Cancer marker screening method based on multi-task sample selection

Abstract

The invention discloses a cancer marker screening method based on multi-task sample selection, which comprises the following steps: step 1, acquiring a cancer gene expression data set, and carrying out data preprocessing; 2, calculating the instance importance degree of the sample to construct an auxiliary task data set; 3, initializing a main task population and an auxiliary task population based on the preprocessed data; 4, respectively calculating an optimization target value of each individual in the two task populations, and adopting different evolutionary strategies to generate filial generations; 5, performing knowledge transfer between two tasks every k generations; the auxiliary tasks are reconstructed every r generations; step 6, iterative optimization is carried out, and an optimal sample subset is selected; and 7, inputting the sample subset into a feature selection algorithm to obtain a final gene expression combination, namely screening out suspected cancer markers. According to the method, the optimal sample subset can be selected, so that the time efficiency of cancer marker screening is improved, and better cancer markers are screened out.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics and machine learning, and in particular to a cancer marker screening method based on multi-task sample selection. Background Art

[0002] In recent years, with the advancement of high-throughput sequencing technology, RNA cancer gene expression profiles have become an important resource for revealing disease mechanisms and screening new markers. Changes in RNA expression levels not only reflect the dynamic regulation of gene activity, but are also closely related to cell physiological state and disease progression. By analyzing the close connection between genes and cancer, exploring the most representative and predictive gene groups from massive RNA expression data as potential cancer markers to enhance predictive ability and clinical practicality has become a hot topic in current research.

[0003] Existing methods for screening potential cancer markers from RNA cancer gene expression profiles using feature selection algorithms have obvious shortcomings.

[0004] 1. Current screening methods do not take the quality of the data set into consideration. Technical limitations of sequencing instruments and incomplete sequencing chemistry can generate noise and erroneous data, which can affect the screening results of cancer markers.

[0005] 2. Each sample contains tens of thousands of genes, and there are a large number of redundant samples in the data set. Processing these data will consume a lot of extra time and affect the screening efficiency of cancer markers. Summary of the invention

[0006] The present invention aims to solve the deficiencies of the above-mentioned prior art and proposes a cancer marker screening method based on multi-task sample selection, so as to first reduce and optimize the sample data set, thereby improving the effect and efficiency of cancer marker screening.

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

[0008] The cancer marker screening method based on multi-task sample selection of the present invention is characterized in that it comprises the following steps:

[0009] Step 1: Obtain cancer gene expression dataset {(x w ,y w )|w=1,2,…,W}, where (x w ,y w ) represents the w-th cancer gene expression sample, x w is the attribute feature set of the wth sample, and represents the dth attribute feature of the wth sample, d∈[1,D], D is the total number of attribute features of cancer gene expression samples, y w represents the category label of the wth sample, y w ∈[1,C], C represents the total number of categories; w∈[1,W], W represents the total number of cancer gene expression samples;

[0010] Step 2: Construct auxiliary task dataset;

[0011] Step 2.1, from {(x w ,y w )|w=1,2,…,W}, and calculate the average value of the attribute features in all samples of the cth category as the center vector center of the cth category. c ; Calculate the w-th cancer gene expression sample and center c After the Euclidean distance is normalized, the importance of the w-th sample is obtained. w , thus obtaining the importance of all samples imp = {imp w |w=1,2,…,W};

[0012] Step 2.2, define the auxiliary task dataset algebra as t, and initialize t = 1;

[0013] Define the tth generation logo vector in, represents the wth sign of the tth generation sign vector;

[0014] Construct random probability vector pos={pos w |w=1,2,…,W}; where pos w represents the wth random number;

[0015] If pos w <imp w , then let Indicates retaining the w-th cancer gene expression sample (x w ,y w ), otherwise, let Indicates deleting the w-th cancer gene expression sample (x w ,y w ), thereby constructing the t-generation cancer gene expression auxiliary dataset in, represents the u-th cancer gene expression auxiliary sample of the t-th generation; represents the attribute feature set of the u-th auxiliary sample of the t-th generation, represents the category label of the u-th auxiliary sample of the t-th generation, U tis the total number of auxiliary samples of cancer gene expression in the tth generation;

[0016] Step 3: Construct the main population and auxiliary population;

