemical degradation, such as diethylhexyl 2,6-naphthalenedicarboxylate or diethylhexyl syringite malonate.

[0127] - Moisturizers (humectants): Compounds or mixtures of compounds that impart the property of reducing transepidermal water loss (TEWL) after UV filter compositions are applied or distributed to the skin surface (e.g., as determined experimentally). These include compounds that reduce transepidermal water loss (TEWL) and / or positively influence the moisture content of the stratum corneum and / or maintain actual hydration, such as polyols (e.g., glycerin or ethylene glycol), lactic acid and / or lactates, particularly sodium lactate, biosaccharide gum-1, soy glycine, ethylhexyloxyglycerol, pyrrolidone carboxylic acid, urea, hyaluronic acid or derivatives thereof, chitosan and / or fucose-rich polysaccharides, and sugars and sugar derivatives, such as glycoisomers.

[0128] - Preservatives / Preservative Enhancers: Substances suitable for improving the shelf life of UV filter compositions, such as ethanol, phenoxyethanol, 1,2-pentanediol, 1,2-hexanediol, 1,2-octanediol (octanediol), 1,2-decanediol, 2-methyl-1,3-propanediol, 1,3-propanediol, 1,2-propanediol and p-hydroxyacetophenone.

[0129] -Fragrance ingredients: These refer to compounds or mixtures of compounds used in a composition to impart a pleasant effect. In other words, for such ingredients to be considered fragrance ingredients, they must be deemed by those skilled in the art to impart or alter the odor of the composition in an active or pleasant manner, rather than merely having an odor. Examples include: limonene, citral, linalool, α-isomethylionone, geraniol, citronellol, 2-isobutyl-4-hydroxy-4-methyltetrahydropyran, 2-tert-pentylcyclohexyl acetate, 3-methyl-5-phenyl-1-pentanol, 7-acetyl-1,1,3,4,4,6-hexamethyltetrahydronaphthalene, adipate, cinnamaldehyde, amyl salicylate, α-pentylcinnamaldehyde, α-methylionone, butylphenylmethylpropionaldehyde, cinnamaldehyde, pentylcinnamyl alcohol, anisole, benzoin, benzyl alcohol, benzyl benzoate, benzyl cinnamate, benzyl salicylate, bergamot oil, bitter orange oil, butylphenylmethylpropionaldehyde, cardamom oil, cedrol, cinnamaldehyde, cinnamyl alcohol, methyl crotonate citronellol ester, citrus oil, coumarin, diethyl succinate, ethyl linalool, eugenol, and tree moss (Evernia) extracts of *Furfuracea*, *Evernia prunastri*, farnesol, guaiac oil, hexylcinnamaldehyde, hexyl salicylate, hydroxycitronellol, lavender oil, lemon oil, linalyl acetate, mandarin oil, menthol PCA, methyl hexadecanoate, nutmeg oil, rosemary oil, sweet orange oil, terpineol, tonka bean oil, triethyl citrate, and vanillin.

[0130] - Emollients: Oils, especially those used to dissolve solid UV filters in UV filter compositions, such as dibutyl adipate, diisopropyl sebacate, dibutyl sebacate, isopropyl palmitate, isopropyl myristate, octyl decanoate triglyceride, cetearyl isononanoate, coconut oil glyceride, isonononanoate, propylene glycol dicaprylate / dicaprylate, phenethyl benzoate, C12-C15 alkyl benzoate, C16-17 alkyl benzoate, dioctyl carbonate, C12-13 alkyl tartarate, hydrogenated castor oil diester. Triheptanoic acid glycerides, C12-13 alkyl lactates (such as lauryl lactate), caprylic / capric triglycerides, dioctyl 2,6-naphthalenedicarboxylate, octyl dodecanol and ethylhexyl cocoate, dibutyl adipate, and oils with soothing and softening effects on the skin, such as silicones (dimethicone, cyclodimethicone), vegetable oils (grapeseed oil, sesame oil, jojoba oil, etc.), milk fats (cocoa butter, shea butter), and petrolatum derivatives (petroleum jelly, mineral oil).

[0131] - Emulsifiers / surfactants: Compounds suitable for preparing emulsions, especially oil-in-water (O / W) or water-in-oil (W / O) emulsions, such as phosphate ester surfactants (e.g., C8-10 alkyl ethyl phosphate, C9-15 alkyl phosphate, cetearyl alcohol polyether-2 phosphate, cetearyl alcohol polyether-5 phosphate, cetearyl alcohol polyether-8 phosphate, cetearyl alcohol polyether-10 phosphate, cetyl phosphate, C6-10 alkanol polyether-4 phosphate, C12-15 alkanol polyether-2 phosphate, C12-15 alkanol polyether-3 phosphate, DEA-cetearyl alcohol polyether-2 phosphate, DEA-cetyl phosphate). DEA-oleyl alcohol polyether-3 phosphate, cetyl phosphate, decanol polyether-4 phosphate, decanol polyether-6 phosphate and trilauryl alcohol polyether-4 phosphate), PEG-30 dihydroxystearate, PEG-4 dilaurate, PEG-8 dioleate, PEG-40 sorbitan peroleate, PEG-7 glyceryl cocoate, PEG-20 almond oil glyceride, PEG-25 hydrogenated castor oil, glyceryl stearate (and) PEG-100 stearate, PEG-7 olive oil ester, PEG-8 oleate, PEG-8 laurate, PEG-60 almond oil glyceride, PEG-20 methyl glucose sesquiterpenes Stearate, PEG-40 stearate, PEG-100 stearate, PEG-80 sorbitan lauryl ester, stearyl alcohol polyether-2, stearyl alcohol polyether-12, oleyl alcohol polyether-2, cetyl alcohol polyether-2, lauryl alcohol polyether-4, oleyl alcohol polyether-10, oleyl alcohol polyether-10 / polyoxyethylene ether 10 oleyl base ether, cetyl alcohol polyether-10, isostearyl alcohol polyether-20, cetyl stearyl alcohol polyether-20, oleyl alcohol polyether-20, stearyl alcohol polyether-20, stearyl alcohol polyether-21, cetyl alcohol polyether-20, isocetyl alcohol polyether-20, lauryl alcohol polyether-23, stearyl alcohol polyether-100, glyceryl citrate stearate, self-emulsifying stearate Glyceryl esters, stearic acid, salts of stearic acid, polyglycerol-3-methylglucose distearate, sorbitan oleate, sorbitan sesquioleate, sorbitan isostearate, sorbitan trioleate, lauryl glucoside, decyl glucoside, sodium stearoyl glutamate, sucrose polystearate, hydrated polyisobutylene, synthetic polymers (e.g., eicosene copolymers, acrylate / C10-30 alkyl acrylate crosspolymers, acrylate / stearyl alcohol polyether-20 methacrylate copolymers), PEG-22 / dodecyl ethylene glycol copolymers, PEG-45 / dodecyl ethylene glycol copolymers), polymer emulsifiers (e.g., hydrophobically modified polyacrylic acid,Such as acrylate / C10-30 alkyl acrylate crosspolymer), polyalkyl glycol ether (e.g., Brij 72 (polyoxyethylene (2) stearate ether) or Brij 721 (polyoxyethylene (21) stearate), polyglycerol-3-methylglucose distearate, lauryl glucoside (and) polyglycerol-2-dipolyhydroxystearate, glyceryl citrate, sodium cetearyl sulfate, cetearyl glucoside; polyglycerol-6 stearate (and) polyglycerol-6 behenate, cetearyl olive oil ester (and) sorbitol olive oil ester, arachidonic acid (and) behenol (and) arachidonic acid glucoside, cetearyl alcohol (and) cocoa glucoside, cocoa glucoside (and) coconut alcohol, PEG-100 stearate (and) glyceryl stearate, sodium stearoyl glutamate, stearyl alcohol polyether-20, stearyl alcohol polyether-21, stearyl alcohol polyether-25, stearyl alcohol polyether-2, cetearyl alcohol polyether-25 and cetearyl alcohol polyether-6. Polyglycerol esters or diesters of fatty acids (e.g., polyglycerol-4 diisostearate / polyhydroxystearate / sebate, PEG-30 dihydroxystearate, polyglycerol-3 diisostearate), polyglycerol esters of oleic acid / isostearate, polyglycerol-6 hexenyl ester, polyglycerol-4 oleate, magnesium stearate, sodium stearate, potassium laurate, potassium ricinoleate, sodium cocoate, sodium shea butter, potassium ricinoleate, sodium oleate, lauryl polyglycerol-3 polydimethylsiloxane and / or PEG-9 polydimethylsiloxane and / or cetyl PEG / PPG-10 / 1 polydimethylsiloxane and / or PEG-12 polydimethylsiloxane crosspolymers and / or PEG / PPG-18 / 18 polydimethylsiloxane),

[0132] - Co-surfactants, such as compounds selected from monoglycerides and diglycerides and / or fatty alcohols, such as behenol, cetyl alcohol, cetearyl alcohol, stearyl alcohol and glyceryl stearate.

