Methods, systems, and computer program products for generating compound structures

Abstract

The invention provides a method, a system and a computer program product for generating a compound structure. The method comprises the following steps: providing a compound structure material library and a background compound with known physical and chemical parameters for a large language model; providing limiting conditions and expected physical and chemical parameters for the large language model; generating candidate compounds conforming to the provided physical and chemical parameters by using a large language model based on a compound structure material library and the physical and chemical parameters under limiting conditions; combining the candidate compound with the background compound to form a new combination of background compounds; the combination of the compound structure material library and the new background compound is provided to the large language model to iteratively generate the structure of the expected compound.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computational chemistry, in particular, to a method, system and computer program product for generating compound structures using a large language model. BACKGROUND

[0002] In the field of computational chemistry, compound screening can be regarded as a search problem in a restricted space, and the importance of artificial intelligence in this field is increasingly valued. With the development of large language models (LLM), whether LLM can be used to generate compounds efficiently and accurately has become a problem of concern. SUMMARY

[0003] The present application provides a method for generating compound structures using an LLM, comprising: providing a compound structure material library and a background compound with known physical and chemical parameters to the LLM; providing a restriction condition and an expected physical and chemical parameter to the LLM; generating a candidate compound that meets the provided physical and chemical parameters under the restriction condition based on the compound structure material library and the physical and chemical parameters using the LLM; combining the candidate compound with the background compound to form a new background compound combination; providing the compound structure material library and the new background compound combination to the LLM to iteratively generate the structure of the expected compound.

[0004] According to an embodiment of the present application, when the background compound with known physical and chemical parameters is provided to the LLM, the physical and chemical parameters of the background compound are also provided to the LLM; when the compound structure material library and the new background compound combination are provided to the LLM, the physical and chemical parameters of the new background compound are also provided to the LLM.

[0005] According to an embodiment of the present application, the restriction condition includes the charge of a transition metal complex (TMC).

[0006] According to an embodiment of the present application, the expected physical and chemical parameters include at least two different physical and chemical parameters.

[0007] According to an embodiment of the present application, the compound includes an organic compound, an inorganic compound, a metal-organic complex, a polymer compound, a small molecule compound, and a biological macromolecule.

[0008] According to an embodiment of the present application, combining the candidate compound with the background compound to form a new background compound combination includes: combining the candidate compound and the top K compounds with the best physical and chemical parameters in the background compound to form a new background compound combination.

[0009] According to an embodiment of the present application, generating the candidate compound conforming to the provided physical and chemical parameters under the restriction condition by the LLM based on the compound structure material library and the physical and chemical parameters comprises: generating the candidate compound based on new compound materials outside the compound structure material library.

[0010] According to an embodiment of the present application, after generating the candidate compound based on the new compound materials outside the compound structure material library, the method further comprises: integrating the new compound materials into the compound structure material library.

[0011] According to an embodiment of the present application, the compound is a TMC, the compound structure material library is a ligand library, and the physical and chemical parameters comprise a HOMO-LUMO energy gap and a polarizability.

[0012] The present application also provides a system for generating a compound structure by an LLM, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the following steps: providing a compound structure material library and background compounds with known physical and chemical parameters to the LLM; providing a restriction condition and expected physical and chemical parameters to the LLM; generating a candidate compound conforming to the provided physical and chemical parameters under the restriction condition by the LLM based on the compound structure material library and the physical and chemical parameters; combining the candidate compound with the background compounds to form a combination of new background compounds; and providing the combination of the compound structure material library and the new background compounds to the LLM to iteratively generate a structure of an expected compound.

[0013] The present application also provides a computer program product comprising computer program instructions which, when executed by a processor, implement the following steps: providing a compound structure material library and background compounds with known physical and chemical parameters to the LLM; providing a restriction condition and expected physical and chemical parameters to the LLM; generating a candidate compound conforming to the provided physical and chemical parameters under the restriction condition by the LLM based on the compound structure material library and the physical and chemical parameters; combining the candidate compound with the background compounds to form a combination of new background compounds; and providing the combination of the compound structure material library and the new background compounds to the LLM to iteratively generate a structure of an expected compound.

[0014] The method, system and computer program product for generating a compound structure provided by the present application can utilize the knowledge inside the LLM and external chemical data, bypass the design of complex mathematical formulas, and provide efficient and accurate compound structure generation. BRIEF DESCRIPTION OF DRAWINGS

[0015] Other features, objects, and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the following drawings:

[0016] Figure 1 is a flowchart of a method for generating compound structures using LLM according to embodiments of the present application;

[0017] Figure 2 is a schematic diagram of a method for generating compound structures using LLM according to embodiments of the present application;

[0018] Figure 3 is a schematic diagram of a system for generating compound structures using LLM according to embodiments of the present application;

[0019] Figure 4 is a structural diagram of 50 ligands and their complexes in embodiments of the present application;

[0020] Figure 5 is a structural diagram of 10 new ligands generated using LLM;

[0021] Figure 6 is a comparison of the effects of a method for generating compound structures using LLM and other methods;

[0022] Figure 7 is a schematic diagram of an analysis of the top 20 TMCs out of 200 generated candidate TMCs. DETAILED DESCRIPTION

[0023] For a better understanding of the present application, reference will be made to the detailed description of the technical solutions of the present application in conjunction with the accompanying drawings. It should be understood that the detailed description is only a description of exemplary embodiments of the present application, and is not intended to limit the scope of the present application in any way. Throughout the specification, the same reference signs refer to the same elements. The expression “and / or” includes any combination or all combinations of one or more of the associated listed items.

