A method for sorting branch pathways in metabolic networks based on atom cluster tracking
By combining the reaction segment score, compound similarity, reaction thermodynamic freedom and the transfer information of the atomic group, sorting the path of metabolic network branch paths, solving the problem of insufficient relevance of metabolic path biochemistry in existing technologies, and realizing a higher metabolic path recovery Accuracy and biochemical correlation.
Patent Information
- Application Number
- CN202210360183.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-07
- Publication Date
- 2025-01-28
- Estimated Expiration
- 2042-04-07
AI Technical Summary
When sequential metabolic path prediction methods fail to effectively consider functional reaction segments when sorting metabolic paths, resulting in insufficient biochemical correlation of the found metabolic paths and difficulty in recovering known paths.
Adopt the method of metabolic network branch path sorting method based on the atomic group tracking, combine the reaction section score with the similarity of the compound, the reaction thermodynamics, and the transfer information of the atomic group. The accuracy of the path path.
Effectively improved the biochemical correlation of the metabolic path, enhanced the accuracy of the recovery of the known path, and returned a branch metabolic path with biochemical correlation.
Smart Images
Figure 220401175756 
Figure 220401175803 
Figure 220401175810
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of bioinformatics, and in particular relates to a metabolic network branch path sorting method based on atomic group tracking. Background Art
[0002] In the problem of metabolic pathway prediction, metabolic pathway prediction methods may return hundreds of branch metabolic pathways connecting the starting compound and the target compound. In order to select metabolic pathways with greater biochemical relevance, metabolic pathway prediction methods often need to sort the resulting pathways. Although existing metabolic pathway prediction methods usually select biochemical characteristics such as compound structural similarity and reaction free energy to sort the resulting pathways, they do not consider combining functional reaction segments to sort the resulting pathways.
[0003] Atom group transfer refers to a group of atoms transferred from a starting compound to a target compound, where the chemical bonds connecting these atoms are not broken in the chemical reaction of the alternative pathway. Finding a path by tracing the trajectory of atom group transfer can bypass cluster metabolites and find alternative pathways with stronger biochemical relevance without defining the target atoms to be tracked. For metabolic pathways involving atom group transfer, the more the minimum number of atom groups transferred from the starting compound to the target compound, the better the biochemical properties of the metabolic pathway.
[0004] The similarity between compounds refers to the similarity between the molecular structures of substrates and products in the reaction of the metabolic pathway. Here, the SMSD (Small molecule subgraph detector) tool is used to calculate the compound similarity between the reaction substrate and product. The thermodynamic free energy of a reaction is often used to measure whether a metabolic reaction can occur spontaneously. If the thermodynamic free energy of a reaction is negative, it means that the reaction is likely to occur spontaneously without external work.
[0005] Information entropy refers to the amount of information contained in a message, which reflects the reduction in uncertainty about an event after learning a message; mutual information refers to the difference between two discrete random variables. VA 1 and VA 2 Measurement of correlation degree,In order to better evaluate the correlation degree between reaction segments composed of multiple reactions,,the improved mutual information calculation formula proposed by Shi Jing was adopted. Summary of the invention
[0006] The purpose of the present invention is to provide a method for sorting branch pathways of metabolic networks based on atom group tracking, which uses reaction segment scores together with compound similarity, reaction thermodynamic free energy, and atom group transfer information as path sorting weights, and sorts the branch pathways obtained based on atom group tracking, thereby effectively improving the accuracy of restoring known pathways found by metabolic pathways and returning branch metabolic pathways from starting compounds to target compounds with biochemical relevance.
[0007] The technical solution adopted by the present invention to achieve the above-mentioned purpose of the invention is: first, a branch path is generated by branching and merging the linear path obtained based on atomic group tracking; secondly, the original reaction segment dictionary is constructed by extracting the high-frequency reaction segments of 63 known metabolic networks; the information entropy and mutual information of the high-frequency reaction segments are calculated; the next step is to screen out the reaction segments whose information entropy and mutual information are higher than the predetermined threshold as the key reaction segments, and store the key reaction segments and their left and right information entropy and mutual information values into the key reaction segment dictionary; then calculate the reaction segment score in the key reaction segment dictionary; finally, the reaction segment score is used together with the compound similarity, reaction thermodynamic free energy, and atomic group transfer information as the sorting weight of the branch path to sort the branch path.