[0017] Step 3.1, define the current algebra as g, and initialize g = 1; define the maximum algebra as g max ;

[0018] Step 3.2: Define and randomly initialize the g-th generation main population as in, Indicates P g The nth principal individual in express The wth major chromosome in express The w-th cancer gene expression sample (x w ,y w ),like express Select the w-th cancer gene expression sample (x w ,y w );

[0019] Step 3.3: Define and randomly initialize the g-th generation auxiliary population as in, Indicates Q g The nth auxiliary individual in , and express The u-th auxiliary chromosome in express Do not select the u-th cancer gene expression auxiliary sample in like express Select the u-th cancer gene expression auxiliary sample

[0020] Step 4: If g is not a multiple of k and r, execute steps 4.1, 4.2 and 4.3 to calculate the optimal target values ​​of the individuals of the two populations of the gth generation for evolving the two populations of the g+1th generation; otherwise, execute step 5;

[0021] Step 4.1, calculate the individual optimization target value;

[0022] Step 4.1.1. The corresponding cancer gene expression sample set is input into the SVM classifier for training to obtain the nth cancer gene expression classification model of the gth generation. And calculate In {(x w ,yw )|w=1,2,…,W} and compression ratio Thus we get The optimization target value

[0023] Step 4.1.2: The corresponding cancer gene expression auxiliary sample set is input into the SVM classifier for training to obtain the nth cancer gene expression auxiliary classification model of the gth generation And calculate exist Classification accuracy on and compression ratio Thus we get The optimization target value

[0024] Step 4.2: Evolve the g-th generation main population P g , and get the main population P of the g+1th generation g+1 , and record the main population of the frontier surface of the g+1th generation in, represents the fth frontier master individual of the g+1th generation, and F represents the size of the frontier master population;

[0025] Step 4.3: Evolve the g-th generation auxiliary population Q g , and obtain the auxiliary population Q of the g+1th generation g+1 , and record the auxiliary population of the frontier surface of the g+1th generation in, represents the e-th frontier auxiliary individual in the g+1th generation, and E represents the size of the frontier auxiliary population;

[0026] Step 5: If g is a multiple of k and g is not a multiple of r, execute the main population P g and auxiliary population Q g The knowledge transfer between them is used to obtain the g+1th generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation

[0027] If g is a multiple of r, evolve the g+1th generation main population and obtain the g+1th generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation

[0028] Step 6: When g>g max When the gth max The main population P gmax The individual with the largest sum of optimization target values ​​is the gth max Best master individual like the w-th main chromosome in Select the w-th cancer gene expression sample (x w , y w ), otherwise, do not select the w-th cancer gene expression sample (x w , y w ), so as to construct the optimal subset of cancer gene expression samples;

[0029] Step 7: Input the optimal subset of cancer gene expression samples into a feature selection algorithm for processing, and screen out the final cancer gene expression combination, that is, the screening result of cancer markers.

[0030] The characteristics of a cancer marker screening method based on multi-task sample selection according to the present invention are that the step 4.2 includes the following steps:

[0031] Step 4.2.1: When g = 1, randomly select the a-th main individual from P g and the b-th main individual and and

[0032] Construct the g-th generation crossover probability vector wherein, represents the w-th crossover random number of the g-th generation;

[0033] Define the crossover probability as pc. If randc w g < pc, then swap and ; otherwise, do not swap;

[0034] Construct the g-th generation a-th mutation probability vector wherein, randm_a w g represents the w-th mutation random number of the a-th main individual of the g-th generation ;

[0035] Define the mutation probability as pm. If randm_a w g < pm, then flip the w-th main chromosome of otherwise, do not flip;

[0036] Construct the g-th generation b-th mutation probability vector wherein, randm_b w g represents the w-th random number of the b-th main individual of the g-th generation ;

[0037] ​If randm_b w g <pm, then flip the w-th main chromosome of Otherwise, do not flip;

[0038] Add to the g-th generation updated main population P′ g ;

[0039] Step 4.2.2, Execute according to the process of Step 4.2.1 times, so as to obtain the g-th generation updated main population P′ g , and after merging with P g , select the top N individuals from the merged population through non-dominated sorting and elitist selection strategy to form the (g + 1)-th generation main population P g+1 , and record the (g + 1)-th generation front main population Wherein, represents the f-th front main individual of the (g + 1)-th generation, and F represents the scale of the front main population;