[0024] It should be noted that in the present specification, the expressions “first”, “second”, “third”, etc. are only used to distinguish one feature from another feature, and do not represent any limitation on the features. In the drawings, the sizes, proportions, and shapes of the legends have been slightly adjusted for ease of illustration. The drawings are merely examples and are not drawn strictly to scale. As used herein, the words “approximately”, “about”, and similar words are used as approximate terms, not as terms of degree, and are intended to account for inherent deviations in measured or calculated values that would be recognized by one of ordinary skill in the art.

[0025] It should also be understood that any reference to or discussion of a term or expression in this specification refers to the meaning of that term or expression in the context of the specification, and not to the literal meaning of the term or expression. It should also be understood that expressions such as “including,” “including but not limited to,” “including one of,” “including one or more of,” “including at least one of,” “including but not limited to one of,” and “including one, some, or all of” are open-ended expressions that are intended to have the same meaning as the term “comprising” as if each of the foregoing expressions were explicitly and individually stated herein. It should also be understood that the term “comprising” as used in this specification is intended to have the same meaning as the term “including,” and that the term “comprising” is intended to be open-ended and allow for the possibility that there are additional elements or steps other than the elements or steps recited in the specification. It should also be understood that the term “method” as used herein refers to a computer-implemented process that is performed by a computer system or other computing device, and includes any acts or operations that are performed by the computer system or other computing device.

[0026] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0027] It should be noted that the features of the embodiments and examples of the present application can be combined with each other as long as there is no conflict. In addition, the specific steps included in the methods described in the present application are not necessarily limited to the order described unless specifically limited or contradicted by the context. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0028] Figure 1 is a flowchart of a method for generating a compound structure using an LLM according to an embodiment of the present application. Figure 2 is a schematic diagram of a method for generating a compound structure using an LLM according to an embodiment of the present application.

[0029] In step S1010, the LLM 2100 is provided with a compound structure material library 2200 and a background compound 2300 of known physical and chemical parameters. The LLM is a deep learning model trained and optimized using a large amount of text data, and can understand, process, and / or generate natural language text. The LLM can handle a variety of natural language tasks such as text classification, question answering, dialogue, etc. The LLM in the present application can be o1 -preview or claud e-3.5-sonnet. However, the embodiments of the present application are not limited thereto. The present application can be applied to any LLM with inherent chemical knowledge. The compound structure material library 2200 can include constituent units for a compound, such as atoms, ions, ligands, etc. The background compound 2300 of known physical and chemical parameters can be a compound of the same class as the intended compound.

[0030] Molecular orbital theory posits that electrons in a molecule do not belong to an atom but move throughout the molecule. Atomic orbitals in a molecule form molecular orbitals through linear combination, which are divided into two types: bonding orbitals and antibonding orbitals. Bonding orbitals allow electrons to form chemical bonds between atoms, while antibonding orbitals cause electrons to move away from the region between atoms. According to molecular orbital theory, electrons in a molecule fill molecular orbitals in order of energy from low to high. Among them, the highest occupied orbital is called the highest occupied molecular orbital (HOMO), and the lowest unoccupied orbital is called the lowest unoccupied molecular orbital (LUMO).

[0031] The HOMO-LUMO energy gap (i.e., the energy gap) refers to the energy difference between the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO). This energy gap has an important influence on the chemical properties and reactivity of the molecule. Generally speaking, the smaller the HOMO-LUMO energy gap, the less stable the chemical bond in the molecule. This is because a small energy gap means that electrons are easily transferred from the HOMO to the LUMO, leading to the breaking of chemical bonds. Conversely, a large energy gap means that chemical bonds are relatively stable. Secondly, the HOMO-LUMO energy gap is an important parameter for measuring the reactivity of a molecule. The smaller the energy gap, the higher the reactivity of the molecule. This is because electrons are easily transferred between the HOMO and the LUMO, thereby participating in chemical reactions. For example, in photochemical reactions, molecules with a smaller HOMO-LUMO energy gap are more likely to absorb photons, triggering a reaction. In addition, the HOMO-LUMO energy gap is also related to many properties of the molecule, such as redox properties, spectral properties, etc. For example, the higher the HOMO energy level, the easier it is for the molecule to lose electrons; the lower the LUMO energy level, the easier it is for the molecule to gain electrons. In addition, the HOMO-LUMO energy gap also affects the absorption and emission spectra of the molecule. Therefore, studying the HOMO-LUMO energy gap and other physicochemical property indicators of metal-organic complexes has practical significance in the application of compounds in catalysts, optoelectronic materials, drug design, etc.

[0032] The physical and chemical parameters can be characteristic parameters of the compound. For example, in the case of a compound being a TMC, the characteristic parameters can be HOMO-LUMO gap, polarizability, etc., and the background compound 2300 can be a TMC constructed using a ligand in the compound structure material library 2200.

[0033] At step S1020, the restriction condition 2400 and the expected physicochemical parameter 2500 are provided to the LLM. It is to be noted that there is no order limitation between step S1010 and step S1020. Further, the order of inputting the compound structure material library 2200, the background compound 2300, the restriction condition 2400, and the physicochemical parameter 2500 to the LLM 2100 is not limited. The restriction condition 2400 can be set according to the purpose of generating the compound. For example, in the case of the compound being a TMC, the restriction condition can be set to be that the total charge of the TMC is -1, 0, or 1. For another example, the restriction condition can also be set to be that the ligand in the generated candidate compound should be a ligand present in the compound structure material library 2200. The expected physicochemical parameter 2500 can be an optimization task set according to the generation purpose. According to the optimization task, one physicochemical parameter 2500 can be set for single-target optimization, or at least two different physicochemical parameters 2500 can be set for multi-target optimization. For example, in the case of the compound being a TMC, the expected physicochemical parameter 2500 can be “a larger energy gap than the HOMO-LUMO energy gap of a given TMC” and / or “a larger polarizability than the polarizability of a given TMC”.

