A method for intelligently predicting unknown chemical reactions based on known reactions

Abstract

This invention discloses a method for intelligently predicting unknown chemical reactions based on known reactions, comprising: a) establishing a database of known reactions; and b) predicting unknown chemical reactions. The method of this invention can predict whether multiple known compounds can react, and can predict the structures of possible products when a reaction occurs. The entire prediction process has low dependence on human intervention and a high degree of intelligence, achieving low-labor-cost, efficient, and highly accurate intelligent prediction of unknown chemical reactions based on known reactions. Furthermore, the reaction database established in this invention is universal and applicable to newly discovered chemical reactions, and is easy to maintain and update. Therefore, this invention represents a significant advancement and has higher application value compared to existing technologies.

Description

Technical Field

[0001] This invention relates to a method for intelligently predicting unknown chemical reactions based on known reactions, and belongs to the field of big data processing technology. Background Technology

[0002] In the pharmaceutical and other chemical engineering fields, the research and development of drugs or products requires the synthesis of various organic molecules. Traditionally, researchers predict unknown reactions by consulting chemical reaction databases and combining this with their own chemical synthesis knowledge and experience, then verify the predictions through experiments. The accuracy of these predictions heavily relies on the researchers' comprehensive grasp of synthetic knowledge and their ability to apply it flexibly, significantly raising the barrier to entry and impacting research efficiency. Therefore, providing a method for intelligently predicting chemical reactions using big data and cheminformatics technologies would have significant application value.

[0003] Currently, the basic principle of mainstream organic synthesis route prediction software is to extract the parts of the reaction that change (called reaction centers) based on a large amount of chemical reaction data, and then match the molecule to be synthesized with the reaction centers, working backward step by step until the available chemical raw materials are available, thereby obtaining the complete synthetic route for synthesizing the target molecule. Therefore, current computer-aided synthesis technologies are all based on retrosynthesis, deducing reaction routes. However, retrosynthesis, or reverse derivation, is based on the limited number of major products, while forward derivation of all possible reactions significantly increases the computational difficulty and workload.

[0004] Chinese patent application CN202011302735.1 discloses a method for predicting chemical reaction products, including: a) establishing a database containing chemical reaction literature, calculating all possible single-step reactions between starting molecules based on the chemical reactions recorded in the literature, and judging the rationality and accuracy of the chemical structure of the reaction products to obtain the screened products; b) performing multiple rounds of calculations on the screened products obtained in step a) with all the starting compounds to obtain all possible reaction products generated in the system during each single-step chemical reaction, and finally screening the product structure at each molecular weight to find the most likely structure. While this prediction method can achieve positive prediction of reaction products based on reactants, it faces challenges due to the potential existence of multiple similar potential reaction centers in the target reactant. Many factors influence which center is active in the prediction scenario, such as the varying degrees of difficulty in participating in the reaction caused by differences in the surrounding three-dimensional space, and the enhancement or weakening of the activity of potential reaction centers by electron-withdrawing, electron-donating, or resonance effects in the surrounding environment. Therefore, this prediction method requires multiple rounds of calculations between the screened products and all starting compounds, and mass spectrometry analysis of the product structures at each molecular weight for thorough screening to ultimately identify the most probable structure. Clearly, this prediction method is extremely complex, and its accuracy heavily relies on the depth and breadth of knowledge about chemical reactions. Furthermore, the difficulty of subsequent maintenance and updates will increase significantly, and it cannot yet effectively achieve low-cost, efficient, and highly accurate intelligent prediction of unknown chemical reactions. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, the purpose of this invention is to provide a method for intelligently predicting unknown chemical reactions based on known reactions, which can achieve low labor costs, high efficiency, and high accuracy.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0007] A method for intelligently predicting unknown chemical reactions based on known reactions includes the following steps:

[0008] a) Establish a database of known reactions

[0009] Collect information on known chemical reactions and use codes to represent the reaction characteristics of each reactant in each known reaction and the reaction rules of each known reaction;

[0010] b) Predicting unknown chemical reactions

[0011] b1) Use each target reactant to sequentially search the known reaction database established in step a) to obtain the set of reaction rules corresponding to each target reactant;

[0012] b2) Determine whether there is any intersection between the sets of reaction rules corresponding to all target reactants to be predicted;

[0013] b3) If the intersection is determined to be empty, it means that according to the current known reaction database, the target reactants to be predicted cannot react with each other, so exit intelligent prediction; if the intersection is determined to be not empty, proceed to step b4).