[0008] The above-mentioned metabolic network branch path sorting method based on atomic group tracking includes the following steps.
[0009] 1) Generate branch paths based on the linear paths obtained by atomic group tracking by branch merging:
[0010] First, a set of linear paths connecting the starting compound and the target compound is found by tracing the atomic groups in the metabolic network. Then, appropriate branch points are selected according to the branch point merging rules, and then the linear paths passing through these branch points are merged at the selected branch points to generate branch paths.
[0011] 2) Extract high-frequency reaction segments of 63 known metabolic networks to construct an original reaction segment dictionary:
[0012] The original reaction segment dictionary is constructed using known metabolic networks. For each metabolic network, all paired combinations of compounds in the network are enumerated, and the two compounds in the paired combination are used as the starting compound and the target compound respectively. The Yen algorithm finds the first five linear metabolic pathways connecting the starting compound and the target compound in the metabolic network and stores the metabolic reactions in the pathway into the original reaction segment dictionary.
[0013] 3) Calculate the information entropy and mutual information of the high-frequency response segment:
[0014] For each high frequency response segment Rs, l is defined as the possible reaction on the left side of the candidate reaction segment, then H( l) is the left information entropy of Rs. Similarly, the right information entropy is defined as H ( r ). For example, suppose there are two reaction groups in the original reaction segment dictionary: (R11, R12, R13) and (R14, R12, R13). For the reaction segment (R12, R13), its left reactions are R11 and R14. The frequencies of the left reactions R11 and R14 in the reaction segment (R12, R13) are 1 and 1 respectively. The reaction segment (R12, R13) has no right reaction. According to the information entropy calculation formula, the left information entropy of the reaction segment (R12, R13) is 1, and the right information entropy is 0. The larger the left and right information entropies, the richer the combination of the reactions before and after the reaction segment, and the greater the possibility that the reaction segment is a key reaction segment.
[0015] The mutual information of high-frequency reaction segments is calculated using the improved mutual information calculation formula proposed by Shi Jing. For example, suppose there are two reaction groups in the original reaction segment dictionary: (R11, R12, R13) and (R14, R12, R13), and there is a candidate reaction segment (R14, R12, R13). The frequency of occurrence of the candidate reaction segment (R14, R12, R13) is 1, and the number of all reaction groups in the original reaction segment dictionary is 2. To calculate the average probability Avg (R14 R12 R13), the reaction segment R14 R12 R13 needs to be divided into (R14, R12R13), (R14 R12, R13), and the average probability of each reaction segment appearing in the original reaction segment dictionary under different divisions is taken as Avg (R14 R12 R13) value. Finally, we can calculate Avg The (R14 R12 R13) value is 0.5. The greater the mutual information, the greater the degree of cohesion between the reactions that constitute the reaction segment, which also means that the reaction segment is more likely to become a key reaction segment.
[0016] 4) Set thresholds to filter out key reaction segments and build a key reaction segment dictionary:
[0017] The candidate reaction segments whose information entropy and mutual information are higher than the set threshold are selected as key reaction segments, and the key reaction segments, information entropy and mutual information are added to the key reaction segment dictionary.
[0018] 5) Calculate the response segment score in the key response segment dictionary.
[0019] 6) The reaction segment score, compound similarity, reaction thermodynamic free energy, and radical transfer information are used as the sorting weights of the branch paths to sort the branch paths.
[0020] The advantages of the present invention are as follows:
[0021] While selecting biochemical characteristics such as compound structure similarity and reaction free energy to sort the result pathways, we also consider combining functional reaction segments to sort the result pathways. This can effectively improve the accuracy of the metabolic pathways found to restore the known pathways and return the branch metabolic pathways with biochemical relevance from the starting compound to the target compound. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 Corresponding formulas for evaluating key response segment scores.
[0023] Figure 2 Scoring formula for linear paths.
[0024] Figure 3 Sorting scoring formula for branching paths.
[0025] Figure 4 Calculate the average R for 30 paths ELCCS Comparison of the values and the number of branching paths in the top five paths.
[0026] Figure 5 This is a comparison chart of the average compound SN value, average compound PPV value and average compound AC value of 30 calculation paths.
[0027] Figure 6 This is a comparison chart of the average reaction R_SN value, average reaction R_PPV value and average reaction R_AC value of 30 calculation paths. DETAILED DESCRIPTION
[0028] The metabolic network branch path sorting method based on atomic group tracking of the present invention has the following specific steps.
