A compound combination screening method based on key genes for resistance to parasitic infections

Abstract

The present invention discloses a compound combination screening method based on key genes for resisting parasitic infections, comprising: 1. After preprocessing biological data, a single-layer differential interaction network is constructed for each individual infected by parasites, and a multi-layer gene interaction network is constructed by connecting the corresponding gene nodes of multiple single-layer networks to achieve the fusion of multiple networks, and then the PageRank algorithm of the multi-layer network is used to calculate the score of the gene in each layer of the network, and the average score of the corresponding same gene in all layers is used to sort and identify the key gene; 2. Based on the key gene and the known FDA-approved compounds and the target gene set of each compound, a multi-objective optimization function model is designed, and a multi-objective optimization algorithm is used to obtain a better compound combination. The present invention can improve the time efficiency of screening the compound combination for resisting parasitic infections and screen out a better compound combination.

Description

Technical Field

[0001] The invention belongs to the field of biological information mining, and specifically is a compound combination screening method based on key genes for resisting parasitic infection. Background Art

[0002] The effect of a single compound on a single gene target will not have a significant effect on combating parasitic infections, and may even activate other factors, causing the compound to lose its effect or produce adverse reactions. Two or more compounds are often used in combination to enhance the effect of combating parasitic infections or reduce adverse reactions. Compared with traditional single-ingredient, single-gene target compounds, compound combinations have the characteristics of multiple components and multiple gene targets, and can more effectively combat parasitic infections.

[0003] Existing methods for screening compound combinations mainly involve high-throughput screening and machine learning. The disadvantages of these two methods mainly include:

[0004] 1. High-throughput screening is a method of obtaining compound combinations through instrumental experiments. Faced with a large number of compounds, there is a problem of combinatorial explosion. The efficiency of screening compound combinations is low and consumes huge manpower, material and financial resources;

[0005] 2. Machine learning refers to training models based on compound similarity weights or other properties. Model training requires a large amount of known compound combination data as a basis, but the currently available compound combination data is limited, which is not conducive to model training. Summary of the invention

[0006] The present invention aims to solve the deficiencies of the above-mentioned prior art and proposes a method for screening compound combinations based on key genes for anti-parasitic infection, in order to obtain a priority key gene set related to parasitic infection, and to improve the time efficiency of screening compound combinations for anti-parasitic infection by using a multi-objective optimization algorithm based on the key gene set, thereby obtaining good time performance on the compound combinations for anti-parasitic infection and screening out a better compound combination.

[0007] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:

[0008] The present invention is characterized in that a compound combination screening method based on key genes for resisting parasitic infection comprises the following steps:

[0009] Step 1: Biological data preprocessing:

[0010] Step 1.1: Based on the known N genes associated with parasite infection in humans {g 1 ,g 2 ,...,g n,...,g N}, and obtain the N parasite-infected human-associated genes {g 1 ,g 2 ,...,g n ,...,g N} expression matrix dataData 1 ={s 1 ,s 2 ,...s c ,...,s C} and the associated genes of N parasites infecting humans in M ​​parasite infection samples {g 1 ,g 2 ,...,g n ,...,g N} expression matrix dataData 2 ={d 1 ,d 2 ,...,d m ,...,d M}; where g n represents the gene associated with the nth parasite infecting the human body, n∈[1,N]; s c represents the expression value vector of N associated genes in the cth normal sample; and Indicates the nth associated gene g in the cth normal sample n The expression value of m represents the expression value vector of N associated genes in the mth infected sample; and Indicates the nth associated gene g in the mth infected sample n The expression value of

[0011] Step 1.2, obtain Z compounds {r 1 ,r 2 ,..,r z ,...,r Z} and the target gene set of each compound {T 1 ,T 2 ,...,T z ,...,T Z}, where r z represents the zth compound, T z represents the zth compound r z Target gene set;

[0012] Step 2: Multi-layer gene network construction:

[0013] Step 2.1: Based on the expression matrix data of N associated genes of C normal samples1 , calculate the Pearson correlation coefficients between the expression values ​​of N associated genes, and then obtain the Pearson correlation coefficient set P = {p 12 ,p 13 ,...,p (n-1)n ,...,p (N-1)N}, p (n-1)n Indicates the n-1th associated gene g n-1 The expression value of the nth gene g n Correlation coefficient of expression values;