[0014] b4) Sequentially apply the reaction rules from the intersection set, and in conjunction with each target reactant, search the known reaction database established in step a) to obtain the reaction rules F for each target reactant. i The reaction instances involved and the reaction characteristics of each target reactant in the reaction instances it participates in; where i is a natural number greater than or equal to 1;

[0015] b5) If the linear combination of the reaction characteristics of all target reactants to be predicted in the reaction instances in which they participate is not equal to reaction rule F i Then it is determined that the target reactants to be predicted cannot undergo a reaction with reaction rule F. i The reaction; if the linear combination of the reaction characteristics of all target reactants in the reaction instances in which they participate equals reaction rule F. i Then it is determined whether the target reactants to be predicted can undergo a reaction with reaction rule F. i The reaction;

[0016] b6) Based on the product structure fragments corresponding to each target reactant in its respective reaction instance, predict the structure of the target reaction product.

[0017] In one embodiment, in step a), the reaction characteristic of each reactant is a linear combination of the reaction characteristics of each reactant at each point, and the reaction characteristics at each point are sorted in ascending order by strings and separated by the character '`'.

[0018] In a further implementation scheme, the code format for the reaction characteristics of each site is as follows: element symbol of the reaction site `T: type of reaction site`environmental information of the reaction site`generative bond information of the reaction site.

[0019] In a further implementation scheme, if the element of the reaction site is fluorine, chlorine, bromine, iodine or astatine, the character 'X' is used as the element symbol of the reaction site; otherwise, the element symbol of the reaction site itself is used as the element symbol of the corresponding reaction site.

[0020] A further implementation scheme categorizes reaction sites into four types based on whether the adjacent atoms and bonds at the reaction sites change after the reaction, denoted by the numbers 1, 2, 3, and 4, as follows:

[0021] 1: This indicates that after the reaction, the original adjacent atoms at the reaction site were not removed, but new adjacent atoms were added, and the bond type of the original adjacent bonds changed.

[0022] 2: This indicates that after the reaction, the original adjacent atoms at the reaction site were not removed, but new adjacent atoms were added, and the bond type of the original adjacent bonds remained unchanged.

[0023] 3: This indicates that after the reaction, the original adjacent atoms at the reaction site are removed, new adjacent atoms are added, and the bond types of the remaining adjacent bonds change.

[0024] 4: This indicates that after the reaction, the original adjacent atoms at the reaction site were removed and new adjacent atoms were added, while the bond type of the remaining adjacent bonds remained unchanged.

[0025] A further implementation scheme uses numbers to represent various bond types, specifically: 1 represents a single bond, 2 represents a double bond, 3 represents a triple bond, and 1.5 represents an aromatic bond.

[0026] In a further implementation scheme, if the original adjacent atoms of the reaction site are not removed after the reaction, the information of double bonds, triple bonds, and aromatic bonds contained in the adjacent bonds of the reaction site before the reaction is used as the environmental information of the reaction site, and the code format is: "E: type of adjacent bond to be output"; if multiple bond types are included, they are output in ascending order of bond type number; if no double bonds, triple bonds, or aromatic bonds are included, the environmental information of the reaction site is expressed by an empty string "".

[0027] In a further implementation scheme, if the original adjacent atoms of the reaction site are removed after the reaction, the information of the removed atoms at the reaction site is used as the environmental information of the reaction site, and the code format is: "D: Element symbol of the removed atom_Bond type between the reaction site and the removed atom".