[0040] Step 4.2.3, When g ≠ 1, randomly select the f-th front main individual from the g-th generation front main population PF g as the guiding individual of ;

[0041] Construct the learning probability vector of the g-th generation Wherein, randl w g represents the w-th random number of the g-th generation;

[0042] Define the learning probability as pl, if randl w g <pl, then the w-th main chromosome in is assigned to Otherwise, do not assign;

[0043] Add to the g-th generation updated main population P′ g ;

[0044] Step 4.2.4, After merging P g and P′ g , select the top N main individuals from the merged population through non-dominated sorting and elitist selection strategy to form the (g + 1)-th generation main population P g+1 , and record the (g + 1)-th generation front main population

[0045] The step 4.3 comprises the following steps:

[0046] Step 4.3.1. From Q g Randomly select the ath auxiliary individual The bth auxiliary individual and

[0047] Construct the g-th generation crossover probability vector in, represents the u-th crossover random number of the g-th generation;

[0048] Define the crossover probability pc′ if Then and Exchange; otherwise, do not exchange;

[0049] Construct the a-th mutation probability vector of the g-th generation in, represents the ath auxiliary individual of the gth generation The uth mutation random number;

[0050] Define the mutation probability as pm′, if Then flip The uth auxiliary chromosome Otherwise, do not flip;

[0051] Construct the b-th mutation probability vector of the g-th generation in, represents the bth auxiliary individual of the gth generation The uth mutation random number;

[0052] like Then flip The uth auxiliary chromosome Otherwise, do not flip;

[0053] Will Join the gth generation to update the auxiliary population Q′ g middle;

[0054] Step 4.3.2: Follow the process of step 4.3.1 times, thus obtaining the g-th generation updated auxiliary population Q′ g , and with Q g After the merger, the top N individuals are selected from the merged population through non-dominated sorting and elite selection strategies to form the g+1 generation auxiliary population Q g+1 , and record the auxiliary population of the frontier surface of the g+1th generation in, represents the e-th frontier auxiliary individual of the g+1th generation, and E represents the size of the frontier auxiliary population.

[0055] In step 5, if g is a multiple of k and g is not a multiple of r, then execute step 5.1 and step 5.2;

[0056] If g is a multiple of r, execute steps 5.2 and 5.3:

[0057] Step 5.1: Calculate the main population P g Each main individual The sum of the optimization target values And P g The main individual corresponding to the maximum value of the sum of the optimization target values ​​is taken as the best main individual of the gth generation

[0058] merge and Q g Finally, through the non-dominated sorting and elite selection strategy, the top N individuals are selected from the merged population to form the g+1 generation auxiliary population Q g+1 , and record the auxiliary population of the frontier surface of the g+1th generation

[0059] according to and S t Update the t+1th generation flag vector in, represents the wth flag of the t+1th generation flag vector;

[0060] like Keep the w-th cancer gene expression sample (x w ,y w ), otherwise, delete the w-th cancer gene expression sample (x w ,y w ), thereby constructing the t+1 generation cancer gene expression auxiliary dataset in, represents the u-th cancer gene expression auxiliary sample of the t+1th generation; represents the attribute feature set of the u-th auxiliary sample of the t+1th generation, represents the category label of the u-th auxiliary sample in the t+1th generation; U t+1 is the total number of auxiliary samples for cancer gene expression;

[0061] Step 5.2: Merge QF g , P g , and through non-dominated sorting and elite selection strategies, select the top N main individuals from the merged population to form the g+1 generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation

[0062] Step 5.3: According to PF g and S t Update the t+1th generation flag vector

[0063] like Keep the w-th cancer gene expression sample (x w ,y w ), otherwise, delete the w-th cancer gene expression sample (x w ,y w ), thereby constructing the t+1 generation cancer gene expression auxiliary dataset

[0064] use Randomly construct the g+1th generation auxiliary population Q g+1 .

[0065] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the cancer marker screening method, and the processor is configured to execute the program stored in the memory.