[0014] Step 2.2: In the expression matrix data 1 The gene expression value vector d of the mth parasite infection sample is added to each expression value vector of m , thus obtaining the expression spectrum matrix data NewData containing a single infected sample m , and recalculate the expression matrix data NewData m The Pearson correlation coefficients between the expression value vectors of the N associated genes in , thus obtaining the Pearson correlation coefficient set NewP containing a single infected sample m ={p′ 12 ,p′ 13 ,...,p′ (n-1)n ,...,p′ (N-1)N}, p′ (n-1)n Indicates the n-1th gene g containing a single infected sample n-1 The expression value vector and the nth gene g n Correlation coefficient of the expression value vector;

[0015] Step 2.3, calculate the difference between two sets of Pearson correlation coefficients |NewP m -P|={|p′ 12 -p 12 |,|p′ 13 -p 13 |,...,|p′ (n-1)n -p (n-1)n |,...,|p′ (N-1)N -p (N-1)N |}, and select all expression value vectors of two related genes that are greater than the set difference threshold, and construct the edge between the selected two related genes, so as to obtain the single-layer individual-specific network G of the mth single infected sample m ;

[0016] Step 2.4: Repeat steps 2.2-2.3 to construct a single-layer individual-specific network {G 1 ,G 2 ,...,Gm ,...,G M};

[0017] Step 2.4: Connect the nodes corresponding to the same associated genes in each single-layer individual-specific network, and set the weights of the edges within and between layers to 1, thereby obtaining an M-layer network G;

[0018] Step 3: Calculate the centrality of the M-layer network G:

[0019] Step 3.1, set the total number of iterations to Iter, the current number of iterations to iter, the absolute error to ε, and the damping coefficient to d;

[0020] Use the fourth-order adjacency tensor E to represent the weighted edges in the M-layer network G, and let the i-th associated gene g in the α-th layer i To the jth associated gene g in the βth layer j The weighted edge of The adjacency tensor E is normalized to obtain the normalized adjacency tensor E′. The i-th associated gene g in the αth layer after normalization is i To the jth associated gene g in the βth layer j The weighted edge of

[0021] Step 3.2, initialize iter = 1, randomly initialize the second-order centrality tensor Φ of the iter-th iteration iter , the second-order centrality tensor Φ of the iter-th iteration iter The i-th associated gene g in the αth layer i The centrality of and

[0022] Step 3.3: Let the second-order influence tensor W of the iter-th iteration be iter Represents the influence between layers. Let the influence of the αth layer on the βth layer in the iterth iteration be recorded as and in, represents the importance of the αth layer at the iterth iteration, and Represents the second-order centrality tensor Φ according to the iter-th iteration iter The sum of the centralities of the associated genes in the αth layer, n α represents the number of associated genes in the αth layer, represents the importance of the βth layer at the iterth iteration, and Represents the second-order centrality tensor Φ according to the iter-th iteration iter The sum of the centralities of the associated genes in the βth layer, nβ represents the number of associated genes in the βth layer;

[0023] Step 3.4: Use the fourth-order interaction tensor H of the iterth iteration iter =EW iter represents the interaction between two associated genes in the iter-th iteration of the M-layer network G. Let the i-th associated gene g in the α-th layer of the iter-th iteration i To the jth associated gene g in the βth layer j The interaction is denoted as and in, Represents a fourth-order tensor of all “1s” with the same dimension as the normalized adjacency tensor E′.

[0024] Step 3.5: Solve the tensor equation H for the iterth iteration iter Φ iter+1 =λΦ iter+1 Get the second-order tensor Φ of the iter+1th iteration iter+1 ,in, represents the i-th associated gene g of the α-th layer at the iter-th iteration i The centrality of , λ represents the characteristic coefficient.