[0034] At step S1030, the LLM 2100 is used to generate candidate compounds 2600 that meet the provided physicochemical parameters under the restriction condition based on the compound structure material library 2200 and the physicochemical parameter 2300. Specifically, the number of candidate compounds 2600 to be generated can be set when generating by using the LLM 2100.

[0035] At step S1040, the candidate compounds 2600 are combined with the background compound 2300 to form a new combination of background compounds. The combination can be in various ways. For example, assuming that the background compound 2300 includes 20 TMCs and the LLM 2100 generates 10 TMCs each time, the top 20 compounds with the best physicochemical parameters among the total 30 TMCs can be combined to form a new combination of background compounds. For another example, the above total 30 TMCs can be provided to the LLM 2100 as a new background compound, and at the same time, the physicochemical parameters of the generated 10 TMCs are provided. Correspondingly, the initial background compound 2300 is provided to the LLM 2100 at the same time as the physicochemical parameters of the initial background compound 2300. In this case, the LLM 2100 can not only obtain positive learning experience from the TMCs with good physicochemical parameters generated, but also obtain negative learning “lessons” from the TMCs with poor physicochemical parameters generated, thereby optimizing the next round of generation.

[0036] In step S1050, the combination of the compound structure material library and the new background compound is provided to the LLM 2100 to iteratively generate the structure of the intended compound. The iteration termination condition can be set according to the actual situation, for example, set to a predetermined number of iterations. For single-objective optimization, a smaller number of iterations can be set, for example, 20 times; for multi-objective, a larger number of iterations can be set, for example, 40 times.

[0037] The compounds involved in the present application include but are not limited to organic compounds, inorganic compounds, metal organic complexes, polymer compounds, small molecule compounds, biological macromolecules, etc., including but not limited to TMC.

[0038] In the task of generating TMC, the compound structure material library 2200 can contain 50 ligands. The 50 ligands are all monodentate ligands, which are mainly composed of carbon, hydrogen, nitrogen, oxygen, phosphorus, sulfur and other elements, and some molecules also contain halogens (such as fluorine, chlorine, bromine, iodine). The molecular weight varies from light single-atom small molecules to medium molecular weight compounds. The charge state is mainly 0 (25) or -1 (25).

[0039] Further, the ligands of the complexes in the compound structure material library 2200, the new background compound 2300, and the candidate compound 2600 can be monodentate ligands, polydentate ligands, etc., the coordination atoms include but are not limited to O, N, S, etc. Donor atoms, the charge state of the ligand includes but is not limited to 0, negative charge, multiple negative charge, etc., the type of ligand can include single ligand, multiple different ligands, etc. In addition, the charge state of the metal atom can be 0, positive charge, multiple positive charge, etc., and the type of ligand atom includes but is not limited to multiple metal atoms such as Pd. Further, the complex form can be a single-core, double-core, or multi-core complex. In the case of a multi-core complex, the complex can exhibit a one-dimensional chain structure, a two-dimensional planar structure, and a three-dimensional stereoscopic structure. In the multi-metal center atom complex, the metal center atoms can be the same or different.

[0040] Further, the compound structure material library 2200, the new background compound 2300, and the candidate compound 2600 are not limited to metal organic complexes or transition metal organic complexes, but also include but are not limited to organic compounds, inorganic compounds, metal organic complexes, polymer compounds, small molecule compounds, biological macromolecules, etc. For example, when generating, first generate a certain number of new compounds using the LLM 2100, then construct a new compound combination, thereby forming a set of new background compounds 2300. In this case, the newly generated compounds can be incorporated into the compound structure material library 2200 to improve the diversity of the generated candidate compounds 2600.

[0041] According to another embodiment of the present application, when generating new compounds, it is possible to allow the generation of candidate compounds 2600 using new compound materials outside of the compound structure material library 2200. These new materials can come from the knowledge base provided by the LLM 2100, or can be directly generated by the LLM 2100.

[0042] For example, in the task of generating TMCs, the compound structure material library 2200 can have a fixed 50 ligands and their complexes. Correspondingly, the background compounds 2300 can be a number of TMCs (e.g., 20 TMCs) from the 1.37M Pd(II) square planar structures [i.e., Pd(II) square planar] TMCs formed by the 50 ligands and central atoms. When generating, first generate 10 new ligands using the LLM 2100, for example, 5 neutral ligands and 5 ionic ligands. Then, use these 10 new ligands to construct new combinations of possible TMCs, thereby forming a new set of background compounds 2300. In this case, the newly generated ligands can be incorporated into the compound structure material library 2200 to improve the diversity of the generated candidate compounds 2600.

[0043] Figure 4 is an example of the compound structure material library 2200, containing but not limited to the specific chemical structures of 50 ligands and their complexes. Figure 4 In the example compound, L x L1-L4 can be the same or different from each other for the same complex. Exemplary, Figure 4 The 50 ligands in are all monodentate ligands, which are mainly composed of carbon, hydrogen, nitrogen, oxygen, phosphorus, sulfur, etc., and some molecules also contain halogens (such as fluorine, chlorine, bromine, iodine). The molecular weight varies from light single-atom small molecules to medium molecular weight compounds. The charge state is 0 (25) or -1 (25). Figure 4 The central metal atom in can be Pd or other metal atoms. Figure 3 is a schematic diagram of a system for generating compound structures using an LLM according to another embodiment of the present application. Figure 5 is a structure diagram of 10 new ligands generated using an LLM.