[0066] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the cancer marker screening method when executed by a processor.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] 1. The method proposed in the present invention is based on multi-objective optimization. The multi-objective optimization algorithm can significantly improve the efficiency of sample selection. Two objective functions are constructed by the classification accuracy of the sample subset and the compression rate of the sample data set, and the problem is converted into an intuitive mathematical model. The sample subset that performs well in both objectives is selected for the screening of cancer markers.

[0069] 2. The present invention designs a multi-task based sample selection method, which constructs a simplified auxiliary task dataset through the importance of cancer gene expression sample data. Two task populations are constructed to evolve independently, and knowledge transfer between task populations is performed at a fixed number of generations. The main task population can perform a detailed search on the entire cancer gene expression sample dataset, while the auxiliary tasks can speed up the efficiency and effect of the main task population search. In order to select a sample subset that is more representative of the entire cancer gene expression sample dataset.

[0070] 3. The present invention proposes a method for optimizing the screening effect and efficiency of cancer markers from the perspective of improving the quality of sample data sets. Different from existing studies, this method improves the quality of sample data sets by removing noise and erroneous samples in the sample data sets, screening and deleting a large number of redundant samples. This improves the screening effect of cancer markers while greatly reducing the time consumed for screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is a flow chart of the method of the present invention;

[0072] Figure 2 This is the cancer marker finally screened by the present invention. DETAILED DESCRIPTION

[0073] In this embodiment, a cancer marker screening method based on multi-task sample selection is provided. Figure 1 As shown, the following steps are included:

[0074] Step 1: Obtain cancer gene expression dataset {(x w ,y w )|w=1,2,…,W}, where (x w ,y w ) represents the w-th cancer gene expression sample, x w is the attribute feature set of the wth sample, and represents the dth attribute feature of the wth sample, d∈[1,D], D is the total number of attribute features of cancer gene expression samples, y w represents the category label of the wth sample, y w ∈[1,C], C represents the total number of categories; w∈[1,W], W represents the total number of cancer gene expression samples;

[0075] In this embodiment, the cancer gene expression dataset uses the gene expression cancer RNA-Seq dataset as the standard dataset. In the gene expression cancer RNA-Seq dataset, there are a total of 801 data, corresponding to 801 independent cancer gene expression data biological samples collected.

[0076] Step 2: Construct auxiliary task dataset;

[0077] Step 2.1, from {(x w ,y w )|w=1,2,…,W}, and calculate the average value of the attribute features in all samples of the cth category as the center vector center of the cth category. c; Calculate the w-th cancer gene expression sample and center c After the Euclidean distance is normalized, the importance of the w-th sample is obtained. w , thus obtaining the importance of all samples imp = {imp w |w=1,2,…,W};

[0078] Step 2.2, define the auxiliary task dataset algebra as t, and initialize t = 1;

[0079] Step 2.2, define the auxiliary task dataset algebra as t, and initialize t = 1;

[0080] Define the tth generation logo vector in, represents the wth sign of the tth generation sign vector;

[0081] Construct random probability vector pos={pos w |w=1,2,…,W}; where pos w represents the wth random number;

[0082] If pos w <imp w , then let Indicates retaining the w-th cancer gene expression sample (x w ,y w ), otherwise, let Indicates deleting the w-th cancer gene expression sample (x w ,y w ), thereby constructing the t-generation cancer gene expression auxiliary dataset in, represents the u-th cancer gene expression auxiliary sample of the t-th generation; represents the attribute feature set of the u-th auxiliary sample of the t-th generation, represents the category label of the u-th auxiliary sample of the t-th generation, U t is the total number of auxiliary samples of cancer gene expression in the tth generation;

[0083] Step 3: Construct the main population and auxiliary population;

[0084] Step 3.1, define the current algebra as g, and initialize g = 1; define the maximum algebra as g max ;

[0085] Step 3.2: Define and randomly initialize the g-th generation main population as in, Indicates P g The nth principal individual in express The wth major chromosome in express The w-th cancer gene expression sample (x w ,y w ),like express Select the w-th cancer gene expression sample (x w ,y w );

[0086] Step 3.3: Define and randomly initialize the g-th generation auxiliary population as in, Indicates Q g The nth auxiliary individual in , and express The u-th auxiliary chromosome in express Do not select the u-th cancer gene expression auxiliary sample in like express Select the u-th cancer gene expression auxiliary sample

[0087] Step 4: Assign 5 to k and 20 to r. If g is not a multiple of k and r, execute steps 4.1, 4.2, and 4.3 to calculate the optimal target values ​​of the individuals of the two populations in the gth generation, which are used to evolve the two populations in the g+1th generation.