[0133] - Thickeners, commonly used to adjust the viscosity of UV filter compositions, such as xanthan gum, microcrystalline cellulose, cellulose, cellulose gum, cross-linked acrylate / C10-C30 alkyl acrylate polymers, hydroxyethyl acrylate / sodium acryloyl dimethyl taurine copolymers, preferably xanthan gum and hydroxyethyl acrylate / sodium acryloyl dimethyl taurine copolymers.

[0134] -Alkanediols, such as 1,3-propanediol, butanediol and 1,2-butanediol.

[0135] - Fillers, such as starch and starch derivatives (e.g., tapioca starch, distarch phosphate, aluminum octenyl succinate, or sodium starch), pigments that have neither primary UV filtering nor coloring effects (e.g., Valvance Touch 210 or 250 available at dsm-firmenich), fumed Aerosils®, talc, polyethylene, nylon, and dimethylsilyl silica.

[0136] - Film-forming agents, such as those selected from polymers such as polyvinylpyrrolidone (PVP), capryloyl glycerol / sebaic acid copolymer, diisostearoyl polyglycerol-3 dimer linoleate, and ethyl cellulose.

[0137] - Skin-active ingredients, such as those used for whitening, preventing sunburn, treating hyperpigmentation, preventing or reducing acne, wrinkles, and fine lines, as well as those used for atrophy and / or anti-inflammation, anti-cellulite and slimming (e.g., phytidine), firming, energizing, self-tanning and soothing, and those used to improve elasticity and skin barrier.

[0138] - Chelating agents and / or isolating agents.

[0139] UV filter compositions can be formulated as liquids or emulsions (water-in-oil emulsions, etc.). After application, the composition forms a thin film on the skin surface, thereby providing UV protection.

[0140] Other UV filter composition types may include, but are not limited to: water-in-oil emulsions, water-in-oil-in-oil emulsions, water-in-oil-in-water emulsions, silicone oil-in-water emulsions, silicone oil-in-water emulsions, polymer gel creams, lipophilic single-phase oils, lipophilic single-phase gels, lipophilic single-phase rods, lipophilic-alcohol mixtures, hydrophilic single-phase liquids, hydrophilic single-phase gels, and powders.

[0141] UV filter compositions can be sprays, creams, lotions, mousses (foams) or powders, and can be packaged accordingly in bottles, jars, pump sprayers, or aerosols with suitable applicators.

[0142] Examples of UV filter compositions include, in particular, photoprotective agents (sunscreen products, sunscreens), such as sunscreen lotions, creams, sunscreen creams, sunscreen blocks, or topical medications or day creams, whether or not they bear an SPF (Sun Protection Factor) label, such as anti-aging agents, photoaging treatment agents, etc. Skin tanning agents (i.e., compositions used for artificial / non-sun tanning and / or to darken human skin), such as self-tanning creams and skin whitening agents.

[0143] "Ingredient function" refers to the role of an ingredient in a composition. Ingredients can perform multiple functions. For example, an ingredient function can be at least one of the following (but not limited to): UVA and / or UVB filter, emollient, emulsifier, aqueous phase thickener, oil phase thickener, preservative, complexing agent.

[0144] "Physicochemical properties of components" refers to the properties that can characterize the physical and / or chemical properties of components through empirical measurement or calculation. These properties can be obtained from the *Handbook of Organic Solvent Properties* or the *Handbook of Chemistry and Physics*. The following are some examples (but not limited to) the physicochemical properties of liquid components (such as emollients and solvents): viscosity, density, boiling point, molecular weight, freezing point, coefficient of thermal expansion, flash point, specific gravity, electrical conductivity, saturated vapor concentration, miscibility with water, solubility parameter, dipole moment, dielectric constant, polarity, refractive index, surface tension, dipole moment, latent heat, specific heat capacity, van der Waals volume, van der Waals surface area, and log Pow.

[0145] Physicochemical properties may include (but are not limited to): viscosity, density, boiling point, molecular weight, freezing point, coefficient of expansion of liquid, flash point, specific gravity, electrical conductivity, saturated vapor concentration, miscibility with water, solubility parameter, dipole moment, dielectric constant, polarity, refractive index, surface tension, dipole, latent heat, specific heat, van der Waals volume, van der Waals surface area, spreading properties, and spreading value.

[0146] Surfactants (such as wetting and dispersing agents, emulsifiers, foaming and defoaming agents, lubricants) may have the following properties (but are not limited to):

[0147] Solubility, surface tension, chemical properties, ionic characteristics (nonionic, anionic, cationic), HLB (hydrophilic-lipophilic balance), emulsification mechanism, ionization potential, ionic charge, and film-forming properties.

[0148] Other possible properties of UV filters include (but are not limited to): chemical structure, photostability, solubility, and hydrolytic stability.

[0149] Possible properties of humectants include (but are not limited to): hygroscopicity, water absorption, hydrophilicity, and osmotic pressure.

[0150] Possible properties of thickeners include (but are not limited to): viscosity, solubility, hydration, pH sensitivity, thermal stability, shear thinning / rheology, film-forming properties, and ionic charge.

[0151] Possible properties of polymer components include (but are not limited to): molecular weight, number-average molecular weight, weight-average molecular weight, polydispersity, molecular weight distribution, dielectric constant, elastic modulus, tensile strength, crystallinity, resilience, ultimate elongation, yield stress, yield strain, Rockwell hardness, notched impact strength, specific gravity, specific heat capacity, coefficient of thermal expansion, maximum continuous operating temperature, transparency, color, refractive index, heat capacity / thermal conductivity, heat of fusion, iodine value, acid value, saponification value, glass transition temperature, degree of branching, and degree of crosslinking.

[0152] Possible characteristics of insoluble pigment components include (but are not limited to): particle size, particle size distribution, polydispersity, coating chemistry, porosity, and particle characteristics.

[0153] A “value” can be defined relative to other components in the composition, or it can be defined in absolute units.

[0154] "Quantity" can be defined relative to other components or in absolute units; in addition, "quantity" can also refer to weight or concentration.

[0155] “Feature” refers to any data associated with a component numerical identifier or a set of UV filter composition numerical identifiers used to train predictive machine learning models.

[0156] The “overall UV spectrum” of a UV filter composition can be calculated using a variety of analytical techniques and instruments designed to measure the absorptivity or transmittance of ultraviolet radiation across a range of wavelengths. The overall UV spectrum of the composition can be calculated by considering the empirically measured individual absorption spectra of its components (e.g., UV filters, additives) and their concentrations in the UV filter composition. This calculation may involve summing the absorption contributions of each component to obtain the overall absorption spectrum of ultraviolet light. The “UV spectrum” of a single filter or any combination of UV filters can also serve as input features for predictive models.