[0044] As Figure 3As shown, the computer system includes one or more processors, such as one or more central processing units (CPUs) 301 and / or one or more graphics processors (GPUs) 313, and the like, which can perform various appropriate actions and processes according to executable instructions stored in a read-only memory (ROM) 302 or loaded into a random access memory (RAM) 303 from a storage 308. The communication unit 312 can include, but is not limited to, a network card, which can include, but is not limited to, an IB (Infiniband) network card.

[0045] The processor can communicate with the read-only memory 302 and / or the random access memory 303 to execute the executable instructions, be connected to the communication unit 312 through the bus 304, and communicate with other target devices through the communication unit 312, so as to complete the operations corresponding to any one of the methods proposed in the embodiments, for example: providing a compound structure material library and a background compound with known physical and chemical parameters to the LLM; providing a restriction condition and an expected physical and chemical parameter to the LLM; generating a candidate compound meeting the provided physical and chemical parameter under the restriction condition based on the compound structure material library and the physical and chemical parameter by using the LLM; combining the candidate compound with the background compound to form a new background compound combination; and providing the compound structure material library and the new background compound combination to the LLM to iteratively generate a structure of an expected compound.

[0046] In addition, various programs and data required for device operation can also be stored in the RAM 303. The CPU 301, the ROM 302, and the RAM 303 are connected to each other through the bus 304. The ROM 302 is an optional module in the case of the RAM 303. The RAM 303 stores executable instructions or writes executable instructions into the ROM 302 at runtime, and the executable instructions make the processor 301 perform the operations corresponding to the above-mentioned communication method. The input / output interface (I / O interface) 305 is also connected to the bus 304. The communication unit 312 can be integrally arranged or arranged as a plurality of sub-modules (for example, a plurality of IB network cards) and linked on the bus.

[0047] The following components are connected to the I / O interface 305: an input section 306 including a keyboard, a mouse, etc.; an output section 307 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as necessary. A removable medium 311 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 310 as necessary.

[0048] Needless to say, the architecture shown is only one optional implementation, and in the course of specific practice, the number and type of the components described above can be selected, reduced, increased, or replaced as necessary; and in terms of the arrangement of different functional components, a separate arrangement or an integrated arrangement can also be implemented, for example, the GPU and the CPU can be arranged separately or the GPU can be integrated on the CPU, the communication section 312 can be arranged separately or can be integrated on the CPU or the GPU, etc. These alternative implementations all fall within the scope of protection of the present disclosure. Figure 3 Figure 3 Needless to say, the architecture shown is only one optional implementation, and in the course of specific practice, the number and type of the components described above can be selected, reduced, increased, or replaced as necessary; and in terms of the arrangement of different functional components, a separate arrangement or an integrated arrangement can also be implemented, for example, the GPU and the CPU can be arranged separately or the GPU can be integrated on the CPU, the communication section 312 can be arranged separately or can be integrated on the CPU or the GPU, etc. These alternative implementations all fall within the scope of protection of the present disclosure.

[0049] In particular, according to the present application, the process described with reference to the flowchart Figure 1 may be implemented as a computer program product. For example, the present application proposes a computer program product comprising computer readable instructions which, when executed by a processor, implement the following operations: providing a library of compound structure motifs and background compounds with known physicochemical parameters to the LLM; providing constraints and desired physicochemical parameters to the LLM; generating candidate compounds that meet the provided physicochemical parameters under the constraints based on the library of compound structure motifs and the physicochemical parameters using the LLM; combining the candidate compounds with the background compounds into a new combination of background compounds; providing the library of compound structure motifs and the new combination of background compounds to the LLM to iteratively generate the structure of the desired compound.

[0050] In such an implementation, the computer program product can be downloaded and installed from a network by the communication section 309, and / or read and installed from the removable medium 311. When the computer program product is executed by the central processing unit (CPU) 301, the above-described functions defined in the method of the present application are executed.

[0051] ​The technical solutions of the present application can be implemented in many ways. For example, the technical solutions of the present application can be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The order of steps for describing the method is provided only for the purpose of more clearly illustrating the technical solutions. Unless specifically limited, the method steps of the present application are not limited to the above specific description. In addition, in some embodiments, the present application can also be implemented as a storage medium storing a computer program product.

[0052] The embodiments of the present application are further described below in conjunction with specific examples. In the following examples, the LLM used is o1 -preview. Since this model is currently trained more fully for English prompts and performs better, the prompts in the embodiments of the present application are given in the form of English prompts, supplemented by the meanings of the prompts. It should be particularly noted that the numbers given in the examples are only for the purpose of more specifically explaining the method of the embodiments of the present application, and are not for the purpose of limitation.

[0053] Example 1 - TMC dual-target iterative optimization based on a given compound structure material library using LLM

[0054] The user prompt "I have a pool of 50 ligands in a csv file format below." is input. This prompt is used to introduce a compound structure material library 2200 constructed from 50 ligands.