[0088] Step 4.1, calculate the individual optimization target value;

[0089] Step 4.1.1. The corresponding cancer gene expression sample set is input into the SVM classifier for training to obtain the nth cancer gene expression classification model of the gth generation. And calculate In {(x w ,y w )|w=1,2,…,W} and compression ratio Thus we get The optimization target value

[0090] Step 4.1.2: The corresponding cancer gene expression auxiliary sample set is input into the SVM classifier for training to obtain the nth cancer gene expression auxiliary classification model of the gth generation And calculate exist Classification accuracy on and compression ratio Thus, the optimized target value

[0091] Step 4.2: Evolve the g-th generation of the main population P g ;

[0092] Step 4.2.1: When g = 1, randomly select the a-th main individual g from P and the b-th main individual and

[0093] Construct the g-th generation of the crossover probability vector where randc w g represents the w-th crossover random number of the g-th generation;

[0094] Define the crossover probability as pc, assign 0.9 to pc. If randc w g < pc, then swap and ; otherwise, do not swap;

[0095] Construct the w-th mutation probability vector of the a-th individual in the g-th generation where randm_a w g represents the w-th mutation random number of the a-th main individual in the g-th generation;

[0096] Define the mutation probability as pm, assign 0.02 to pm. If then flip the w-th main chromosome of otherwise, do not flip;

[0097] Construct the w-th mutation probability vector of the b-th individual in the g-th generation where randm_b w g represents the w-th random number of the b-th main individual in the g-th generation;

[0098] If randm_b w g < pm, then flip the w-th main chromosome of otherwise, do not flip;

[0099] Add to the updated main population P' g in the g-th generation;

[0100] Step 4.2.2: Execute according to the process of Step 4.2.1 times to obtain the updated main population P′ of the g-th generation g , and after merging with P g , select the top N individuals from the merged population through non-dominated sorting and elitist selection strategies to form the main population P of the (g + 1)-th generation g+1 , and record the main population of the (g + 1)-th generation front wherein represents the f-th main individual of the (g + 1)-th generation front, and F represents the scale of the main population of the front

[0101] Step 4.2.3: When g ≠ 1, randomly select the f-th main individual of the (g)-th generation front from the main population PF of the (g)-th generation g as the guiding individual of as ;

[0102] Construct the learning probability vector of the g-th generation wherein randl w g represents the w-th random number of the g-th generation

[0103] Define the learning probability as pl, assign 0.7 to pl, if randl w g < pl, then the w-th main chromosome in is assigned to otherwise, no assignment is made

[0104] Add to the updated main population P′ of the g-th generation g ;

[0105] Step 4.2.4: After merging P g and P′ g , select the top N main individuals from the merged population through non-dominated sorting and elitist selection strategies to form the main population P of the (g + 1)-th generation g+1 , and record the main population of the (g + 1)-th generation front

[0106] Step 4.3: Evolve the auxiliary population Q of the g-th generation g ;

[0107] Step 4.3.1: Randomly select the a-th auxiliary individual g from Q the b-th auxiliary individual and

[0108] Construct the crossover probability vector of the g-th generation in, represents the u-th crossover random number of the g-th generation;

[0109] Define the crossover probability pc′ and assign 0.9 to pc′. If Then and Exchange; otherwise, do not exchange;

[0110] Construct the a-th mutation probability vector of the g-th generation in, represents the ath auxiliary individual of the gth generation The uth mutation random number;

[0111] Define the mutation probability as pm', assign 0.02 to pm', if Then flip The uth auxiliary chromosome Otherwise, do not flip;

[0112] Construct the b-th mutation probability vector of the g-th generation in, represents the bth auxiliary individual of the gth generation The uth mutation random number;

[0113] like Then flip The uth auxiliary chromosome Otherwise, do not flip;

[0114] Will Join the gth generation to update the auxiliary population Q′ g middle;