[0157] The overall UV spectrum is defined as follows:

[0158] (1)

[0159] Where, ε i β i and MW i These are the molar extinction coefficients (L mol) of filter agent i. -1 cm -1 The concentration (%) and molecular weight (g / mol) of filter agent i are determined empirically. Similarly, the UV spectrum of filter agent i can be defined as... i (λ) is a linear function proportional to λ, while the UV spectrum of a UV filter combination can be defined as one or more ϵ iA linear combination of (λ), where j takes multiple values.

[0160] Therefore, one possible method for predicting SPF is to use the following method based on the Beer-Lambert law (Gamma):

[0161] Transmittance (T) is calculated on an irregular thin film based on:

[0162] (2)

[0163] Where d is the average thickness of the applied sunscreen (2mg / cm). 2 =>0.002cm), h(F) is the inverse cumulative distribution of the gamma distribution, and its expression is:

[0164] (3)

[0165] When the area under the film profile is normalized to 1, the following relationship must be satisfied:

[0166] (4)

[0167] This is the case when c = 1 / b in formula (3).

[0168] It is important to note that the overall UV spectrum of the UV filter composition can also be measured. In this document, "calculated overall UV spectrum" refers to a value representing the overall UV spectrum, regardless of whether that value is measured or calculated. "Calculated UV spectrum" refers to a value representing the UV spectrum of one or more individual filters, regardless of whether that value is measured or calculated.

[0169] Sun Protection Factor (SPF) is a measure of how well a sunscreen protects the skin from harmful ultraviolet (UV) radiation—primarily UVB radiation that causes sunburn. The SPF value indicates how long a person can remain in the sun without getting sunburned, compared to not taking any protective measures. SPF values ​​are typically determined through in vivo testing, i.e., testing on human subjects.

[0170] The SPF value can then be calculated using the following formula:

[0171] (5)

[0172] Where S er For the erythema effect spectrum, S s This represents the spectrum of the light source.

[0173] The shape parameter c, which describes the shape of the gamma distribution in formula (3), is given by the following formula:

[0174] (6)

[0175] Where OFC is the total amount of UV filter, calculated using the following formula:

[0176] (7)

[0177] Furthermore, REAE represents the proportion of erythema energy absorbed by the oil phase (o) to the total absorbed erythema energy (water phase (w) and oil phase), and its calculation formula is as follows:

[0178] (8)

[0179] (9)

[0180] Finally, the optimizer can be used to calibrate parameters p1 to p5.

[0181] "Oil-phase absorption of erythema energy" refers to the proportion of ultraviolet (UV) radiation that causes erythema (skin redness) that is absorbed by the oily components in sunscreen or other topical product formulations. This indicator is crucial for understanding the effectiveness of sunscreens in protecting the skin from UV-induced erythema.

[0182] "Total UV Filter Content" refers to the combined concentration or quantity of UV filter components present in a UV filter composition. UV filter components are active ingredients that can absorb, reflect, or scatter ultraviolet radiation, thereby protecting the skin from the harmful effects of sunlight.

[0183] A "predictive machine learning model" is a computational algorithm or mathematical framework designed to make predictions or forecasts based on input data. These models learn patterns and relationships from the input data, thereby generalizing and predicting unseen or future data instances. Predictive machine learning models are widely used in various fields such as finance, healthcare, marketing, and natural language processing.

[0184] Generally, a "numerical identifier" refers to any two-way numerical representation of a physical object, such as an ingredient or composition. For example, such a numerical identifier could correspond to an entry in a database. Numerical identifiers can also refer to labels representing the name, chemical structure, or internal reference information of an ingredient or composition.

[0185] In this specification, "materialized" or "assembled" refers to a substance existing outside the digital environment of this invention. "Materialized" or "assembled" can refer, for example, to a substance readily available in nature or a substance synthesized in a laboratory or chemical plant. In any case, the digital identifier of a materialized or assembled UV filter composition has a practical meaning. "To be formulated" or "formulated" refers to the materialization process of the digital identifier of the UV filter composition, whether this process is achieved through the extraction and assembly of components or through the synthesis and assembly of components.

[0186] In this document, the term "input mechanism" refers to, for example, a keyboard, mouse, and / or touchscreen, adapted to interact with a computer system to collect user input. In some cases, input mechanisms are inherently logical, such as a network port of a computer system configured to receive input commands transmitted electronically. Such input mechanisms may be associated with a graphical user interface (GUI) or application programming interface (API) displayed to the user. In other cases, an input mechanism may be a sensor configured to measure specific physical parameters associated with an intended use case. Figure 2 Examples of input mechanisms have been published.

[0187] As used herein, the terms “computing system,” “computer,” or “computer system” refer to any electronic computing device, whether monolithic or distributed, capable of receiving value input and providing value output through any type of interface (digital and / or analog). Typically, a computing system refers to a computer running software that can access data storage, or a client-server architecture where data and / or computation are performed on the server side while the client acts as the interface. Figure 2 Examples of such computing systems have been published.

[0188] As used herein, the term "database" refers to any physical and / or virtual data storage medium. Such a database can refer to computer memory accessible within a computer network, on which a database management system runs. In a simpler implementation, a database can correspond to a table stored in computer memory. Such tables can be stored in documents, such as Microsoft Excel documents.

[0189] Figure 1 A specific embodiment is shown that provides at least one predicted sun protection factor (SPF) value associated with a digital identifier of an ultraviolet (UV) filter composition, the digital identifier representing a materializable UV filter composition, the materializable UV filter composition being associated with at least one ingredient digital identifier representing an existing component of the materializable UV filter composition. Figure 1The different steps are shown, which can be divided into five different stages:

[0190] - The first phase involves building the database, including the following steps:

[0191] - Assembly step 105, which assembles the UV filter composition with at least one component.

[0192] - Calculation step 110, which calculates at least one UV spectrum of the assembled UV filter composition.

[0193] -Empirical measurement step 115, which empirically measures the SPF value of the assembled UV filter composition, and

[0194] -Storage step 120, which stores in a database at least one empirically measured SPF value and at least one calculated UV spectrum, the at least one calculated UV spectrum being associated with at least one digital identifier of the UV filter composition representing the assembled UV filter composition.

[0195] - The second stage involves preprocessing the training set (125), and will combine... Figure 5 The specific implementation methods shown will be described.

[0196] - The third stage involves training the predictive machine learning model, including the following steps:

[0197] - Step 135 is provided, which provides an original exemplary UV filter composition digital identifier set, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, including:

[0198] - At least one value representing an existing component, or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier.

[0199] - At least one calculated UV spectrum value associated with a digital identifier of the UV filter composition, and

[0200] - At least one empirically measured SPF value associated with a digital identifier of the UV filter composition.

[0201] This forms the training set, and

[0202] - Training step 140, which uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value.

[0203] - The fourth stage involves predicting at least one SPF value, including the following steps:

[0204] - Define step 145, which defines at least one digital identifier for the UV filter composition.

[0205] - Prediction step 150, which uses a trained predictive machine learning model to predict at least one SPF value, and

[0206] - Step 155 is provided, which yields the predicted SPF value.

[0207] - The fifth stage involves materializing at least one UV filter composition, including the following steps:

[0208] - Step 160 sends a digital command representing an instruction to materialize at least one UV filter composition, the at least one UV filter composition corresponding to a defined digital identifier for at least one UV filter composition.

[0209] - Materialization step 165, which materializes at least one UV filter composition corresponding to at least one defined UV filter composition digital identifier.

[0210] The first, second, and fifth stages are optional implementations of the method of the present invention.

[0211] The first phase involves building a database based on real-world experiments.

[0212] During the first stage, at least one UV filter composition is assembled 105 by at least one component in a manner known to those skilled in the art. After the UV filter composition is assembled, 110 at least one value representing the UV spectrum is calculated, and 115 at least one value representing the SPF value is empirically measured. Such measurements and calculations are known to those skilled in the art and can be based on Formula 1 as described above or any other method known to those skilled in the art.

[0213] During storage step 115, each assembled composition is associated with a digital identifier for the UV filter composition. The database associates the digital identifier of the UV filter composition with at least one component digital identifier of the UV filter composition, at least one value representing a calculated UV spectrum, and at least one value representing an empirically measured SPF value. The database may also store at least one of the following elements:

[0214] - At least one value representing the quantity of the component.