[0055] Subsequently, the ligand data file in csv format is input. Each row of data is a SMILES string, id, charge, connecting atom, and connecting atom index, as follows:

[0056] c1ccccn1, RUCBEY-subgraph-1, 0, N, 1

[0057] CP(C)C, WECJIA-subgraph-3, 0, P, 1

[0058] N#CC, KEYRUB-subgraph-1, 0, N, 1

[0059] [C-]#[N+]c1c(C)cccc1C, NURKEQ-subgraph-2, 0, C, 1

[0060] O, MEBXUN-subgraph-1, 0, O, 1

[0061] n1c(cccc1C)C, BIFMOV-subgraph-1, 0, N, 1

[0062] CP(C)c1ccccc1, CUJYEL-subgraph-2, 0, P, 1

[0063] n1ccc(cc1)C, EZEXEM-subgraph-1, 0, N, 1

[0064] n1cccc(c1)Cl, FOMVUB-subgraph-2, 0, N, 1

[0065] [C-]#[N+]C(C)(C)C, EFIHEJ-subgraph-3, 0, C, 1

[0066] CN1[C]N(C)C=C1, LETTEL-subgraph-1, 0, C, 2

[0067] n1ccn(c1)C, KAKKIR-subgraph-3, 0, N, 1

[0068] [C-]#[N+]C1CCCCC1, BICRIQ-subgraph-3, 0, C, 1

[0069] S(=O)(C)C, UPEGAZ-subgraph-2, 0, S, 1

[0070] O1CCNCC1, CEVJAP-subgraph-2, 0, N, 1

[0071] n1[nH]c(cc1C)C, BABTUT-subgraph-3, 0, N, 1

[0072] C(C)NCC, ZEJJEF-subgraph-3, 0, N, 1

[0073] n1ccc(cc1)N(C)C, KULGAZ-subgraph-2, 0, N, 1

[0074] [C-]#[O+], CIGDAA-subgraph-1, 0, C, 1

[0075] c1ccnc(c1)N, HOVMIP-subgraph-3, 0, N, 1

[0076] N, ULUSIE-subgraph-1, 0, N, 1

[0077] S(C)C, IBEKUV-subgraph-1, 0, S, 1

[0078] CS(=O)C, REQSUD-subgraph-2, 0, O, 1

[0079] NCC, BOSJIF-subgraph-1, 0, N, 1

[0080] n1c(cc(cc1C)C)C, GUVMEP-subgraph-0, 0, N, 1

[0081] [Cl-], MAZJIJ-subgraph-0, -1, Cl, 1

[0082] [Br-], OBONEA-subgraph-1, -1, Br, 1

[0083] [I-], CORTOU-subgraph-2, -1, I, 1

[0084] [CH3-], LEVGUO-subgraph-2, -1, C, 1

[0085] [C-]1=C(F)C(=C(C(=C1F)F)F)F, REBWEB-subgraph-2, -1, C, 1

[0086] C1=C[C-]=CC=C1, DOGPAS-subgraph-1, -1, C, 3

[0087] O=N(=O)[O-], IJIMIX-subgraph-1, -1, O, 1

[0088] [N-]=[N+]=[N-], PEJGAN-subgraph-1, -1, N, 1

[0089] S1(=O)(=O)[N-]C(=O)c2c1cccc2, BIFZEX-subgraph-0, -1, N, 1

[0090] [C-]#N, IRIXUC-subgraph-3, -1, C, 1

[0091] [S-]C#N, SAYGOO-subgraph-0, -1, S, 1

[0092] [F-], UROGIS-subgraph-1, -1, F, 1

[0093] [C-] (F) (F) F, MAQKEX-subgraph-1, -1, C, 1

[0094] O=N[O-], LUQWUQ-subgraph-1, -1, N, 1

[0095] C[C-]=0, QAYDID-subgraph-2, -1, C, 2

[0096] C1CC(=0)[N-]C1=0, MOYDOV-subgraph-3, -1, N, 1

[0097] [C-]1=CC=C(C=C1)F, NIZQUK-subgraph-1, -1, C, 1

[0098] [S-]C#N, SAYHIJ-subgraph-1, -1, N, 1

[0099] O=[C-]OC, CIQGOY-subgraph-0, -1, C, 1

[0100] [S-]c1ccccc1, VUFZUT-subgraph-1, -1, S, 1

[0101] [O-]c1ccccc1, ZOQFIU-subgraph-0, -1, O, 1

[0102] [S-]c1c(c(cc(c1F)F)F)F, GUQBUQ-subgraph-0, -1, S, 1

[0103] [C-]1=CC=C(C=C1)C, LEZYUM-subgraph-2, -1, C, 1

[0104] [N-]1C(=0)c2c(C1=0)cccc2, RAJXUX-subgraph-2, -1, N, 1

[0105] c1c(C#[C-])cccc1, QEWZOH-subgraph-3, -1, C, 4

[0106] Subsequently, the input generation task, “I am interested in making a Pd(II) square planer transition metal complex (TMC),” was entered, instructing the LLM to generate a Pd(II) square planar TMC.

[0107] Along with the input of the task, the constraints and the expected physicochemical parameters are also inputted.

[0108] The prompt words for the expected physicochemical parameters (or optimization objectives) are as follows:

[0109] The optimization objective of “Design objective: 1. find new TMCs with both larger HOMO-LUMO gap and larger polarizability (these two values will be normalized and evaluated together) than the given TMCs presented later on.” aims to perform double-objective optimization while pursuing a larger HOMO-LUMO energy gap and a higher polarizability, wherein the HOMO-LUMO energy gap and the polarizability of the generated TMCs are required to be higher than the given TMCs presented later.

[0110] The prompt words for the constraints are as follows:

[0111] “Constraints:

[0112] The total charge of the TMC should be-1, 0, or 1.

[0113] All ligands in the TMC need to be those present in this csv file provided above.”

[0114] The above constraints limit the total charge of the generated TMC to be -1, 0, or 1. In addition, the above constraints also limit that all ligands in the TMC should be those covered in the csv file provided to the LLM.