[0115] Step 4.3.2: Follow the process of step 4.3.1 times, thus obtaining the g-th generation updated auxiliary population Q′ g , and with Q g After the merger, the top N individuals are selected from the merged population through non-dominated sorting and elite selection strategies to form the g+1 generation auxiliary population Q g+1 , and record the auxiliary population of the frontier surface of the g+1th generation in, represents the e-th frontier auxiliary individual in the g+1th generation, and E represents the size of the frontier auxiliary population;

[0116] Step 5: If g is a multiple of k and g is not a multiple of r, execute the main population P of step 5.1 and step 5.2. g and auxiliary population Q g knowledge transfer between

[0117] If g is a multiple of r, execute step 5.2 to evolve the g+1th generation main population, and execute step 5.3 to construct the t+1th generation auxiliary population;

[0118] Step 5.1: Calculate the main population P g Each main individual The sum of the optimization target values And P g The main individual corresponding to the maximum value of the sum of the optimization target values ​​is taken as the best main individual of the gth generation

[0119] merge and Q g Finally, through the non-dominated sorting and elite selection strategy, the top N individuals are selected from the merged population to form the g+1 generation auxiliary population Q g+1 , and record the auxiliary population of the frontier surface of the g+1th generation

[0120] according to and S t Update the t+1th generation flag vector in, represents the wth flag of the t+1th generation flag vector;

[0121] like Keep the w-th cancer gene expression sample, otherwise delete the w-th cancer gene expression sample, so as to construct the t+1-th generation cancer gene expression auxiliary dataset in, represents the u-th cancer gene expression auxiliary sample of the t+1th generation; represents the attribute feature set of the u-th auxiliary sample of the t+1th generation, represents the category label of the u-th auxiliary sample in the t+1th generation; U t+1 is the total number of auxiliary samples for cancer gene expression;

[0122] Step 5.2: Merge QF g , P g , through non-dominated sorting and elite selection strategies, the top N main individuals are selected from the merged population to form the g+1 generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation

[0123] Step 5.3: According to PF g and S t Update the t+1th generation flag vector

[0124] like Keep the w-th cancer gene expression sample, otherwise delete the w-th cancer gene expression sample, so as to construct the t+1-th generation cancer gene expression auxiliary dataset

[0125] use Randomly construct the g+1th generation auxiliary population Q g+1 ;

[0126] Step 6: When g>g max When the gth max The main population P gmax The individual with the largest sum of optimization target values ​​is the gth max Best master individual like The wth major chromosome in Select the wth cancer gene expression sample (x w ,y w ), otherwise, the w-th cancer gene expression sample (x w ,y w ), thereby constructing the best subset of cancer gene expression samples;

[0127] As shown in Table 1, the classification accuracy and compression rate of the best cancer gene expression sample subset are calculated. Compared with the original data set, it shows that the best cancer gene expression sample subset removes redundant and erroneous samples in the original sample set;

[0128] Table 1

[0129]

[0130] Step 7: Input the best cancer gene expression sample subset into the feature selection algorithm for processing to screen out the final cancer gene expression combination, i.e., the screening result of the cancer marker.

[0131] like Figure 2 As shown, the binary chromosome sequence obtained by the final feature selection is (1, 0, 1, ..., 0, 0, 1), which corresponds to the set composed of the 1st, 3rd, ..., Lth gene expression in the cancer gene expression data set. This set is a potential cancer marker. The comparison between the method of the present invention and the traditional feature selection screening method on the gene expression cancer RNA-Seq data set is shown in Table 2;

[0132] Table 2

[0133]

[0134] As can be seen from Table 2: Compared with the traditional feature selection screening method, the cancer markers screened out by the method of the present invention using the cancer gene expression data set have a higher classification accuracy for the entire data set, which means that the gene expression combination corresponding to the cancer marker has a closer connection with cancer, that is, the screening effect is better. At the same time, the screening time of the method of the present invention is shorter than that of the traditional feature selection screening method. At present, with the development of high-throughput sequencing technology, cancer gene expression data is constantly increasing. The method of the present invention can effectively improve the effect and efficiency of screening cancer markers using cancer gene expression data.