[0025] Step 3.6: If Φ iter -Φ iter+1 If the 1-norm of is less than the absolute error ε, it indicates that the centrality of the M-layer network G is obtained, and step 3.7 is executed; otherwise, iter+1 is assigned to iter and it is determined whether iter>Iter is established. If so, it indicates that the centrality of the M-layer network G is obtained, and step 3.7 is executed. Otherwise, return to step 3.3 and execute sequentially;

[0026] Step 3.7: According to the second-order tensor Φ of the iter+1th iteration iter+1 Get the i-th associated gene g i The centrality of the M-layer network G is calculated and the mean is obtained. After sorting all the associated genes in descending order according to their centrality mean, the top H key genes are selected to form the optimal key gene set Y;

[0027] Step 4: Based on the preferred key gene set Y, two objective functions F are constructed using formula (1): 1 and F 2 :

[0028]

[0029] In formula (1), T u and T v are the uth compound r uand the vth compound r v Gene target collection;

[0030] Step 5: Use the improved SparseEA genetic algorithm to solve the two objective functions F 1 and F 2 To solve:

[0031] Step 5.1: Initialize the population using binary encoding:

[0032] Set the population size to P, and define the population individual set as A = {A 1 ,A 2 ,...,A p ,...,A P},A p represents the pth individual in the population, with the pth individual A p represents the pth compound combination; and A p ={X 1 ,X 2 ,...,X u ,...,X v ...,X Z}, where X u represents the uth compound, X v represents the vth compound, and X 1 ~X Z ∈{0,1}, Z represents the number of compounds;

[0033] Step 5.2: Set the maximum number of iterations to I max , set the current iteration to I, and initialize I = 1;

[0034] Step 5.3: Randomly generate P individuals of the I-th generation population as the initial solution:

[0035] Set the individual A of the I generation population p (I) = {X 1 (I),X 2 (I),...,X u (I),...,X v (I)...,X Z (I)}, only two compounds have a value of 1, and the other compounds have a value of 0. u (I) = 1 and X v (I) = 1, which means that the u-th compound X in the individual of the I-th generation population u (I) and the vth compound X v The combination of (I) is denoted as (u,v);

[0036] Step 5.4, cross:

[0037] From the I generation population individual A(I) = {A 1 (I),A 2 (I),...,A p (I),...,A P (I)} Randomly select two individuals from the population and

[0038] in, It means randomly selecting the nth individual A from the Ith generation population individual A(I) η The ath compound in (I), represents the nth individual A in the population η The bth compound in (I), It means randomly selecting the μth individual A from the Ith generation population individual A(I) μ The kth compound in (I), represents the μth individual A in the population μ The first compound in (I), and and According to the following three situations, η (I) and A μ (I) Crossover:

[0039] If there is an identical compound in the combination (a, b) and (k, l), the other two different compounds are combined to complete the crossover and obtain the individual after the crossover;

[0040] If there are no identical compounds in the combinations (a, b) and (k, l), the compounds in one combination are randomly combined with the compounds in the other combination to complete the crossover and obtain the individuals after the crossover;

[0041] If there are two identical compounds in the combination (a, b) and (k, l), no crossover is performed and they are directly used as individuals after crossover;

[0042] Step 5.5, mutation:

[0043] The random probability is used to determine whether the individual after the crossover will mutate. If it mutates, the mutation method is to randomly select a compound with a value of 1 in the individual after the crossover to change it to 0, and at the same time randomly select a compound with a value of 0 to change it to 1, so as to obtain the mutated individual; otherwise, the individual after the crossover is directly used as the mutated individual;

[0044] Step 5.6: Repeat steps 5.4 and 5.5 until P crossover mutations are completed, and obtain the population A′(I) after crossover mutation;

[0045] Step 5.7: After combining the I-generation population A(I) with the I-generation crossover population A′(I), calculate the non-dominated sorting and crowding distance between individuals, and select the first P individuals as the I+1 generation population individuals to determine if I+1>I max Is it true? If true, it means that the I max Generation of individuals and obtaining the I max The compound combination represented by the individuals in the generation population is the optimal compound combination, otherwise, assign I+1 to I and return to step 5.4 to execute sequentially.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The method proposed in the present invention is based on multi-objective optimization model screening, which is not limited to the number of compounds. Even if a large number of compounds are used to screen compound combinations, the multi-objective optimization algorithm can significantly improve the efficiency of compound screening. In addition, two objective functions are constructed based on the compound gene target set and the key gene set of parasitic infection, considering the effect of anti-parasitic infection and possible adverse reactions, and the multi-objective optimization algorithm is used to solve the optimal compound combination, thereby converting the problem into an intuitive mathematical model. The rapid convergence of the multi-objective optimization algorithm can improve the speed of compound combination screening, effectively avoiding the combinatorial explosion problem of a large number of compounds.