[0215] - At least one value representing the function of a component, associated with at least one component numerical identifier.

[0216] - At least one value representing the physicochemical properties of a component is associated with at least one component digital identifier or a combination of at least two component digital identifiers;

[0217] - At least one value representing the proportion of erythema energy absorbed in the oil phase, associated with at least one digital identifier for the UV filter composition, and / or

[0218] - At least one value representing the total amount of UV filter in the composition is associated with at least one digital identifier for the UF filter composition.

[0219] For example, in assembly step 105 of assembling a UV filter composition, the value of each at least one representative component is empirically measured.

[0220] For example, based on component numerical identifiers, at least one value representing a component function can be entered by the user or automatically through a user interface. Automatic input can be achieved, for example, by retrieving data from a database that associates at least one component numerical identifier with at least one component function.

[0221] For example, based on component numerical identifiers, at least one value representing the physicochemical properties of a component can be entered by the user or automatically via a user interface. For instance, automatic input can be achieved by retrieving data from a database that associates at least one component numerical identifier with at least one component physicochemical property.

[0222] A database that associates at least one component numerical identifier with at least one component physicochemical property, and a database that associates at least one component numerical identifier with at least one component function, may be the same database. In this database, at least one component physicochemical property may be associated with at least one component function. Examples of such associations are shown in Table 1 below.

[0223] [Table 1]

[0224]

[0225] After storage step 115, this database will be used in step 135 of the third phase, which provides the training set. Therefore, the predictive machine learning model is trained based at least on a portion of the elements constituting the aforementioned database.

[0226] The inclusion of the overall UV spectrum and OFC is due to their strong correlation with SPF values. Specifically, in the example dataset considered in this invention, E7 showed the highest correlation at ρ=0.756, while OFC showed a correlation of ρ=0.739. On the other hand, among the features other than OFC and extinction coefficient, one component showed the highest correlation at ρ=0.422. These findings are summarized in Table 2 below. Therefore, any predictive model will benefit from the inclusion of extinction coefficient and OFC, which is the strategy proposed in this invention.

[0227] [Table 2]

[0228]

[0229] In a particular embodiment, the computer-implemented method 100 further includes a preprocessing step 125 of preprocessing the training set, wherein the preprocessed training set is used during the providing step 135 of providing the training set. Therefore, the predictive machine learning model is trained based on at least a portion of the elements constituting the aforementioned reduced training set.

[0230] In certain implementations, the training set comprises a large amount of data, and a predictive machine learning model is used to determine which features are strongly correlated with the SPF value. Predictive machine learning models such as random forest regressors, gradient boosting regressors, and neural networks can learn the importance of features during training and distinguish unimportant features.

[0231] In certain implementations, the amount of training data is limited, thus posing a risk of overfitting due to the high dimensionality of the corresponding model. Feature engineering becomes particularly important in these implementations.

[0232] In a particular implementation, preprocessing step 125 includes:

[0233] - Grouping step 505, which groups at least one component digital identifier based on at least one value representing a component function associated with at least two component digital identifiers.

[0234] - The removal step uses domain expert knowledge or cross-validation to remove unimportant features.

[0235] - Optional determination step 510, which determines a value representing the influence of at least one group on the predicted SPF value.

[0236] - Optional selection step 515, which uses cross-validation to select at least one of the most influential groups, and

[0237] -Optional deletion step 520, which removes at least one group with less influence from the feature set.

[0238] In a particular implementation, during selection step 515, each group with a determined value higher than a predefined influence selection value is selected, which represents the influence on the predicted SPF value.

[0239] In a particular implementation, during the selection step 515, the method includes: a sorting step that sorts each group from highest to lowest value based on determined values ​​representing the impact on the predicted SPF value; and a selection step that selects at least a predefined number of groups having the highest value representing the impact on the predicted SPF value.

[0240] Preprocessing step 125 results in a significant reduction in the number of features. Therefore, in the example training set S1 described below, 430 components can be reduced to 130.

[0241] In a particular implementation, preprocessing step 125 includes a substitution step 525, which at least partially replaces the component numerical identifiers with values ​​representing at least one physicochemical property. Thus, the exemplary training set S2 described below produces the 21 features shown in Table 1 without loss of accuracy.

[0242] In certain embodiments, in the database, at least one value representing a function is associated with at least one value representing a physicochemical property. In these embodiments, alternative step 525 is performed before the grouping step, and during grouping step 505, at least one value representing a physicochemical property is grouped according to the value of at least one component function, which is associated with at least one value representing the same physicochemical property.

[0243] Preprocessing step 125 is optional and depends on the content of the training set used.

[0244] It should be noted that predictive machine learning models can be trained using only a training set initially formed using at least one value representing the UV spectrum and at least one of the following values:

[0245] -At least one value representing the function of the component,

[0246] - At least one value representing the physicochemical properties of the component is associated with at least one value representing the function of the component.

[0247] In the embodiments described above, the training set may further include:

[0248] - At least one value representing the physicochemical properties of the component, associated with at least one value representing the function of the component.

[0249] - At least one value representing the proportion of erythema energy absorbed in the oil phase, associated with at least one digital identifier for the UV filter composition, and / or

[0250] - At least one value representing the total amount of UV filter in the composition is associated with at least one digital identifier for the UV filter composition.

[0251] In other words, the database is independent of the component numerical identifiers.

[0252] Figure 1 A computer-implemented method 130 is also shown for training a predictive machine learning model to provide at least one value associated with a digital identifier of an ultraviolet (UV) filter composition, the digital identifier representing a materializable UV filter composition, the method comprising the following steps:

[0253] - Step 135 is provided, which provides an original exemplary UV filter composition digital identifier set, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, including:

[0254] - At least one existing component or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier, and

[0255] - The digital identifier of the UV filter composition is associated with at least one value representing the calculated UV spectrum, and

[0256] - The digital identifier of the UV filter composition is associated with at least one value representing an empirically measured SPF value.

[0257] This forms the training set.

[0258] - Prediction step 140, which uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value.

[0259] This method for training can itself be the method objective of the present invention, independent of the method 100 for providing at least one predicted SPF value.

[0260] For example, step 135 may be performed by a computer program executed by an electronic computing device (e.g., a microprocessor). During step 135, at least one digital identifier for a UV filter composition is provided to the predictive machine learning model to be trained. This step 135 may correspond to transferring the memory of an exemplary UV filter composition digital identifier from one digital storage location to another, the latter dedicated to the training set of the predictive machine learning model. During step 135, a large sample of UV filter composition digital identifiers representing different UV filter compositions may be divided into a training set and a validation set.

[0261] During the provisioning step, the provided training set may be the preprocessed training set obtained in the preprocessing step 125 described above, or the dataset independent of the component number identifiers described above.

[0262] Training step 140 can use any training method suitable for training predictive machine learning models.

[0263] Figure 1 The fourth stage of the method described relates to the prediction and provision of at least one predicted SPF value by a trained predictive machine learning model.

[0264] For example, prediction step 150 is performed by a computer program running on an electronic computing unit or computing system (e.g., a computer). During this step, a trained predictive machine learning model is executed, thereby providing a predicted SPF value as output.

[0265] Step 145 can be performed similarly to step 135 of method 100.

[0266] In a particular implementation, the trained predictive machine learning model is a random forest regressor.

[0267] Random forest regressors are an ensemble learning method that improves performance by combining the predictions of a group of decision trees. This is achieved by creating multiple bootstrap samples in the training set and training random trees on each sample. Specifically, the random forest algorithm uses two techniques to reduce the variance of the prediction function: bootstrapping and decorrelation of the trees. The variance of the forest can be expressed as:

[0268] (10)

[0269] in:

[0270] σ 2 It is the variance of a single tree.

[0271] ρ represents the pairwise correlation of the tree.

[0272] Where B is the number of trees.