Claims

1. A cancer marker screening method based on multi-task sample selection, characterized in that: The following steps are involved: Step 1: Obtain cancer gene expression dataset {(x w ,y w )|w=1,2,…,W}, where (x w ,y w ) represents the w-th cancer gene expression sample, x w is the attribute feature set of the wth sample, and represents the dth attribute feature of the wth sample, d∈[1,D], D is the total number of attribute features of cancer gene expression samples, y w represents the category label of the wth sample, y w ∈[1,C], C represents the total number of categories; w∈[1,W], W represents the total number of cancer gene expression samples; Step 2: Construct auxiliary task dataset; Step 2.1, from {(x w ,y w )|w=1,2,…,W}, and calculate the average value of the attribute features in all samples of the cth category as the center vector center of the cth category. c ; Calculate the w-th cancer gene expression sample and center c After the Euclidean distance is normalized, the importance of the w-th sample is obtained. w , thus obtaining the importance of all samples imp = {imp w |w=1,2,…,W}; Step 2.2, define the auxiliary task dataset algebra as t, and initialize t = 1; Define the tth generation logo vector in, represents the wth sign of the tth generation sign vector; Construct random probability vector pos={pos w |w=1,2,…,W}; where pos w represents the wth random number; If pos w <imp w , then let Indicates retaining the w-th cancer gene expression sample (x w ,y w ), otherwise, let Indicates deleting the w-th cancer gene expression sample (x w ,y w ), thereby constructing the t-generation cancer gene expression auxiliary dataset in, represents the u-th cancer gene expression auxiliary sample of the t-th generation; represents the attribute feature set of the u-th auxiliary sample of the t-th generation, represents the category label of the u-th auxiliary sample of the t-th generation, U t is the total number of auxiliary samples of cancer gene expression in the tth generation; Step 3: Construct the main population and auxiliary population; Step 3.1, define the current algebra as g, and initialize g = 1; define the maximum algebra as g max ; Step 3.2: Define and randomly initialize the g-th generation main population as in, Indicates P g The nth principal individual in express The wth major chromosome in express The w-th cancer gene expression sample (x w ,y w ),like express Select the w-th cancer gene expression sample (x w ,y w ); Step 3.3: Define and randomly initialize the g-th generation auxiliary population as in, Indicates Q g The nth auxiliary individual in , and express The u-th auxiliary chromosome in express Do not select the u-th cancer gene expression auxiliary sample in like express Select the u-th cancer gene expression auxiliary sample Step 4: If g is not a multiple of k and r, execute steps 4.1, 4.2 and 4.3 to calculate the optimal target values ​​of the individuals of the two populations of the gth generation for evolving the two populations of the g+1th generation; otherwise, execute step 5; Step 4.1, calculate the individual optimization target value; Step 4.1.

1. The corresponding cancer gene expression sample set is input into the SVM classifier for training to obtain the nth cancer gene expression classification model of the gth generation And calculate In {(x w ,y w )|w=1,2,…,W} and compression ratio Thus we get The optimization target value Step 4.1.2: The corresponding cancer gene expression auxiliary sample set is input into the SVM classifier for training to obtain the nth cancer gene expression auxiliary classification model of the gth generation And calculate exist Classification accuracy on and compression ratio Thus we get The optimization target value Step 4.2: Evolve the g-th generation main population P g , and get the main population P of the g+1th generation g+1 , and record the main population of the frontier surface of the g+1th generation in, represents the fth frontier master individual of the g+1th generation, and F represents the size of the frontier master population; Step 4.3: Evolve the g-th generation auxiliary population Q g , and obtain the g+1th generation auxiliary population Q g+1 , and record the auxiliary population of the frontier surface of the g+1th generation in, represents the e-th frontier auxiliary individual in the g+1th generation, and E represents the size of the frontier auxiliary population; Step 5: If g is a multiple of k and g is not a multiple of r, execute the main population P g and auxiliary population Q g The knowledge transfer between them is used to obtain the g+1th generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation If g is a multiple of r, evolve the g+1th generation main population and obtain the g+1th generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation Step 6: When g>g max When the gth max The main population P gmax The individual with the largest sum of optimization target values ​​is the gth max Best master individual like The wth major chromosome in Select the wth cancer gene expression sample (x w ,y w ), otherwise, the w-th cancer gene expression sample (x w ,y w ), thereby constructing the best subset of cancer gene expression samples; Step 7: Input the best cancer gene expression sample subset into the feature selection algorithm for processing to screen out the final cancer gene expression combination, i.e., the screening result of the cancer marker.