[0048] 2. The method for selecting key gene sets proposed in the present invention is to construct a multi-layer network based on the gene expression profile data of normal samples and parasite infection samples to calculate the centrality of associated genes. The number of layers of the network is equal to the number of parasite infection samples. Therefore, this method is still applicable to expression profile data of only a small number of parasite infection samples;

[0049] 3. The method for screening compound combinations proposed in the present invention is based on a priority key gene set related to parasitic infection, which is different from existing studies that consider all genes related to parasitic infection. This enables the compound combination to act more specifically on important genes associated with parasitic infection, thereby obtaining a compound combination with better anti-parasitic infection effect; BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flow chart of the method for combinatorial screening of compounds of the present invention;

[0051] Figure 2 A schematic diagram of constructing a multi-layer gene network according to the present invention;

[0052] Figure 3 This is a schematic diagram of the genes associated with the combination of compounds of the present invention. DETAILED DESCRIPTION

[0053] In this embodiment, refer to Figure 1 A method for screening compound combinations based on key genes for resistance to parasitic infections comprises the following steps:

[0054] Step 1: Biological data preprocessing:

[0055] Step 1.1: Based on the known N genes associated with parasite infection in humans {g 1 ,g 2 ,...,g n ,...,g N}, and obtain the N parasite-infected human-associated genes {g 1 ,g 2 ,...,g n ,...,g N} expression matrix dataData 1 ={s 1 ,s 2 ,...s c ,...,s C} and the associated genes of N parasites infecting humans in M ​​parasite infection samples {g 1 ,g 2 ,...,g n ,...,g N} expression matrix dataData 2 ={d 1 ,d 2 ,...,d m ,...,d M}; where g n represents the gene associated with the nth parasite infecting the human body, n∈[1,N]; s c represents the expression value vector of N associated genes in the cth normal sample; and Indicates the nth associated gene g in the cth normal sample n The expression value of m represents the expression value vector of N associated genes in the mth infected sample; and Indicates the nth associated gene g in the mth infected sample n The expression value of

[0056] Step 1.2, obtain Z compounds {r 1 ,r 2 ,..,r z ,...,r Z} and the target gene set of each compound {T 1 ,T 2 ,...,T z ,...,TZ}, where r z represents the zth compound, T z represents the zth compound r z The target gene set;

[0057] Step 2: Figure 2 The multi-layer gene network is constructed as shown below:

[0058] Step 2.1: Based on the expression matrix data of N associated genes of C normal samples 1 , calculate the Pearson correlation coefficients between the expression values ​​of N associated genes, and then obtain the Pearson correlation coefficient set P = {p 12 ,p 13 ,...,p (n-1)n ,...,p (N-1)N}, p (n-1)n Indicates the n-1th associated gene g n-1 The expression value of the nth gene g n Correlation coefficient of expression values;

[0059] Step 2.2: In the expression matrix data 1 The gene expression value vector d of the mth parasite infection sample is added to each expression value vector of m , thus obtaining the expression spectrum matrix data NewData containing a single infected sample m , and recalculate the expression matrix data NewData m The Pearson correlation coefficients between the expression value vectors of the N associated genes in , thus obtaining the Pearson correlation coefficient set NewP containing a single infected sample m ={p′ 12 ,p′ 13 ,...,p′ (n-1)n ,...,p′ (N-1)N}, p′ (n-1)n Indicates the n-1th gene g containing a single infected sample n-1 The expression value vector and the nth gene g n Correlation coefficient of the expression value vector;

[0060] Step 2.3, calculate the difference between two sets of Pearson correlation coefficients |NewP m -P|={|p′ 12 -p 12 |,|p′ 13 -p 13 |,...,|p′ (n-1)n -p (n-1)n |,...,|p′ (N-1)N -p (N-1)N|}, this difference reflects the specificity of the correlation coefficient after adding the parasite infection sample, and selects all the expression value vectors of two related genes that are greater than the set difference threshold, and constructs the edge between the selected two related genes, thereby obtaining the single-layer individual specific network G of the mth single infection sample m ;