[0273] Increasing the number of trees and their pairwise correlations can reduce variance. The random forest regressor is constructed using the following algorithm:

[0274] First, for b=1 to B:

[0275] (i) Draw a pilot sample Z of size N from the complete dataset. b

[0276] (ii) By randomly selecting feature m in Z bTree growing on top T b (x, θ b (, until the minimum node size n is reached) min until

[0277] Secondly, the random forest is defined as:

[0278] (11)

[0279] For example, the scikit-learn Python library provides a public implementation of a random forest regressor.

[0280] In a particular implementation, the trained predictive machine learning model is a gradient boosting regressor.

[0281] Gradient boosting regressors belong to a class of machine learning models that improve performance by combining the predictions of multiple models; this concept is known as ensemble learning. Specifically, a gradient boosting regressor is a sequential ensemble method that builds a series of models, each dedicated to minimizing the error of the preceding models. An additive model of the following form can be constructed:

[0282] (12)

[0283] in:

[0284] x is the input vector

[0285] c m m=1,2,…M are the expansion coefficients.

[0286] b m Let m = 1, 2, ..., M be basis functions, and

[0287] α is a set of parameters.

[0288] Then, by minimizing the training data Use the average loss function on the tagged surface to train the machine learning model:

[0289] (13)

[0290] In the boosting algorithm, minimizing the loss function is efficiently achieved through an iterative process called additive modeling in the forward phase. Optimization is performed sequentially by adding new basis functions to the additive model and adjusting only the new basis functions in each iteration of the training phase.

[0291] The gradient boosting regressor is constructed using the following algorithm:

[0292] First, initialize

[0293] Secondly, for m=1 to M:

[0294] -calculate (15)

[0295] -set up

[0296] The lifting model is F(x) = f M (x)

[0297] The optimization of the second step can be achieved by using a first- or higher-order algorithm.

[0298] A popular variation is gradient boosting, where, in each iteration, the gradient...

[0299] (16)

[0300] Used to fit the base learner to the dataset b_m. And the expansion coefficient c m Then, by solving a one-dimensional optimization problem (17) to fit. There are higher-order variants that can be scaled well to very large datasets and parallel environments.

[0301] In a preferred embodiment, the gradient boosting regressor uses a tree function.

[0302] Public implementations of gradient boosting machine learning models are provided in the scikit-learn or XGBoost libraries.

[0303] In a preferred embodiment, the trained predictive machine learning model is a neural network.

[0304] Neural networks represent a broad class of machine learning models. Their strong scalability and expressive power, combined with powerful optimization algorithms, enable them to drive numerous cutting-edge applications across various fields.

[0305] The basic form of a neural network consists of multiple layers of interconnected nodes (called neurons). These layers are typically referred to as input layers, hidden layers, and output layers. The input layer contains the input features, the hidden layers apply continuous (non-linear) transformations to the input layer, and the output layer is the result of these transformations. A neural network with L-1 hidden layers can be recursively defined as follows:

[0306] (18)

[0307] in:

[0308] a (l) It is the output of the l-th layer.

[0309] f(l) It is a non-linear activation function, typically including Sigmoid, Sinh, and ReLU.

[0310] W (l) It is a weight matrix

[0311] a (l-1) It is the output of layer l-1.

[0312] b (l) It is a bias vector

[0313] The input vector and the output vector are a (0) a (L)

[0314] a (l) The group of (x) is usually referred to as the activation value.

[0315] Given training data Neural networks are trained by minimizing a loss function:

[0316] (19)

[0317] Optimization typically uses first-order optimization algorithms, such as gradient descent.

[0318] Regularization techniques are crucial for preventing overfitting when training neural networks. Overfitting occurs when a model learns noise and spurious patterns in the data, causing it to perform poorly on unseen data. Overfitting is particularly problematic when the number of training samples is much smaller than the number of parameters in the neural network. This is common in SPF prediction because in vivo measurement data is very limited, while the effective size of the neural network is large. Common regularization techniques include:

[0319] -L2 regularization (weight decay): It introduces a penalty term into the loss function that is proportional to the sum of squared weights (elements of the weight matrix), thereby encouraging the optimization algorithm to keep smaller weights and effectively obtain a simpler machine learning model.

[0320] -L1 regularization: Introduces a penalty term proportional to the sum of the absolute values ​​of the model weights, thereby prompting the optimization algorithm to set many weights to zero, effectively simplifying the model; in addition, it also allows the model to ignore unrelated / redundant features, effectively achieving feature selection.

[0321] -Dropout: Randomly deactivates a subset of neurons during training; random deactivation is repeated in each training cycle and iteration, effectively applying training to a simpler model and preventing the network from fitting spurious and noisy patterns in the data.

[0322] An example neural network configuration that can be used for SPF quantization prediction can be described as follows. The number of neurons in the input layer corresponds to the number of features. The number of hidden layers can be set to three, including 64, 32, and 16 neurons respectively. The output node corresponds to a neuron used for SPF quantization.

[0323] An ensemble method configuration for SPF quantization prediction can be described as follows: an XGBoost regressor with an input vector dimension corresponding to the number of features; 100 tree estimators with a tree depth of 6; and output node quantization of SPF values.

[0324] The following example uses a training set based on 291 (N) UV filter compositions. Two training sets with eigenvectors can be considered, labeled S1 and S2 respectively.

[0325] The first training set (S1) includes the following features:

[0326] a) Weighted average of extinction coefficients n w =23 is defined as:

[0327]

[0328] in, is n f Multiply by n w The matrix, including n w n at each wavelength f The extinction coefficient of a UV filter, and is n f Vector, including the percentage concentration of UV filter in the UV filter composition.

[0329] b) OFC:

[0330] (25)

[0331] c) REAE, defined as follows:

[0332]

[0333] Among them, EAE i The erythematous activity of filter i is extinct, while S o and S w These are indexes for oil-soluble and water-soluble filter media, respectively. Erythematous extinction is defined as:

[0334]

[0335] Among them, E 11 (i,λ) is the specific extinction coefficient i of the filter at wavelength λ, S s(λ) is the intensity of the light source, while s er (λ) is the spectrum of erythema effects.

[0336] d) List of ingredients for this formula

[0337] The second training set (S2) includes features a)-c) and:

[0338] e) Feature list P1, where the function and feature names are shown in Table 1. For categorical variables, report the corresponding categories.

[0339] Given a set of input features, the following preprocessing steps can be applied to the training set:

[0340] - The standardization step, which standardizes each feature to a standard normal distribution.

[0341] - Removal step, which removes elements with a given associative threshold ρ t Highly correlated features

[0342] In this embodiment, the standardization step of standardizing each feature is achieved by calculating the mean and standard deviation of each feature and applying the following transformation:

[0343] (27)

[0344] The collinearity removal step is performed on the collinearity features and is achieved by calculating pairwise correlations:

[0345] (28)

[0346] Where, σ i σ j They are X i X j The standard deviation of cov(X), while i X j ) is the covariance:

[0347] (29)

[0348] Where N is the number of observations.

[0349] Given pairwise correlations, a set of uncorrelated features is generated sequentially by adding one feature at a time, such that the correlation between a feature and any previously selected feature is no greater than a certain threshold ρ. t The addition of features will continue until no new features satisfy the associativity constraints.

[0350] Both of these preprocessing steps are standard practice when developing machine learning models. These steps ensure good convergence of the training method 130 and result in better in-sample and out-of-sample model performance. Furthermore, other dimensionality reduction transformations can be applied to the dataset, such as t-distributed random neighborhood embedding (t-SNE), uniform manifold approximation and projection (UMAP), principal component analysis (PCA), kernel principal component analysis (KPCA), and linear discriminant analysis (LDA).

[0351] In a particular implementation, to obtain robust estimates of the performance and generalization ability of a predictive machine learning model, a K-fold cross-validation method known to those skilled in the art can be employed. This method divides the dataset into K equal-sized subsets, called "folds." In one cross-validation iteration, the model is trained on K-1 folds and validated on the remaining folds (the test set). This method is repeated K times, such that each fold is used only once for validation, while the remaining K-1 folds are used for training. The performance scores of the corresponding K training and test sets are then averaged to obtain a final estimate of the predictive machine learning model's performance. Optionally, K-fold cross-validation can be repeated multiple times by randomly shuffling the dataset before each repetition.