[0115] Then, the background compounds (in this example, TMCs) with known physicochemical parameters are provided. The prompt words are as follows:

[0116] "Here is some TMCs their measures of total charge, polarisability, and HOMO-LUMO gap. They are provided in a format of {TMC, total charge, polarisability, HOMO-LUMO gap}. The TMC should be in a format of Pd_$L1_$L2_$L3_$L4, where Pd is the metal center, and $L1, $L2, $L3, and $L4 are the id of the ligands (listed in the csv file) and follow a clockwise ordering. Note that the TMC has cyclic symmetry for the ligands, so that Pd_$L1_$L2_$L3_$L4, Pd_$L2_$L3_$L4_$L1, Pd_$L3_$L4_$L1_$L2, and Pd_$L4_$L1_$L2_$L3 are the same TMC. Below are the TMCs and their ground-truth total charge, polarizability, and HOMO-LUMO gap."

[0117] The above prompt words explain the data format of the TMCs to be provided. Subsequently, the TMC data is provided:

[0118] {Pd_KEYRUB-subgraph-1_MEBXUN-subgraph-1_BIFMOV-subgraph-1_RAJXUX-subgraph-2, 1, 250.599, 2.465}

[0119] {Pd_MOYDOV-subgraph-3_ULUSIE-subgraph-1_NURKEQ-subgraph-2_NURKEQ-subgraph-2, 1, 307.955, 2.784}

[0120] {Pd_MEBXUN-subgraph-1_BABTUT-subgraph-3_BOSJIF-subgraph-1_REBWEB-subgraph-2, 1, 217.567, 2.94}

[0121] {Pd_RUCBEY-subgraph-1_ZEJJEF-subgraph-3_EFIHEJ-subgraph-3_DOGPAS-subgraph-1, 1, 282.377, 2.245}

[0122] {Pd_KAKKIR-subgraph-3_QAYDID-subgraph-2_IRIXUC-subgraph-3_GUQBUQ-subgraph-0, -1, 229.135, 1.674}

[0123] {Pd_SAYHIJ-subgraph-1_LUQWUQ-subgraph-1_SAYGOO-subgraph-0_REQSUD-subgraph-2, -1, 178.701, 2.251}

[0124] {Pd_REQSUD-subgraph-2_MAQKEX-subgraph-1_OBONEA-subgraph-1_LUQWUQ-subgraph-1, -1, 146.398, 2.21}

[0125] {Pd_CIQGOY-subgraph-0_BOSJIF-subgraph-1_UROGIS-subgraph-1_NURKEQ-subgraph-2, 0, 209.993, 2.11}

[0126] {Pd_MAQKEX-subgraph-1_DOGPAS-subgraph-1_KULGAZ-subgraph-2_ZEJJEF-subgraph-3, 0, 272.928, 2.306}

[0127] {Pd_ZOQFIU-subgraph-0_HOVMIP-subgraph-3_NURKEQ-subgraph-2_MEBXUN-subgraph-1, 1, 285.01, 1.786}

[0128] Finally, some additional information can be provided to the LLM, such as the number of TMCs that need to be generated, the generation approach suggestion, and the data format to limit the output. The LLM can be asked to generate 5 new TMCs, and it can be suggested that the LLM utilize crossover and / or mutation in genetic algorithms to generate the TMCs. Additionally, the LLM can be asked not to generate duplicate TMCs. The prompt words are as follows:

[0129] "Please propose 5 new TMCs that achieve my design objective. You can make ligand crossover (i.e., swap ligands in the TMCs) or ligand mutations (i.e., substitute ligands in TMCs with some other ligands in the pool) based on your chemistry knowledge to achieve the design objective. You can also directly propose new TMCs by taking advantage of your chemistry knowledge to propose Pd TMCs with maximize HOMO-LUMO gap and maximize polarizability. Be sure to use the id of the provided pool of 50 ligands and control the total charge of the TMC to be -1, 0, or 1. In addition, please DO NOT propose duplicated TMCs that I have listed above.

[0130] Your output should follow the format: {<< <explanation>>>: $EXPLANATION, << <tmc>>>:[$TMC],<<<TOTAL_CHARGE>>>:${total charge},<< <polarisability>>>:$POLARISABILITY,<<<HOMO-LUMO gap>>>:${HOMO-LUMO gap}}.Here are therequirements that you should fulfill:

[0131] 1.$EXPLANATION should be your analysis of why the new TMC wouldachieve my design objective.

[0132] 2.The$TMC should be in a format of[Pd_$L1_$L2_$L3_$L4],where Pd isthe center metal.$L1,$L2,$L3,and$L4 should be the id of the ligands(listed inthe csv file)and follow a clockwise ordering.

[0133] 3$POLARISABILITY and${HOMO-LUMO gap}are your predictions of thepolarisability and HOMO-LUMO gap for$TMC based on your chemistry knowledgeand provided data.”

[0134] After that, the LLM will generate the candidate compounds according to the requirements. The following gives the form of the candidate compounds generated by the LLM:

[0135] {<< <explanation>This TMC combines the large, aromatic BIFMOV-subgraph-1 ligand with the strongly electron-withdrawing EFIHEJ-subgraph-3 ligand. The combination should increase polarizability due to the extended π-system, while the electron-withdrawing groups should widen the HOMO-LUMO gap. The neutral MEBXUN-subgraph-1 (water) ligand balances the charge. <tmc>>>:[Pd_BIFMOV-subgraph-1_EFIHEJ-subgraph-3_BIFMOV-subgraph-1_MEBXUN-subgraph-1],<<<TOTAL_CHARGE>>>:1,<< <polarisability>>>:310.5,<<<HOMO-LUMO gap>>>:2.95}

[0136] Then, the above generated TMCs can be combined with the background compounds initially provided to the LLM, and iteratively generated according to the above procedure.

[0137] Table 1 is an example of complexes 1-10 generated by the LLM model according to the dual optimization target of polarity and HOMO-LUMO energy gap. The newly generated complexes 1-10 are the top 10 complexes in the product of the energy gap and the polarity in the complex structure generated by multiple iterations, i.e., the complexes 1-10 have both a large polarity and a relatively large HOMO-LUMO energy gap.