[0061] Step 2.4: Repeat steps 2.2-2.3 to construct a single-layer individual-specific network {G 1 ,G 2 ,...,G m ,...,G M};

[0062] Step 2.5, connect the nodes corresponding to the same associated genes in each single-layer individual-specific network, and set the weights of the edges within and between layers to 1, thereby obtaining an M-layer network G;

[0063] Step 3: Calculate the centrality of the M-layer network G:

[0064] Step 3.1, set the total number of iterations to Iter, the current number of iterations to iter, the absolute error to ε, and the damping coefficient to d;

[0065] Use the fourth-order adjacency tensor E to represent the weighted edges in the M-layer network G, and let the i-th associated gene g in the α-th layer i To the jth associated gene g in the βth layer j The weighted edge of The adjacency tensor E is normalized to obtain the normalized adjacency tensor E′. The i-th associated gene g in the αth layer after normalization is i To the jth associated gene g in the βth layer j The weighted edge of

[0066] Step 3.2, initialize iter = 1, randomly initialize the second-order centrality tensor Φ of the iter-th iteration iter , the second-order centrality tensor Φ of the iter-th iteration iter The i-th associated gene g in the αth layer i The centrality of and

[0067] Step 3.3: Let the second-order influence tensor W of the iter-th iteration be iter Represents the influence between layers. Let the influence of the αth layer on the βth layer in the iterth iteration be recorded as and in, represents the importance of the αth layer, and Represented by the second-order centrality tensor Φ iter The sum of the centralities of the associated genes in the αth layer, n α represents the number of associated genes in the αth layer, represents the importance of the βth layer, and Represented by the second-order centrality tensor Φ iter The sum of the centralities of the associated genes in the βth layer, n β represents the number of associated genes in the βth layer;

[0068] Step 3.4: Use the fourth-order interaction tensor H iter =EW iter Represents the interaction between two associated genes in the M-layer network G. Let the i-th associated gene g in the α-th layer of the iter-th iteration i To the jth associated gene g in the βth layer j The interaction is denoted as and in, Represents a fourth-order tensor of all “1s” with the same dimension as the normalized adjacency tensor E′.

[0069] Step 3.5: Solve the tensor equation H iter Φ iter+1 =λ·Φ iter+1 Get the second-order tensor Φ of the iter+1th iteration iter+1 ,in, represents the i-th associated gene g in the αth layer i The centrality, λ represents the characteristic coefficient, where λ is taken as 1 to ensure that the second-order tensor Φ iter+1 existence and uniqueness.

[0070] Step 3.6: If Φ iter -Φ iter+1 If the 1-norm of is less than the absolute error ε, it indicates that the centrality of the M-layer network G is obtained, and step 3.7 is executed; otherwise, iter+1 is assigned to iter and it is determined whether iter>Iter is established. If so, it indicates that the centrality of the M-layer network G is obtained, and step 3.7 is executed. Otherwise, return to step 3.3 and execute sequentially;

[0071] Step 3.7: According to the second-order tensor Φ of the iter+1th iteration iter+1 Get the i-th associated gene g i The centrality of the M-layer network G is calculated and the mean is obtained. After sorting all the associated genes in descending order according to their centrality mean, the top H key genes are selected to form the optimal key gene set Y;

[0072] Step 4: Based on the preferred key gene set Y, two objective functions F are constructed using formula (1): 1 and F 2 :

[0073]

[0074] In formula (1), T u and T v are the uth compound r u and the vth compound r v Gene target collection. 1 Represents compound r u and compound r v The proportion of overlapping genes between the combined target gene set and the preferred key gene set Y indicates the effect of the compound group combination on anti-parasitic infection, F 2 Represents compound r u and compound r v The proportion of genes in the combined target gene set other than the preferred key gene set Y indicates the possible adverse reactions of the compound combination, such as Figure 3 shown.