[0352] In certain implementations, when the predictive machine learning model includes feature engineering and hyperparameter tuning, a nested cross-validation strategy known to those skilled in the art is employed to evaluate the model's generalization ability. For K-fold cross-validation, the dataset is divided into K folds. In each iteration, K-1 folds are used for hyperparameter tuning, feature engineering, and training of the predictive machine learning model, while the remaining folds are used for predictive machine learning model validation (test set). In these implementations, the selection of model hyperparameters and feature engineering does not depend on any knowledge of the validation data, thus enabling a more accurate evaluation of the predictive machine learning model's generalization ability.

[0353] The four tables below illustrate the performance of the predictive machine learning models trained on datasets S1 and S2 using the nested cross-validation method described above.

[0354] In Example 1, the dataset used is S1. The table below illustrates the nested cross-validation results R based on the cross-validation training and test sets described above. 2 .

[0355]

[0356] Figure 3 This demonstrates the correlation between the measured SPF value and the SPF value calculated by a trained predictive machine learning model based on training set S1. Figure 3The accuracy of the trained predictive machine learning model used in the method of the present invention and the corresponding training method are shown.

[0357] The table below illustrates the results of nested cross-validation based on the training and test sets. 2 These cross-validation training and test sets only include the features c and d of S1.

[0358]

[0359] As can be seen from the figure, the correlation between the results is much lower, resulting in poor performance of the trained predictive machine learning model.

[0360] In Example 2, the dataset used is S2. The table below illustrates the results of nested cross-validation based on the cross-validation training and test sets, R. 2 .

[0361]

[0362] Figure 4 This demonstrates the correlation between measured SPF values ​​and SPF values ​​calculated by a trained predictive machine learning model based on training set S2. Figure 4 The accuracy of the trained predictive machine learning model used in the method of the present invention and the corresponding training method are shown.

[0363] The table below illustrates the results of nested cross-validation of R^2 based on cross-validation training and test sets, where the cross-validation training and test sets only include the S2 features d and e.

[0364]

[0365] As can be seen from the figure, the correlation between the results is much lower, resulting in poor performance of the trained predictive machine learning model.

[0366] Another approach to analyzing the correlation between additional features and model performance is to calculate feature importance scores. Several algorithms exist for calculating these scores, such as the SHAP method, importance scores based on feature permutations, or tree-based algorithms. Figure 6 The importance score calculated using the XGBoost library is displayed. It can be seen that UV spectra, as well as some other component and property characteristics, have a significant impact on the prediction results.

[0367] For example, features that can be used to train machine learning algorithms include at least one of the following:

[0368] - E290 capability of the component or composition

[0369] - Parameters of the liquid filter agent in the oil phase of the component or composition.

[0370] - E325 capability of the component or composition

[0371] - E330 capability of the ingredients or composition

[0372] - E335 capability of the ingredients or composition

[0373] - E375 capability of the ingredients or composition

[0374] -OFC of ingredients or compositions

[0375] - The viscosity of the component or composition,

[0376] - A parameter relating to the viscosity of the liquid filter agent in the composition or formulation.

[0377] -WoS parameters of the ingredients or composition

[0378] -REAE of ingredients or composition,

[0379] -A parameter representing the total weight of all filter media in a composition or component.

[0380] - E395 capability of the ingredients or composition

[0381] - The diffusion rate of the component or composition,

[0382] - The E400 capability of the ingredients or composition,

[0383] - The density of the component or composition, and / or,

[0384] - A filter agent in the aqueous phase of a component or composition.

[0385] The following functions are available:

[0386]

[0387] It should be noted that most of the features listed in this article are different from existing technologies.

[0388] This method can utilize the following new features:

[0389] -Calculated UV spectrum,

[0390] -Concentration of a single UV filter,

[0391] - Physicochemical properties of the components or compositions thereof

[0392] -Concentration of a single emollient

[0393] - The ratio of erythema absorbance to oil phase absorbance.

[0394] - Concentration of ingredients other than UV filters and emollients.

[0395] It should be noted that, for better generalization, physicochemical properties can be used to replace individual components.

[0396] The prediction method presented in this paper is a method for indirectly measuring the SPF value of sunscreen formulations. Specifically, this method first directly measures the physicochemical properties of the formulation or its components using specific sensors or experimental procedures. Then, the prediction method uses these data to indirectly (through computer simulation) measure the SPF value.

[0397] The method proposed in this paper also addresses the inherent limitations of machine learning algorithms. Specifically, when data is scarce (very common in the SPF field) and the correlation between features and output is poor, the performance of learning algorithms is often unsatisfactory. Therefore, even if all information about the input compound is implicitly or indirectly included in the input features, the model's performance may still be unsatisfactory because limited data availability and the limited complexity of the model do not allow for sufficient modeling of this implicit information. The method proposed in this paper addresses this problem by identifying features that are both highly informative and highly correlated with the output. The examples above clearly demonstrate this. When using only the ingredients and REAE (which indirectly includes all information about the formulation), the model's performance is relatively poor, while the performance is significantly improved by adding UV spectroscopy.

[0398] Importantly, SPF prediction methods can be used to develop new sunscreen formulations. Specifically, formulators may need to develop dozens or even hundreds of formulations with only a specific target SPF value. This method can be used to select only a subset of formulations that meet the required SPF value. The selected formulations can then be further optimized by the formulator and characterized through in vivo experiments. Therefore, this prediction method can be directly integrated into the development process, reducing unnecessary and costly screening rounds and improving development efficiency.

[0399] During the fifth optional stage, there is a sending step 160, which generates and sends a digital command representing an instruction to materialize at least one UV filter composition to a device capable of materializing at least one UV filter composition, the UV filter composition corresponding to a defined digital identifier for at least one UV filter composition.

[0400] The materialization step 165 can be performed manually (e.g., via a user interface) or automatically (via a device capable of materializing at least one UV filter composition corresponding to at least one digital identifier of a UV filter composition defined based on a sent command).

[0401] Figure 2The diagram illustrates an example computer system 200 that can be used to implement one embodiment of the present invention. In the example of FIG8, computer system 205 and instructions for implementing the disclosed techniques in hardware, software, or a combination of hardware and software are represented schematically, for example by boxes and circles, at a level of detail similar to that typically used by those skilled in the art when discussing computer architecture and computer system implementation.

[0402] Computer system 205 includes an input / output (I / O) subsystem 220, which may include a bus and / or other communication mechanisms for transmitting information and / or instructions between various components of computer system 205 via electronic signal paths. I / O subsystem 220 may include an I / O controller, a memory controller, and at least one I / O port. Electronic signal paths are schematically represented in the accompanying drawings, for example, by lines, unidirectional arrows, or bidirectional arrows.

[0403] At least one hardware processor 210 is coupled to the I / O subsystem 220 for processing information and instructions. The hardware processor 210 may include, for example, a general-purpose microprocessor or microcontroller, and / or a special-purpose microprocessor, such as an embedded system, a graphics processing unit (GPU), a digital signal processor, or an ARM processor. The processor 210 may include an integrated arithmetic logic unit (ALU) or may be coupled to a separate ALU.

[0404] Computer system 205 includes one or more memory units 225, such as main memory, connected to I / O subsystem 220 for electronically storing data and instructions to be executed by processor 210. Memory 225 may include volatile memory, such as various forms of random access memory (RAM) or other dynamic storage devices. Memory 225 may also be used to store temporary variables or other intermediate information during processor 210 instruction execution. When these instructions are stored in a non-transitory computer-readable storage medium accessible to processor 210, computer system 205 can become a dedicated machine specifically for performing the operations specified in the instructions.