[0138]

[0139]

[0140]

[0141]

[0142]

[0143]

[0144] Example 2 - Single-target TMC iterative optimization with new ligands using LLM

[0145] The input user prompt "I have a pool of 50 ligands in a csv file format below." is used to introduce the chemical structure material library 2200 constructed from 50 ligands. It should be noted that in the process of subsequent iterations, newly generated ligands will be added to the chemical structure material library 2200. For example, in the process of the first round of generation, if m new ligands are generated, then in the process of subsequent iterations, the LLM will be provided with a chemical structure material library constructed from 50+m ligands.

[0146] Subsequently, the ligand data file in csv format is input. Each row of data is a SMILES string, id, charge, connection atom, and connection atom number, and the prompt is similar to the format given in Example 1, which will not be repeated here.

[0147] Subsequently, an input generation task, "I am interested in making a Pd(II) square planer transition metal complex (TMC)", is inputted, thereby instructing the LLM to generate a Pd(II) square planar TMC.

[0148] Along with the input of the generation task, a constraint condition and an expected physicochemical parameter are inputted, with the prompt word as follows:

[0149] "My design objective is to maximize its polarisability while making the total charge of the TMC to be -1, 0 or 1." The above optimization objective aims to perform single-objective optimization, only maximizing the polarizability; the constraint condition is to ensure that the total charge of the generated TMC is -1, 0 or 1.

[0150] Then, a background compound (in this example, a TMC) with known physicochemical parameters is provided. The prompt word is as follows:

[0151] "I have made 100 TMCs and measured their total charge and polarisability. They are provided in a format of {$TMC, ${total charge}, ${polarisability}, {error message (if any)}}.

[0152] The TMC should be in a format of Pd_$L1_$L2_$L3_$L4, where Pd is the metal center, $L1, $L2, $L3, and $L4 are the id of the ligands (listed in the csvfile) and follow a clockwise ordering. Note that the TMC has cyclic symmetry for the ligands so that Pd_$L1_$L2_$L3_$L4, Pd_$L2_$L3_$L4_$L1, Pd_$L3_$L4_$L1_$L2, and Pd_$L4_$L1_$L2_$L3 are the same TMC. Below are the TMCs and their ground-truth total charge and polarisability.

[0153] The above prompt explains the data format of the TMCs that will be provided. The prompt that accompanies the provision of the TMC data is similar to the format given in Example 1 and is not repeated here. Similarly, in the process of the first round of generation, if n new TMCs are generated, then in the process of subsequent iterative optimization, the n newly generated TMCs will be added to the background compounds. At this time, the TMCs will have 100 + n.

[0154] Subsequently, the LLM can be asked in natural language to construct TMCs using new ligands beyond the initial 50 ligands provided, with the following prompt:

[0155] "Grounded on your chemistry knowledge, look at the pattern of the provided data and think about what makes a ligand combination give extremely large polarisability for a TMC. Then based on your chemistry knowledge and 100 example TMCs provided, can you give me TEN new TMCs that would help me further maximize the polarisability? I am expecting that the new proposed TMC has a larger polarisability than the largest in given TMCs. For the TEN proposed TMCs, you must design new ligands to achieve the objective."

[0156] Finally, some additional information can be provided to the LLM, which can be given according to the generation purpose. For example, the number of TMCs to be generated, the generation method suggestion and the data format of the output limit. For example, the LLM can be required to generate 5 or 10 new TMCs. For another example, the LLM can be suggested to use crossover and / or mutation in genetic algorithm to generate TMC. In addition, the LLM can also be required not to generate duplicate TMCs.

[0157] Then, the new prompt words can be constructed using the above generated chemical structure material library containing new ligands and TMCs containing new ligands, and the iteration generation is carried out according to the above process.

[0158] Table 2 is an example of complexes 1-10 generated by optimizing the polarity optimization target through the LLM model. The newly generated complexes 1-10 all show excellent polarity.

[0159]

[0160]

[0161]

[0162]

[0163]

[0164] Example 3 - Superior performance of LLM in generating compounds

[0165] LLM in generating compounds has significant advantages over other algorithms. To show the advantages of LLM in generating compounds over other algorithms, Figure 6 The effect of generating compounds using a single-step method is compared with other methods of generating compound structures.

[0166] In the process of generating compounds using LLM, a non-iterative method (i.e., a single-step method) can also be used to generate compounds. For example, based on steps similar to those given in Example 1, the difference is that after the LLM generates a candidate compound each time, the candidate compound is no longer fed back to the LLM for iterative optimization. LLM can be used to generate 10 compounds each time, and after repeating 20 times, the top 20 compounds out of the 200 generated compounds are analyzed and compared. Figure 6 It is shown that even with this single-step method, LLM in generating compounds exhibits superior performance compared to other algorithms.