[0075] Step 5: Use the improved SparseEA genetic algorithm to solve the two objective functions F 1 and F 2 To solve:

[0076] Step 5.1: Initialize the population using binary encoding:

[0077] Set the population size to P, and define the population individual set as A = {A 1 ,A 2 ,...,A p ,...,A P},A p represents the pth individual in the population, with the pth individual A p represents the pth compound combination; and A p ={X 1 ,X 2 ,...,X u ,...,X v ...,X Z}, where X u represents the uth compound, X v represents the vth compound, and X 1 ~X Z ∈{0,1}, Z represents the number of compounds;

[0078] Step 5.2: Set the maximum number of iterations to I max, set the current iteration to I, and initialize I = 1;

[0079] Step 5.3: Randomly generate P individuals of the I-th generation population as the initial solution:

[0080] Set the individual A of the I generation population p (I) = {X 1 (I),X 2 (I),...,X u (I),...,X v (I)...,X Z (I)}, only two compounds have a value of 1, and the other compounds have a value of 0. u (I) = 1 and X v (I) = 1, which means that the u-th compound X in the individual of the I-th generation population u (I) and the vth compound X v The combination of (I) is denoted as (u,v);

[0081] Step 5.4, cross:

[0082] From the I generation population individual A(I) = {A 1 (I),A 2 (I),...,A p (I),...,A P (I)} Randomly select two individuals from the population and

[0083] in, It means randomly selecting the nth individual A from the Ith generation population individual A(I) η The ath compound in (I), represents the nth individual A in the population η The bth compound in (I), It means randomly selecting the μth individual A from the Ith generation population individual A(I) μ The kth compound in (I), represents the μth individual A in the population μ The first compound in (I), and and According to the following three situations, η (I) and A μ (I) Crossover:

[0084] If there is an identical compound in the combination (a, b) and (k, l), then the other two different compounds are combined to complete the crossover and obtain the individual after the crossover; for example, if a = k and b ≠ l, the crossover result is and If a=l and b≠k, the crossover result is and If b = k and a ≠ l, the crossover result is and If b = l and a ≠ k, the crossover result is and

[0085] If there are no identical compounds in the combinations (a, b) and (k, l), that is, a≠b≠k≠l, the compounds in one combination are randomly combined with the compounds in the other combination to complete the crossover and obtain the individuals after the crossover; that is, the crossover result is and or and or and or and

[0086] If there are two identical compounds in the combination (a, b) and (k, l), that is, a=k and b=l, no crossover is performed and they are directly used as individuals after crossover; that is, the crossover result is and

[0087] Step 5.5, mutation:

[0088] The random probability determines whether the individual after the crossover will mutate. If it mutates, the mutation method is to randomly select the individual B′ after the crossover. ω (I) A compound with a value of 1 is changed to 0, and a compound with a value of 0 is randomly selected and changed to 1 to obtain the mutated individual; otherwise, the individual after crossover is directly used as the mutated individual;

[0089] Step 5.6: Repeat steps 5.4 and 5.5 until P crossover mutations are completed, and obtain the population A′(I) after crossover mutation;

[0090] Step 5.7: After combining the I-generation population A(I) with the I-generation crossover population A′(I), calculate the non-dominated sorting and crowding distance between individuals, and select the first P individuals as the I+1 generation population individuals to determine if I+1>I max Is it true? If true, it means that the I max Generation population individuals and obtain the I max The compound combination represented by the individuals in the generation population is the optimal compound combination, otherwise, assign I+1 to I and return to step 5.4 to execute sequentially.