[0029] 1) Generate branch paths by branching and merging the linear paths obtained based on the atomic group tracking:
[0030] When searching for a linear metabolic pathway containing atom group transfer from a starting compound to a target compound in a metabolic network, first find the k shortest paths from the starting compound to the target compound in the metabolic network composed of compounds and reactions in the KEGG RPAIR database. Then, the Ullman algorithm is used for each shortest path to calculate the atom group transferred from the starting compound to the target compound after a series of reactions in the path, and the linear paths that retain the atom group from the starting compound in the target compound are taken as candidate linear paths. Then, the compounds that appear in at least two linear paths are taken as candidate branch points, and these candidate branch points are used to form a candidate branch point set. Then, the branch points that meet the rules are screened out from the candidate branch point set according to the branch point merging rules, and then the linear paths passing through these branch points are merged at the selected branch points to generate branch paths.
[0031] 2) Extract high-frequency reaction segments of 63 known metabolic networks to construct an original reaction segment dictionary:
[0032] The original reaction segment dictionary was constructed using 63 metabolic networks from the PATHWAY database in KEGG. For each metabolic network, all compounds contained in the metabolic network were obtained, and all paired combinations of compounds were enumerated. The two compounds in each compound combination were used as the starting compound and the target compound, respectively, and the first five linear metabolic pathways connecting the starting compound and the target compound in the metabolic network were found by the Yen algorithm. Finally, the metabolic reactions in all these found linear pathways were stored in the original reaction segment dictionary.
[0033] 3) Calculate the information entropy and mutual information of the high-frequency response segment.
[0034] 4) Set thresholds to filter out key reaction segments and build a key reaction segment dictionary:
[0035] The left and right information entropies and mutual information of all candidate reaction segments are counted, thresholds are set for the left and right information entropies and mutual information of the candidate reaction segments, and the candidate reaction segments whose left and right information entropies and mutual information are higher than the thresholds are selected as candidate key reaction segments. Finally, the candidate key reaction segments and their left and right information entropies and mutual information values are stored in the key reaction segment dictionary.
[0036] 5) Calculate the response segment score in the key response segment dictionary:
[0037] After constructing the key reaction segment dictionary, we can obtain the left and right information entropy and mutual information of the reaction segments in the key reaction segment dictionary, and then use the attached Figure 1 The calculation formulas (1), (2), and (3) in the above formula evaluate the score of the key reaction segment. r1…rm is the reaction section, SKRS(r1…rm) Refers to the score of the key reaction segment. H'(r1…rm) Reaction section r1…rm The statistical score of the left information entropy and the right information entropy. I (r1…rm) Reaction segment r1…rm The normalized value of the mutual information of . Hl(r1…rm) Reaction segment r1…rm The left information entropy of Hr(r1…rm) Reaction segment r1…rm The right information entropy. minHl represents the minimum left information entropy among all the reaction segments in the key reaction segment dictionary, maxHl represents the maximum left information entropy among all reaction segments in the key reaction segment dictionary. minHr represents the minimum right information entropy among all reaction segments in the key reaction segment dictionary. maxHr Represents the maximum right information entropy among all the reaction segments in the key reaction segment dictionary. minMMI Refers to the minimum mutual information value among all the reaction segments in the key reaction segment dictionary, maxMMI Refers to the maximum mutual information value among all the reaction segments in the key reaction segment dictionary.
[0038] 6) The reaction segment score, compound similarity, reaction thermodynamic free energy, and atom group transfer information are used as the sorting weight of the branch path to sort the branch path. Linear path L P Rating Score(L P ) By attaching Figure 2 Calculate using formula (4). c i and c j It is a linear path L P The two compounds connected by the reaction, sim(c i , c j ) Represents compound c i and c j The similarity of compounds n r Refers to the linear path L P The number of reactions in . r i Represents a linear path L P One of the reactions, f (r i ) Reaction r i Thermodynamic free energy, the range of thermodynamic free energy is between -2233.7 and 10194.7, (3200+fe( r i )) / 10000 is the reaction r i Thermodynamic free energy fe(r i ) Regularization makes (3200+ fe( r i )) / 10000 has a value between 0.097 and 1.339. t represents the path L P The number of the smallest atomic group transferred from the starting compound to the target compound in , b represents the number of bonds connecting atoms in the starting compound, and t / b is the regularization of t. SKRS(Rs)is the score of the reaction segment obtained in step (5) as the key reaction segment , Rs Refers to the key reaction segment dictionary and the online linear path L P The reaction segments that appear simultaneously in the . α s 、α fe 、α t 、α r are weight parameters, which are used to adjust compound similarity, reaction thermodynamic free energy, atom group transfer information and key reaction segment. Score(L P ) degree of impact. α s 、α fe 、α t 、α r The value range is between 0 and 1, and can be adjusted according to user needs.