[0028] A further implementation scheme uses the bond type information between the reaction site and the newly added adjacent atom after the reaction as the generated bond information of the reaction site. The code format is: "Nbd: bond type between the reaction site and the newly added adjacent atom". If there are multiple bond types, they are output in ascending order of bond type number.

[0029] In one implementation, in step a), the reaction rule for each known reaction refers to a linear combination of the reaction characteristics of all reactants participating in the reaction, sorted in ascending order by strings of reaction characteristics and separated by the character '|'.

[0030] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0031] The method described in this invention has low dependence on human intervention and a high degree of intelligence. It can achieve intelligent prediction of unknown chemical reactions based on known reactions with low labor costs, high efficiency, and high accuracy. Furthermore, the reaction database described in this invention is universal and can be applied to newly discovered chemical reactions in the future, and is easy to maintain and update in the future. Therefore, this invention has significant progress and higher application value compared with the prior art. Detailed Implementation

[0032] The technical solution of the present invention will be further described in detail and completely below with reference to the embodiments.

[0033] Example 1

[0034] A method for predicting whether two known compounds q1 and q2 can react and the possible structures of the products includes the following steps:

[0035] a) Establish a database of known reactions

[0036] Collect information on known chemical reactions and use codes to represent the reaction characteristics of each reactant in each known reaction and the reaction rules of each known reaction;

[0037] The reaction characteristics of each reactant are a linear combination of the reaction characteristics of each reactant at each point, and the reaction characteristics at each point are sorted in ascending order by strings and separated by the character '`'.

[0038] The code format for each site's reaction characteristics is: element symbol of the reaction site `T: type of reaction site `environmental information of the reaction site `formative bond information of the reaction site;

[0039] In this invention:

[0040] The reaction site refers to the molecular structure of the reactant, where the atom at that site will enter the product structure after the reaction, and the atoms and / or adjacent bonds of the site will change.

[0041] If the element at the reaction site is fluorine, chlorine, bromine, iodine or astatine, the character 'X' is used as the element symbol for that reaction site; otherwise, the element symbol of the reaction site itself is used as the element symbol for the corresponding reaction site.

[0042] Based on whether the adjacent atoms and bonds at the reaction site change after the reaction, reaction sites are classified into four types, represented by the numbers 1, 2, 3, and 4, as follows:

[0043] 1: This indicates that after the reaction, the original adjacent atoms at the reaction site were not removed, but new adjacent atoms were added, and the bond type of the original adjacent bonds changed; as shown in site A in the following example:

[0044]

[0045] 2: This indicates that after the reaction, the original adjacent atoms at the reaction site were not removed, but new adjacent atoms were added, and the bond type of the original adjacent bonds remained unchanged; such as site B in the following example:

[0046]

[0047] 3: This indicates that after the reaction, the original adjacent atoms at the reaction site are removed, new adjacent atoms are added, and the bond types of the remaining adjacent bonds change; as shown in the C site in the following example:

[0048]

[0049] 4: This indicates that after the reaction, the original adjacent atoms at the reaction site are removed and new adjacent atoms are added, while the bond types of the remaining adjacent bonds remain unchanged; as shown in the D site in the following example:

[0050]

[0051] Furthermore, the present invention uses numbers to represent various bond types, specifically: 1 represents a single bond, 2 represents a double bond, 3 represents a triple bond, and 1.5 represents an aromatic bond;

[0052] Furthermore, if the original adjacent atoms of the reaction site are not removed after the reaction, the information of double bonds, triple bonds, and aromatic bonds contained in the adjacent bonds of the reaction site before the reaction is used as the environmental information of the reaction site. The code format is: "E: type of adjacent bond to be output", for example, the environmental information of site A in the above example is: "E: 2"; if multiple bond types are included, they are output in ascending order of bond type number; if double bonds, triple bonds, and aromatic bonds are not included, the environmental information of the reaction site is expressed by an empty string "", for example, the environmental information of site B in the above example is empty, represented by an empty string ""; if the original adjacent atoms of the reaction site are removed after the reaction, the information of the removed atoms of the reaction site is used as the environmental information of the reaction site. The code format is: "D: element symbol of removed atom_bond type between reaction site and removed atom", for example, the environmental information of site C in the above example is: "D: O_2";

[0053] Furthermore, the bond type information between the reaction site and the newly added adjacent atom after the reaction is used as the generated bond information of the reaction site. The code format is: "Nbd: bond type between the reaction site and the newly added adjacent atom". For example, the generated bond information of site A in the above example is: "Nbd:1". If there are multiple bond types, they are output in ascending order of bond type number.