Claims

1. A method for screening compound combinations based on key genes for resistance to parasitic infections, characterized by comprising the following steps: Step 1: Biological data preprocessing: Step 1.1: Based on the known N genes associated with parasite infection in humans {g1, g2, ..., g n ,...,g N }, and obtain the N parasite-infected human-associated genes {g1,g2,...,g n ,...,g N } expression spectrum matrix data Data1 = {s1, s2, ...s c ,...,s C } and the associated genes {g1,g2,...,g n ,...,g N } expression spectrum matrix data Data2 = {d1, d2, ..., d m ,...,d M }; where g n represents the gene associated with the nth parasite infecting the human body, n∈[1,N]; s c represents the expression value vector of N associated genes in the cth normal sample; and Indicates the nth associated gene g in the cth normal sample n The expression value of m represents the expression value vector of N associated genes in the mth infected sample; and Indicates the nth associated gene g in the mth infected sample n The expression value of Step 1.2, obtain Z compounds {r1, r2, .., r z ,...,r Z } and the target gene set of each compound {T1,T2,...,T z ,...,T Z }, where r z represents the zth compound, T z represents the zth compound r z Target gene set; Step 2: Multi-layer gene network construction: Step 2.1, according to the expression spectrum matrix data Data1 of N associated genes of C normal samples, calculate the Pearson correlation coefficients between the expression values ​​of N associated genes, so as to obtain the Pearson correlation coefficient set P = {p 12 ,p 13 ,...,p (n-1)n ,...,p (N-1)N }, p (n-1)n Indicates the n-1th associated gene g n-1 The expression value of the nth gene g n Correlation coefficient of expression values; Step 2.2: Add the gene expression value vector d of the mth parasite infection sample to each expression value vector of the expression spectrum matrix data Data1. m , thus obtaining the expression spectrum matrix data NewData containing a single infected sample m , and recalculate the expression matrix data NewData m The Pearson correlation coefficients between the expression value vectors of the N associated genes in , thus obtaining the Pearson correlation coefficient set NewP containing a single infected sample m = {p′ 12 ,p′ 13 ,...,p′ (n-1)n ,...,p′ (N-1)N }, p′ (n-1)n Indicates the n-1th gene g containing a single infected sample n-1 The expression value vector and the nth gene g n Correlation coefficient of the expression value vector; Step 2.3, calculate the difference between two sets of Pearson correlation coefficients |NewP m -P|={|p′ 12 -p 12 |,|p′ 13 -p 13 |,...,|p′ (n-1)n -p (n-1)n |,...,|p′ (N-1)N -p (N-1)N |}, and select all expression value vectors of two related genes that are greater than the set difference threshold, and construct the edge between the selected two related genes, so as to obtain the single-layer individual-specific network G of the mth single infected sample m ; Step 2.4: Repeat steps 2.2-2.3 to construct a single-layer individual-specific network {G1, G2, ..., G m ,...,G M }; Step 2.4: Connect the nodes corresponding to the same associated genes in each single-layer individual-specific network, and set the weights of the edges within and between layers to 1, thereby obtaining an M-layer network G; Step 3: Calculate the centrality of the M-layer network G: Step 3.1, set the total number of iterations to Iter, the current number of iterations to iter, the absolute error to ε, and the damping coefficient to d; Use the fourth-order adjacency tensor E to represent the weighted edges in the M-layer network G, and let the i-th associated gene g in the α-th layer i To the jth associated gene g in the βth layer j The weighted edge of The adjacency tensor E is normalized to obtain the normalized adjacency tensor E′. The i-th associated gene g in the αth layer after normalization is i To the jth associated gene g in the βth layer j The weighted edge of Step 3.2, initialize iter = 1, randomly initialize the second-order centrality tensor Φ of the iter-th iteration iter , the second-order centrality tensor Φ of the iter-th iteration iter The i-th associated gene g in the αth layer i The centrality of and Step 3.3: Let the second-order influence tensor W of the iter-th iteration be iter Represents the influence between layers. Let the influence of the αth layer on the βth layer in the iterth iteration be recorded as and in, represents the importance of the αth layer at the iterth iteration, and Represents the second-order centrality tensor Φ according to the iter-th iteration iter The sum of the centralities of the associated genes in the αth layer, n α represents the number of associated genes in the αth layer, represents the importance of the βth layer at the iterth iteration, and Represents the second-order centrality tensor Φ according to the iter-th iteration iter The sum of the centralities of the associated genes in the βth layer, n β represents the number of associated genes in the βth layer; Step 3.4: Use the fourth-order interaction tensor H of the iterth iteration iter =EW iter represents the interaction between two associated genes in the iter-th iteration of the