[0405] Computer system 205 also includes non-volatile memory, such as read-only memory (ROM) 230 or other static storage devices coupled to I / O subsystem 220, for storing information and instructions required by processor 210. ROM 230 may include various forms of programmable read-only memory (PROM), such as erasable PROM (EPROM) or electrically erasable PROM (EEPROM). Persistent storage unit 215 may include various forms of non-volatile random access memory (NVRAM), such as flash memory, solid-state memory, magnetic disk, or optical disk (e.g., CD-ROM or DVD-ROM), and may be coupled to I / O subsystem 220 to store information and instructions. Storage unit 215 is an example of a non-volatile computer-readable medium that can be used to store instructions and data that, when executed by processor 210, trigger computer-implemented execution methods to perform the techniques described herein.

[0406] Instructions in memory 225, ROM 230, or memory 215 may include one or more instruction sets, which may be organized into modules, methods, objects, functions, routines, or calls. These instructions may be organized into one or more computer programs, operating system services, or applications (including mobile applications). These instructions may include operating system and / or system software; one or more libraries for supporting multimedia, programming, or other functions; data protocol instructions or protocol stacks for implementing TCP / IP, HTTP, or other communication protocols; file format processing instructions for parsing or rendering files encoded using HTML, XML, JPEG, MPEG, or PNG; user interface instructions for rendering or interpreting graphical user interface (GUI), command-line interface, or text user interface commands; and application software such as office suites, internet access applications, design and manufacturing applications, graphical applications, audio applications, software engineering applications, educational applications, games, or other applications. These instructions may implement web servers, web application servers, or web clients. Instructions may be organized into presentation layers, application layers, and data storage layers, such as using Structured Query Language (SQL) or SQL-free relational database systems, object stores, graph databases, flat file systems, or other data storage.

[0407] Computer system 205 can be connected to at least one output device 235 via I / O subsystem 220. In one embodiment, output device 235 is a digital computer display or human-machine interface. Examples of displays that can be used in various embodiments include touch screen displays, light-emitting diode (LED) displays, liquid crystal displays (LCDs), or electronic paper displays. Computer system 205 may include other types of output devices 235 as alternatives to or supplements to the display device. Examples of other output devices 235 include printers, ticket printers, plotters, projectors, sound cards or video cards, speakers, buzzers or piezoelectric devices or other sound devices, indicator lights or LED or LCD indicators, haptic devices, actuators, or servo mechanisms.

[0408] At least one input device 240 is connected to the I / O subsystem 220 for transmitting signals, data, command selections, or gestures to the processor 210. Examples of input devices 240 include touchscreens, microphones, still and video digital cameras, alphanumeric keys and other keys, keypads, keyboards, graphics tablets, image scanners, joysticks, clocks, switches, buttons, knobs, and sliders.

[0409] Another input device is a control device 245, which can perform cursor control or other automatic control functions, such as navigation in a graphical interface on the display screen. These functions can be performed in parallel with or independently of the input function. Control device 245 can be a touchpad, mouse, trackball, or cursor arrow keys, used to transmit directional information and command selection to processor 210 and control cursor movement on display 235. The input device has at least two degrees of freedom on two axes, namely a first axis (e.g., the x-axis) and a second axis (e.g., the y-axis), which allows the device to specify a position in a plane. Another type of input device is a wired, wireless, or optical control device, such as a joystick, handle, console, steering wheel, pedals, gear shift mechanism, or other types of control devices. Input device 240 can include a combination of multiple different input devices, such as a camera and a depth sensor.

[0410] In another embodiment, computer system 205 may include Internet of Things (IoT) devices, wherein one or more of output device 235, input device 240, and / or control device 245 are omitted. Alternatively, in such an embodiment, input device 240 may include one or more cameras, motion detectors, thermometers, microphones, seismic detectors, other sensors or detectors, measuring devices, or encoders, while output device 235 may include a dedicated display (e.g., a single-line LED or LCD display), one or more indicators, display panels, instruments, valves, solenoid valves, actuators, or servo mechanisms.

[0411] Computer system 205 may implement the techniques described herein using custom hardwired logic, at least one application-specific integrated circuit (ASIC) or field-programmable gate array (FPGA), firmware, and / or program instructions or logic. When this hardware or logic is loaded and used or executed in conjunction with the computer system, the computer system operates as a dedicated machine. According to one embodiment, computer system 205 executes the techniques described herein in response to processor 210 executing at least one sequence of at least one instruction included in main memory 225. These instructions may be read into main memory 225 from other storage media, such as memory 215. Executing the sequence of instructions included in main memory 225 causes processor 210 to perform the processing steps described herein. In other embodiments, hardwired circuitry may be used in place of or in combination with software instructions.

[0412] As used herein, the term "storage medium" refers to any non-volatile medium used to store data and / or instructions, thereby enabling a machine to operate in a particular manner. Such storage media can include non-volatile media and / or volatile media. Non-volatile media include, for example, optical discs or magnetic disks, such as memory 215. Volatile media include dynamic memory, such as memory 225. Common forms of storage media include, for example, hard disks, solid-state drives, flash drives, magnetic data storage media, any optical or physical data storage media, memory chips, etc.

[0413] Storage media and transmission media are different, but can be used in combination. Transmission media are used to transfer information between storage media. Examples of transmission media include coaxial cables, copper wires, and optical fibers, including the conductors that form the I / O subsystem 220 bus. Transmission media can also be sound waves or light waves, such as those generated during radio wave and infrared data communication.

[0414] When transmitting at least one sequence of instructions to processor 210 for execution, various media may be required. For example, the instructions may initially be stored on a disk or solid-state drive of a remote computer. The remote computer may load the instructions into its dynamic memory and transmit them using a modem via a communication link (e.g., fiber optic, coaxial cable, or telephone line). A modem or router local to computer system 205 may receive data on the communication link and convert it into a format readable by computer system 205. For example, a receiver (e.g., an RF antenna or infrared detector) may receive data transmitted in a wireless or optical signal, and appropriate circuitry may provide the data to I / O subsystem 220 (e.g., placing the data on a bus). I / O subsystem 220 transmits the data to memory 225, from which processor 210 retrieves and executes the instructions. Instructions received by memory 225 may be selectively stored in memory 215, either before or after processor 210 executes the instructions.

[0415] Computer system 205 also includes a communication interface 260 connected to bus 220. Communication interface 260 provides bidirectional data communication and connects to a network link 265 that is directly or indirectly connected to at least one communication network (e.g., network 270 or a public or private cloud on the Internet). For example, communication interface 260 may be an Ethernet interface, an Integrated Services Digital Network (ISDN) card, a cable modem, a satellite modem, or a modem for providing data communication connectivity with a corresponding type of communication line (e.g., Ethernet cable, any type of metal cable, fiber optic line, or telephone line). Network 270 broadly refers to a local area network (LAN), a wide area network (WAN), a campus network, an interconnected network, or any combination thereof. Communication interface 260 may include a LAN card for providing data communication connectivity with a compatible LAN; or a cellular wireless telephone interface for sending or receiving cellular data according to cellular wireless telephone network standards; or a satellite radio interface for sending or receiving digital data according to satellite wireless network standards. In any such implementation, communication interface 260 sends and receives electrical, electromagnetic, or optical signals through signal paths carrying digital data streams representing various types of information.

[0416] Network link 265 typically provides data communication, including electrical, electromagnetic, or optical signals, to other data devices, either directly or via at least one network (e.g., using satellite, cellular, Wi-Fi, or Bluetooth technologies). For example, network link 265 can be connected to host 250 via network 270.