[0054] In step a), the reaction rule for each known reaction refers to a linear combination of the reaction characteristics of all reactants participating in the reaction, sorted in ascending order by the strings of reaction characteristics and separated by the character '|'.

[0055] b) Predicting unknown chemical reactions

[0056] b1) Using target reactant q1 Sequentially search the known reaction database established in step a) to obtain the set of reaction rules corresponding to target reactants q1 and q2 respectively;

[0057] b2) Determine whether there is an intersection between the sets of reaction rules corresponding to the target reactants q1 and q2 to be predicted;

[0058] b3) If the intersection is determined to be empty, it means that according to the current known reaction database, the target reactants q1 and q2 cannot react, and the intelligent prediction is exited; if the intersection is determined to be not empty, such as when reaction rule F exists in the intersection of the reaction rules corresponding to q1 and q2. i “C`T:4`D:O_1`Nbd:1|O`T:1`E:2`Nbd:1”, then proceed to step b4);

[0059] b4) Using this reaction rule F from the intersection. i Search the known reaction database established in step a) with compounds q1 and q2 (as reactants) respectively to obtain reaction examples in which compounds q1 and q2 are reactants and participate under reaction rule Fi. For example, the reaction examples in which compound q1 is reactant and participate under reaction rule Fi are as follows:

[0060]

[0061] (Reaxys Reaction ID:5203266, Source: Pesticide Science, 1997, 50(3), 258–274); and, the reaction characteristics of compound q1 as a reactant in the above-mentioned reaction examples it participates in are: "C`T:4`D:O_1`Nbd:1", and the corresponding product fragment is the part within the solid box; and the reaction examples in which compound q2 is used as a reactant and participates in the reaction according to reaction rule Fi are as follows:

[0062]

[0063] (Reaxys Reaction ID:37087, Source: Synthetic Communications, 1988, 18(4), 1625–1636); and, the reaction characteristics of compound q2 as a reactant in the above reaction examples it participates in are: “O`T:1`E:2`Nbd:1”, and the corresponding product fragment is the part inside the dashed box;

[0064] b5) Since the linear combination of the reaction characteristics of compounds q1 and q2 as reactants in the reaction example with reaction rule Fi is: "C`T:4`D:O_1`Nbd:1|O`T:1`E:2`Nbd:1", which is exactly equal to reaction rule Fi, it can be predicted that compounds q1 and q2, as reactants, can undergo a reaction with reaction rule F. i The reaction;

[0065] b6) Based on the product structure fragments corresponding to compounds q1 and q2 as reactants in their respective reaction examples, and considering the respective reaction characteristics of compounds q1 and q2, it can be known that each product fragment has one reaction site, and the formation bonds at both reaction sites are single bonds, which is consistent with each other. Therefore, by connecting the two reaction sites in the product fragments with single bonds, the product structure of the predicted reaction can be obtained. The specific predicted reaction is shown below:

[0066]

[0067] The reaction predicted by the method of this invention has been verified in the recent literature "Advanced Synthesis and Catalysis, 2019, 361(1), 185–191, Reaxys Reaction ID: 50362153", which shows that the method of this invention is reliable and can achieve intelligent prediction of unknown chemical reactions based on known reactions with low labor cost, high efficiency and high accuracy.

[0068] Example 2

[0069] Predicting target reactant q3 With q4 Can they react and what are the structures of the possible reaction products?