[0039] The present invention adopts Figure 3 The formula in the branch path calculation B P Ranking score Score_br(B P ) .in, n b Indicates branch path B P The number of branch points in n l Refers to the branch path B P The number of linear paths, l i Indicates the branch path B P Linear path. Score(L i ) Indicates that by attaching Figure 2 The linear path calculated by formula (4) l i Linear path score. α b It is a weight parameter used to adjust the influence of branch point information and linear path score on the branch path sorting formula. The value range is between 0 and 1 and can be adjusted according to user needs. Finally, the branch paths with high scores in the calculation results are ranked first.
[0040] The present invention compares seven indicators of known branch paths and calculated branch paths: the proportion of the largest connected subgraph edges R ELCCS , compound sensitivity SN, compound positive prediction value PPV, compound accuracy AC, reaction sensitivity R_SN, reaction positive prediction value R_PPV, reaction accuracy R_AC to evaluate the biochemical relevance of the found branch pathways. The branch pathways found by the method CRSPFinder of the present invention with added key reaction scores and other methods for finding branch metabolic pathways BPFinder, BPAT-M, BPAT-S, Retrace, PhdSeeker and MetQuest were compared with 30 known branch pathways in the KEGG PATHWAY database to evaluate the performance of CRSPFinder.
[0041] CRSPFinder uses the following search parameters for branch paths: the number of candidate linear paths k to be searched is 2000, the minimum length of a candidate linear path is 2, the maximum length is 15, the minimum number of atomic groups in a branch path from the starting compound to the target compound is 2 and α s is 0.1, α fe is 0.2, α t is 0.2, α b is 0.8, α r is 0.1, and the threshold is 0. The operating parameters of other methods involved in the comparison are set with reference to the operating parameter values given by the authors in the paper.
[0042] Attached Figure 4 is the average R of 30 calculation paths ELCCS Comparison chart of the value and the number of branch paths in the first five paths; Attachment Figure 5 This is a comparison chart of the average compound SN value, average compound PPV value and average compound AC value of the 30 calculation paths; Figure 6 The figure is a comparison chart of the average reaction R_SN value, average reaction R_PPV value and average reaction R_AC value of 30 calculated pathways. The above result analysis and comparison show that the pathway found by the method of the present invention is closer to the known pathway, and can effectively improve the accuracy of restoring the known pathway by the found metabolic pathway.
Claims
1. A metabolic network branch path sorting method based on atom group tracking, characterized in that: include: 1) Generate branch paths by branching and merging the linear paths obtained based on the atomic group tracking: First, a set of linear paths connecting the starting compound and the target compound is found by tracing the atomic groups in the metabolic network. Then, appropriate branch points are selected according to the branch point merging rules, and the linear paths passing through these branch points are merged at the selected branch points to generate branch paths. 2) Extract high-frequency reaction segments of 63 known metabolic networks to construct an original reaction segment dictionary: Use known metabolic networks to construct the original reaction segment dictionary. For each metabolic network, enumerate all the paired combinations of compounds in the network. The two compounds in the paired combination are used as the starting compound and the target compound respectively. The Yen algorithm is used to find the first five linear metabolic pathways connecting the starting compound and the target compound in the metabolic network and store the metabolic reactions in the pathway into the original reaction segment dictionary. 3) Calculate the information entropy and mutual information of the high-frequency response segment; 4) Set thresholds to filter out key reaction segments and build a key reaction segment dictionary: Select candidate reaction segments whose information entropy and mutual information are higher than a set threshold as key reaction segments, and add the key reaction segments, information entropy and mutual information into the key reaction segment dictionary; 5) calculating the response segment scores in the key response segment dictionary; 6) The reaction segment score, compound similarity, reaction thermodynamic free energy, and radical transfer information are used as the sorting weights of the branch paths to sort the branch paths.