[0417] Furthermore, network link 265 can connect to other computing devices via network 270 or via interconnected devices and / or computers operated by Internet Service Provider (ISP) 275. ISP 275 provides data communication services via a global packet data communication network (i.e., the Internet 280). Server 255 can connect to the Internet 280. Server 255 broadly refers to any computer, data center, virtual machine, or virtual computing instance (with or without a hypervisor), or computer running a containerized program system (such as DOCKER or KUBERNETES). Server 255 can also refer to an electronic digital service implemented using multiple computers or instances, which can be accessed and used by sending Web service requests, Uniform Resource Locator (URL) strings with HTTP payload parameters, API calls, application service calls, or other service calls. Computer system 205 and server 255 can constitute components of a distributed computing system that includes other computers, processing clusters, server clusters, or other computer organizations that work together to perform tasks or run applications or services. Server 255 may include one or more instruction sets organized as modules, methods, objects, functions, routines, or calls. These instructions can be organized into one or more computer programs, operating system services, or applications, including mobile applications. These instructions may include operating systems and / or system software; one or more libraries for supporting multimedia, programming, or other functions; data protocol instructions or protocol stacks for implementing TCP / IP, HTTP, or other communication protocols; file format processing instructions for parsing or rendering files encoded using HTML, XML, JPEG, MPEG, or PNG; user interface instructions for rendering or interpreting graphical user interface (GUI), command-line interface, or text user interface commands; and application software such as office suites, internet access applications, design and manufacturing applications, graphics applications, audio applications, software engineering applications, educational applications, games, or other applications. Server 255 may include a web application server that hosts the presentation layer, application layer, and data storage layer, such as relational database systems using Structured Query Language (SQL) or SQL-free relational database systems, object stores, graph databases, flat file systems, or other data storage.

[0418] Computer system 205 can send messages and receive data and instructions, including program code, via a network, network link 265, and communication interface 260. In the Internet example, server 255 can send application request code via Internet 280, ISP 275, local network 270, and communication interface 260. Received code can be executed immediately by processor 210 and / or stored in memory 215 or other non-volatile memory for subsequent execution.

[0419] The execution of the instructions described in this section can realize a process, which runs as an instance of a computer program, including program code and its current activity. Depending on the operating system (OS), a process can consist of multiple execution threads that execute instructions concurrently. In this context, a computer program is a passive collection of instructions, while a process is the actual execution of those instructions. Multiple processes can be associated with the same program; for example, opening multiple instances of the same program typically means that multiple processes are executing. Multitasking can be implemented, allowing multiple processes to share processor 210. While each processor 210 or processor core can only execute one task at a time, computer system 205 can be programmed to implement multitasking, allowing each processor to switch between executing tasks without waiting for each task to complete. In one implementation, switching can occur when a task performs an input / output operation, indicates that it can switch, or a hardware interrupt occurs. To achieve rapid responsiveness in interactive user applications, time-sharing techniques can be used to create the illusion of parallel execution by rapidly performing context switching, allowing multiple processes to execute simultaneously. In one implementation, to ensure security and reliability, the operating system can prevent direct communication between independent processes, thereby providing strictly controlled inter-process communication functionality.

Claims

1. A computer-implemented method (100) for providing at least one predicted sun protection factor (SPF) value associated with a digital identifier of an ultraviolet (UV) filter composition, the digital identifier representing a materializable UV filter composition, the materializable UV filter composition being associated with at least one component digital identifier representing an existing ingredient of the materializable UV filter composition, characterized in that, The method includes: - Providing step (135), which provides an original exemplary UV filter composition digital identifier set, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, including: - At least one value representing an existing component, or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier. - At least one calculated UV spectrum value associated with a digital identifier of the UV filter composition, and - At least one value representing an empirically measured SPF value associated with a digital identifier of the UV filter composition. This forms the training set. - Training step (140), which uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value. - Define step (145), which defines at least one digital identifier for the UV filter composition. - Prediction step (150), which uses a trained predictive machine learning model to predict at least one SPF value; - Provide step (155), which provides at least one predicted SPF value.

2. The computer-implemented method (100) according to claim 1, wherein, In the training set, at least one value representing an existing component is the number of components.

3. The computer-implemented method (100) according to claim 1 or 2, wherein, In the training set, at least one value representing an existing component is a value representing the function of an existing component.

4. The computer-implemented method (100) according to any one of claims 1 to 3, wherein, In the training set, at least one value representing an existing component or combination of existing components is a value representing the physicochemical properties of an existing component or combination of existing components.

5. The computer-implemented method (100) according to claims 2, 3, and 4, wherein, The training set includes more than 20 features, which are selected from at least one of each value representing the quantity of existing components, the function of existing components, and the physicochemical properties of existing components or representing combinations of components.

6. The computer-implemented method (100) according to any one of claims 1 to 5, wherein, The training set includes at least one value representing the proportion of erythema energy absorbed in the oil phase, which is associated with at least one composition digital identifier; and / or at least one value representing the total amount of UV filter in the composition, which is associated with at least one UV filter composition digital identifier.

7. The computer-implemented method (100) according to any one of claims 1 to 6, comprising: - Assembly step (105), which assembles at least one component into a UV filter composition. - Calculation step (110), which calculates at least one UV spectrum of the assembled UV filter composition, or at least one UV spectrum of a combination of one or more components. -Empirical measurement step (115), which empirically measures the SPF value of the assembled UV filter composition, and - Storage step (120), which stores in a database at least one empirically measured SPF value and at least one calculated UV spectrum, the UV spectrum being associated with at least one digital identifier of the UV filter composition representing the assembled UV filter composition. The database is used during the provisioning step (135) of providing the training set.

8. The computer-implemented method (100) according to claim 7 further includes a preprocessing step (125) which preprocesses the training set, and wherein, The preprocessed training set is used in the provision step (135) of providing the training set.

9. The computer-implemented method (100) according to any one of claims 1 to 8, wherein, The trained predictive machine learning model is the random forest regressor.

10. The computer-implemented method (100) according to any one of claims 1 to 8, wherein, The trained predictive machine learning model is a gradient boosting regressor.

11. The computer-implemented method (100) according to any one of claims 1 to 8, wherein, A trained predictive machine learning model is a neural network.

12. The computer-implemented method (100) according to any one of claims 1 to 11, further comprising a sending step (160) of sending a digital command representing an instruction for materializing at least one UV filter composition, the at least one UV filter composition corresponding to a defined at least one UV filter composition digital identifier.

13. The computer-implemented method (100) according to claim 12 further includes a materialization step (165) which materializes at least one UV filter composition corresponding to a defined digital identifier of at least one UV filter composition.

14. A computer-implemented method (130) for training a predictive machine learning model to provide at least one value associated with a digital identifier of an ultraviolet (UV) filter composition, the digital identifier representing a materializable UV filter composition, the method comprising: - Providing step (135), which provides an original exemplary UV filter composition digital identifier set, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, including: - At least one value representing an existing component, or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier. - At least one calculated UV spectrum value associated with a digital identifier of the UV filter composition, and - At least one value representing an empirically measured SPF value associated with a digital identifier of the UV filter composition. This forms the training set, and - Training step (140), which uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value.

15. A computer program product (200), characterized in that, It includes instructions that, when executed by a computer, cause the computer to perform the method (100) according to any one of claims 1 to 13.

16. A computer-readable storage medium (215) storing programming instructions that, when executed by a computer, cause the computer to perform the method (100) according to any one of claims 1 to 14.

17. An apparatus (200) for providing at least one predicted sun protection factor (SPF) value associated with a digital identifier of an ultraviolet (UV) filter composition, the digital identifier representing a materializable UV filter composition, the materializable UV filter composition being associated with at least one component digital identifier representing an existing component of the materializable UV filter composition, characterized in that, The device includes: - A provider that provides an original set of exemplary UV filter composition digital identifiers, the exemplary UV filter composition digital identifier set representing a materialized UV filter composition, comprising: - At least one value representing an existing component, or at least one value representing a combination of existing components, each existing component being associated with a component numerical identifier. - At least one calculated UV spectrum value associated with a digital identifier of the UV filter composition, and - At least one value representing an empirically measured SPF value associated with a digital identifier of the UV filter composition. This forms the training set, and - A training facility that uses a training set to train a predictive machine learning model, wherein the predictive machine learning model is trained to predict at least one SPF value. - Define an organization that defines at least one digital identifier for a UV filter composition. - A prediction agency that uses a trained predictive machine learning model to predict at least one SPF value; - Provide an agency that provides at least one predicted SPF value.

Images