[0070] a) Establish a database of known reactions

[0071] The specific details are the same as described in Example 1;

[0072] b) Predicting unknown chemical reactions

[0073] b1) Use the target reactants q3 and q4 to sequentially search the known reaction database established in step a) and obtain the set of reaction rules corresponding to each of the target reactants q3 and q4 respectively;

[0074] b2) Determine whether there is any intersection between the sets of reaction rules corresponding to the target reactants q3 and q4 to be predicted;

[0075] b3) If the intersection is determined to be empty, it means that according to the current known reaction database, the target reactants q3 and q4 cannot react, and the intelligent prediction is exited; if the intersection is determined to be not empty, such as when reaction rule F exists in the intersection of the reaction rules corresponding to q3 and q4. i “C`T:4`D:X_1`Nbd:1|O`T:4`Nbd:1”, then proceed to step b4);

[0076] b4) Using this reaction rule F from the intersection. i Search the known reaction database established in step a) with compounds q3 and q4 (as reactants) respectively to obtain reaction examples in which compounds q3 and q4 are reactants and participate under reaction rule Fi. For example, the reaction examples in which compound q3 is reactant and participate under reaction rule Fi are shown below:

[0077]

[0078] (Reaxys Reaction ID:1789851, Source: Journal of the Chemical Society. Perkin transactions I, 1983(1), 197–200); and, the reaction characteristics of compound q3 as a reactant in the above-mentioned reaction examples it participates in are: "C`T:4`D:X_1`Nbd:1", and the corresponding product fragment is the part within the solid box; and, the reaction examples in which compound q4 is used as a reactant and participates in the reaction according to reaction rule Fi are as follows:

[0079]

[0080] (Reaxys Reaction ID:1613491, Source: Journal of Organic Chemistry USSR (English Translation), 1991, 27(11.1), 2010–2015; Zhurnnal OrganicheskoiKhimii, 1991, 27(11), 2269–2275); and, the reaction characteristics of compound q4 as a reactant in the above-mentioned reaction examples it participates in are: "O`T:4`Nbd:1", and the corresponding product fragment is the part inside the dashed box;

[0081] b5) Since the linear combination of the reaction characteristics of compounds q3 and q4 as reactants in the reaction example with reaction rule Fi is “C`T:4`D:X_1`Nbd:1|O`T:4`Nbd:1”, which is exactly equal to reaction rule Fi, it can be determined that compounds q3 and q4, as reactants, can undergo a reaction with reaction rule F. i The reaction;

[0082] b6) Based on the product structure fragments corresponding to compounds q3 and q4 as reactants in their respective reaction examples, and considering the reaction characteristics of compounds q3 and q4, it can be seen that each product fragment has one reaction site, and the formation bonds at both reaction sites are single bonds, which is consistent with each other. Therefore, by connecting the two reaction sites in the product fragments with single bonds, the product structure of the predicted reaction can be obtained. The specific predicted reaction is shown below:

[0083]

[0084] This reaction has not been found in existing reaction databases and is still considered an unknown chemical reaction.

[0085] In summary, the method described in this invention can achieve intelligent prediction of unknown chemical reactions based on known reactions with low labor costs, high efficiency, and high accuracy. Furthermore, the reaction database established in this invention is universal and applicable to newly discovered chemical reactions in the future, and is easy to maintain and update in the future. Therefore, this invention has significant progress and application value compared to the prior art.