[0167] Figure 6 The a subgraph of shows the HOMO-LUMO energy gap distribution of the top 20 TMCs out of the 200 TMCs generated by each method. Among them, blue represents the random algorithm, red represents the genetic algorithm (GA), orange represents claude-3.5-sonnet, green represents o1-preview, purple represents o1-mini, and sky blue represents gpt-4o. These results reflect the superior performance of LLM in the TMC generation task with few samples. Figure 6 The b subgraph of shows the HOMO-LUMO energy gap cumulative probability of all TMCs proposed by various methods and their overall distribution. Figure 6 The c subgraph of shows the proportion of valid (in the space of 1.37M TMCs) and unique (i.e., not repeated) TMCs out of the 200 generated TMCs. Figure 6 The d-subplot shows the HOMO-LUMO gap distribution of the top 20 TMCs out of 200 generated TMCs when different numbers of known TMCs are provided in the prompt words. Only the results of claude-3.5-sonnet (orange) and o1 -preview (green) are shown in the plot. In addition, the TMCs with the largest HOMO-LUMO gap derived from claude-3.5-sonnet and o1 -preview are shown, respectively. The average HOMO-LUMO gap of the top 20 TMCs derived from the random algorithm is represented by a blue dashed line. It is worth noting that o1 -preview exhibits better performance than the random algorithm using only one initial TMC sample. Increasing the number of initial TMCs generally improves the HOMO-LUMO gap of the TMCs generated by the model (although there is a marginal effect), which indicates that LLMs can achieve satisfactory results with limited initial data, providing researchers with a balance between data requirements and performance. Figure 6 The colors of different atoms in the d-subplot of FIG. 6 are as follows: palladium (Pd) is sky blue, carbon (C) is gray, nitrogen (N) is blue, oxygen (O) is red, phosphorus (P) is purple, fluorine (F) is green, and hydrogen (H) is white. The purple dashed line in the TMC represents the coordination bond between Pd(II) and the ligand.

[0168] As another example, Figure 7 shows the analysis of the top 20 TMCs out of 200 generated candidate TMCs under the method of iterative optimization. o1 -preview identifies better TMCs faster than the genetic algorithm. In Figure 7 In the figure, the random algorithm is represented in blue, the genetic algorithm in orange, and the o1 -preview method in green and grey. The green data represents the case where, in each iteration, only the 20 TMCs with the largest HOMO-LUMO gap from the previous round of candidates and the known TMCs are used in the next round of iteration. The grey data represents the case where, in each iteration, the full history of the generated TMCs (including the physico-chemical parameters) is kept and used in the next iteration. While the two methods of o1 -preview do not show much difference in the early iterations (iteration number 5), the divergence between the performances becomes apparent as the optimization progresses. Keeping all the historical data allows the model to generate TMCs with higher HOMO-LUMO gap. The above description is only an implementation of the present application and an explanation of the principles of the technology used. Those skilled in the art will understand that the scope of protection of the present application is not limited to the technical solutions formed by the specific combination of the technical features described above, and should also cover other technical solutions formed by any combination of the above technical features or equivalent features, without departing from the technical concept. For example, the above features can be replaced with technical features disclosed in the present application (but not limited to) with similar functions to form technical solutions.< / polarisability> < / tmc> < / explanation> < / polarisability> < / tmc> < / explanation>

Claims

1. A method for generating a structure of a compound using a large language model, comprising: providing a library of compound structures and background compounds with known physical-chemical parameters to the large language model; providing constraints and desired physical-chemical parameters to the large language model; generating candidate compounds that meet the provided physical-chemical parameters under the constraints using the large language model based on the library of compound structures and the physical-chemical parameters; combining the candidate compounds with the background compounds into a new set of background compounds; providing the library of compound structures and the new set of background compounds to the large language model to iteratively generate a structure of a desired compound.

2. The method of claim 1, wherein: when providing the background compounds with known physical-chemical parameters to the large language model, the physical-chemical parameters of the background compounds are also provided to the large language model; when providing the library of compound structures and the new set of background compounds to the large language model, the physical-chemical parameters of the new set of background compounds are also provided to the large language model.

3. The method of claim 1, wherein, the desired physical-chemical parameters comprise at least two different physical-chemical parameters.

4. The method of claim 1, wherein, the compounds comprise organic compounds, inorganic compounds, metal-organic complexes, polymeric compounds, small molecule compounds, biological macromolecules.

5. The method of claim 1, wherein, combining the candidate compounds with the background compounds into a new set of background compounds comprises: combining the top K compounds in the candidate compounds and the background compounds with the best physical-chemical parameters into the new set of background compounds.

6. The method of claim 1, wherein, generating candidate compounds that meet the provided physical-chemical parameters under the constraints using the large language model based on the library of compound structures and the physical-chemical parameters comprises: generating the candidate compounds based on new compound materials outside the library of compound structures.

7. The method of claim 6, wherein, after generating the candidate compounds based on new compound materials outside the library of compound structures, the method further comprises integrating the new compound materials into the library of compound structures.

8. The method of claim 1, wherein, the compounds are transition metal complexes, the library of compound structures is a library of ligands, and the physical-chemical parameters comprise HOMO-LUMO energy gap and polarizability.

9. The method of claim 8, wherein, the constraints comprise the charge of transition metal complexes, metal atoms, and / or ligands.

10. A system for generating a compound structure using a large language model, comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program comprises the following steps: the processor executes the computer program to implement the following steps: providing a library of compound structures and background compounds with known physical-chemical parameters to the large language model; providing constraints and desired physical-chemical parameters to the large language model; generating candidate compounds that meet the provided physical-chemical parameters under the constraints using the large language model based on the library of compound structures and the physical-chemical parameters; combining the candidate compounds with the background compounds into a new set of background compounds; providing the library of compound structures and the new set of background compounds to the large language model to iteratively generate a structure of a desired compound.

11. A computer program product comprising computer program instructions which, when executed by a processor, implement the following steps: providing a library of compound structure building blocks and background compounds with known physicochemical parameters to the large language model; providing constraints and desired physicochemical parameters to the large language model; generating candidate compounds that meet the provided physicochemical parameters under the constraints using the large language model based on the library of compound structure building blocks and the physicochemical parameters; combining the candidate compounds with the background compounds into a new set of background compounds; providing the library of compound structure building blocks and the new set of background compounds to the large language model to iteratively generate the structure of the desired compound.

Images