M-layer network G. Let the i-th associated gene g in the α-th layer of the iter-th iteration i To the jth associated gene g in the βth layer j The interaction is denoted as and in, represents a fourth-order tensor of all "1" with the same dimension as the normalized adjacency tensor E'; Step 3.5: Solve the tensor equation H for the iterth iteration iter Φ iter+1 =λΦ iter+1 Get the second-order tensor Φ of the iter+1th iteration iter+1 ,in, represents the i-th associated gene g of the α-th layer at the iter-th iteration i The centrality of , λ represents the characteristic coefficient; Step 3.6: If Φ iter -Φ iter+1 If the 1-norm of is less than the absolute error ε, it indicates that the centrality of the M-layer network G is obtained, and step 3.7 is executed; otherwise, iter+1 is assigned to iter and it is determined whether iter>Iter is established. If so, it indicates that the centrality of the M-layer network G is obtained, and step 3.7 is executed. Otherwise, return to step 3.3 and execute sequentially; Step 3.7: According to the second-order tensor Φ of the iter+1th iteration iter+1 Get the i-th associated gene g i The centrality of the M-layer network G is calculated and the mean is obtained. After sorting all the associated genes in descending order according to their centrality mean, the top H key genes are selected to form the optimal key gene set Y; Step 4: Based on the preferred key gene set Y, two objective functions F1 and F2 are constructed using formula (1): In formula (1), T u and T v are the uth compound r u and the vth compound r v Gene target collection; Step 5: Use the improved SparseEA genetic algorithm to solve the two objective functions F1 and F2: Step 5.1: Initialize the population using binary encoding: Set the population size to P, and define the population individual set as A = {A1, A2, ..., A p ,...,A P },A p represents the pth individual in the population, with the pth individual A p represents the pth compound combination; and A p ={X1,X2,...,X u ,...,X v ...,X Z }, where X u represents the uth compound, X v represents the vth compound, and X1~X Z ∈{0,1}, Z represents the number of compounds; Step 5.2: Set the maximum number of iterations to I max , set the current iteration to I, and initialize I = 1; Step 5.3: Randomly generate P individuals of the I-th generation population as the initial solution: Set the individual A of the I generation population p (I) = {X1(I),X2(I),...,X u (I),...,X v (I)...,X Z (I)}, only two compounds have a value of 1, and the other compounds have a value of 0. u (I) = 1 and X v (I) = 1, which means that the u-th compound X in the individual of the I-th generation population u (I) and the vth compound X v The combination of (I) is denoted as (u,v); Step 5.4, cross: From the I-generation population individuals A(I) = {A1(I), A2(I), ..., A p (I),...,A P (I)} Randomly select two individuals from the population and in, It means randomly selecting the nth individual A from the Ith generation population individual A(I) η The ath compound in (I), represents the nth individual A in the population η The bth compound in (I), X k μ (I) means randomly selecting the μth individual A from the Ith generation population individual A(I) μ The kth compound in (I), represents the μth individual A in the population μ The first compound in (I), and According to the following three situations, η (I) and A μ (I) Crossover: If there is an identical compound in the combination (a, b) and (k, l), the other two different compounds are combined to complete the crossover and obtain the individual after the crossover; If there are no identical compounds in the combinations (a, b) and (k, l), the compounds in one combination are randomly combined with the compounds in the other combination to complete the crossover and obtain the individuals after the crossover; If there are two identical compounds in the combination (a, b) and (k, l), no crossover is performed and they are directly used as individuals after crossover; Step 5.5, mutation: The random probability is used to determine whether the individual after the crossover will mutate. If it mutates, the mutation method is to randomly select a compound with a value of 1 in the individual after the crossover to change it to 0, and at the same time randomly select a compound with a value of 0 to change it to 1, so as to obtain the mutated individual; otherwise, the individual after the crossover is directly used as the mutated individual; Step 5.6: Repeat steps 5.4 and 5.5 until P crossover mutations are completed, and obtain the population A′(I) after crossover mutation; Step 5.7: After combining the I-generation population A(I) with the I-generation crossover population A′(I), calculate the non-dominated sorting and crowding distance between individuals, and select the first P individuals as the I+1 generation population individuals to determine if I+1>I max Is it true? If true, it means that the I max Generation population individuals and obtain the I max The compound combination represented by the individuals in the generation population is the optimal compound combination, otherwise, assign I+1 to I and return to step 5.4 to execute sequentially.

Images