[0086] Finally, it should be pointed out that the above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of intelligently predicting unknown chemical reactions based on known reactions, characterized in that, The method comprises the following steps: a) establishing a known reaction database Collect known chemical reaction information and use codes to represent the reaction characteristics of each reactant in each known reaction and the reaction rules of each known reaction; the reaction characteristics of each reactant are a linear combination of the reaction characteristics of each site, and the string is sorted from small to large, and separated by the character '`'; the code format of each site reaction characteristic is: the element symbol of the reaction site `T: the type of the reaction site` the environment information of the reaction site `the generation bond information of the reaction site; if the element of the reaction site is fluorine, chlorine, bromine, iodine or astatine, the character 'X' is used as the element symbol of the reaction site, and the rest are used as the element symbol of the corresponding reaction site; b) predicting unknown chemical reactions b1) sequentially searching the known reaction database established in step a) with each target reactant to obtain a set of reaction rules corresponding to each target reactant; b2) determining whether there is an intersection between the sets of reaction rules corresponding to all target reactants to be predicted; b3) if the intersection is empty, it means that according to the current known reaction database, the target reactants to be predicted cannot react, then exit the intelligent prediction; If the intersection is not empty, proceed to step b4); b4) sequentially applying each reaction rule in the intersection to combine each target reactant, and searching the known reaction database established in step a) respectively to obtain each target reactant to the same reaction rule F i participating reaction instances and reaction characteristics of each target reactant in its participating reaction instances; wherein i is a natural number greater than or equal to 1; b5) if the linear combination of the reaction properties of all target reactants to be predicted in the reaction instances they participate in does not equal the reaction rule F i , then it is determined that a reaction with reaction rule F i cannot occur between the target reactants to be predicted; if the linear combination of the reaction properties of all target reactants to be predicted in the reaction instances they participate in equals the reaction rule F i , then it is determined that a reaction with reaction rule F i can occur between the target reactants to be predicted; b6) predicting the structure of the target reaction product according to the product structure fragments corresponding to each target reactant in the reaction instance.

2. The method of intelligently predicting unknown chemical reactions based on known reactions according to claim 1, wherein: According to whether the adjacent atoms and bonds of the reaction site change after the reaction, the reaction site is divided into four types, represented by numbers 1, 2, 3 and 4, as follows: 1: indicates that after the reaction, the original adjacent atom of the reaction site is not removed, but a new adjacent atom is added, and the bond type of the original adjacent bond changes; 2: indicates that after the reaction, the original adjacent atom of the reaction site is not removed, but a new adjacent atom is added, and the bond type of the original adjacent bond does not change; 3: indicates that after the reaction, the original adjacent atom of the reaction site is removed, and a new adjacent atom is added, and the bond type of the remaining adjacent bond changes; 4: indicates that after the reaction, the original adjacent atom of the reaction site is removed, and a new adjacent atom is added, and the bond type of the remaining adjacent bond does not change.

3. The method of intelligently predicting unknown chemical reactions based on known reactions according to claim 2, wherein: Various bond types are represented by numbers, specifically: 1 represents a single bond, 2 represents a double bond, 3 represents a triple bond, and 1.5 represents an aromatic bond.

4. The method of intelligently predicting unknown chemical reactions based on known reactions according to claim 2, wherein: If the original adjacent atom of the reaction site is not removed after the reaction, the information of the double bond, triple bond and aromatic bond contained in the adjacent bond of the reaction site before the reaction is used as the environment information of the reaction site, and the code format is: "E: the type of the adjacent bond to be output"; if multiple bond types are included, they are output in order from small to large; if no double bond, triple bond or aromatic bond is included, the environment information of the reaction site is represented by an empty string "".

5. The method of intelligently predicting unknown chemical reactions based on known reactions as claimed in claim 2, wherein: If the original adjacent atom of the reaction site is removed after the reaction, the removed atom information of the reaction site is used as the environment information of the reaction site, and the code format is: "D: the element symbol of the removed atom _ the bond type between the reaction site and the removed atom".

6. The method of intelligently predicting unknown chemical reactions based on known reactions as claimed in claim 2, wherein: The bond type information between the reaction site and the newly added adjacent atom after the reaction is taken as the generated bond information of the reaction site, and the code format is: "Nbd: bond type between reaction site and newly added adjacent atom", and if there are multiple bond types, the bond type numbers are output in ascending order.

7. The method of intelligently predicting unknown chemical reactions based on known reactions as claimed in claim 1, wherein: In step a), the reaction rule of each known reaction refers to the linear combination of the reaction characteristics of all reactants participating in the reaction, and the string of reaction characteristics is sorted in ascending order and separated by the character '|'.

Images