2. The method for screening cancer markers based on multi-task sample selection according to claim 1, characterized in that: The step 4.2 comprises the following steps: Step 4.2.1: When g = 1, from P g Randomly select the ath master individual and the bth principal individual and Construct the g-th generation crossover probability vector Among them, randc w g represents the wth crossover random number of the gth generation; Define the crossover probability as pc. If randc w g < pc, then swap and ; otherwise, do not swap. Construct the a-th mutation probability vector of the g-th generation Among them, randm_a w g represents the ath master individual of the gth generation The w-th mutation random number; Define the mutation probability as pm. If randm_a w g < pm, then flip the w-th main chromosome of Otherwise, do not flip; Construct the b-th mutation probability vector of the g-th generation Among them, randm_b w g represents the bth primary individual of the gth generation The wth random number of ; If randm_b w g <pm, then flip the w-th main chromosome of Otherwise, do not flip; Will Join the gth generation to update the main population P′ g middle; Step 4.2.2: Follow the process of step 4.2.1 times, thus obtaining the g-th generation updated main population P′ g , and with P g After the merger, the top N individuals are selected from the merged population through non-dominated sorting and elite selection strategies to form the g+1th generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation in, represents the fth frontier master individual of the g+1th generation, and F represents the size of the frontier master population; Step 4.2.3: When g≠1, the main population PF of the g-th generation frontier g Randomly select the fth frontier master individual As guiding individuals; Construct the learning probability vector of the gth generation Among them, randl w g represents the wth random number of the gth generation; Define the learning probability as pl. If randl w g < pl, then the w-th main chromosome in is assigned to Otherwise, no assignment is made; Will Join the gth generation to update the main population P′ g middle; Step 4.2.4: P g and P′ g After the merger, the top N main individuals are selected from the merged population through non-dominated sorting and elite selection strategies to form the g+1 generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation 3. The method for screening cancer markers based on multi-task sample selection according to claim 5, characterized in that: The step 4.3 comprises the following steps: Step 4.3.

1. From Q g Randomly select the ath auxiliary individual The bth auxiliary individual and Construct the g-th generation crossover probability vector in, represents the u-th crossover random number of the g-th generation; Define the crossover probability pc′ if Then and Exchange; otherwise, do not exchange; Construct the a-th mutation probability vector of the g-th generation in, represents the ath auxiliary individual of the gth generation The uth mutation random number; Define the mutation probability as pm′, if Then flip The uth auxiliary chromosome Otherwise, do not flip; Construct the b-th mutation probability vector of the g-th generation in, represents the bth auxiliary individual of the gth generation The uth mutation random number; like Then flip The uth auxiliary chromosome Otherwise, do not flip; Will Join the gth generation to update the auxiliary population Q′ g middle; Step 4.3.2: Follow the process of step 4.3.1 times, thus obtaining the g-th generation updated auxiliary population Q′ g , and with Q g After the merger, the top N individuals are selected from the merged population through non-dominated sorting and elite selection strategies to form the g+1 generation auxiliary population Q g+1 , and record the auxiliary population of the frontier surface of the g+1th generation in, represents the e-th frontier auxiliary individual of the g+1th generation, and E represents the size of the frontier auxiliary population.

4. The method for screening cancer markers based on multi-task sample selection according to claim 3, characterized in that: In step 5, if g is a multiple of k and g is not a multiple of r, then execute step 5.1 and step 5.2; if g is a multiple of r, then execute step 5.2 and step 5.3: Step 5.1: Calculate the main population P g Each main individual The sum of the optimization target values And P g The main individual corresponding to the maximum value of the sum of the optimization target values ​​is taken as the best main individual of the gth generation merge and Q g Finally, through the non-dominated sorting and elite selection strategy, the top N individuals are selected from the merged population to form the g+1 generation auxiliary population Q g+1 , and record the auxiliary population of the frontier surface of the g+1th generation according to and S t Update the t+1th generation flag vector in, represents the wth flag of the t+1th generation flag vector; like Keep the w-th cancer gene expression sample (x w ,y w ), otherwise, delete the w-th cancer gene expression sample (x w ,y w ), thereby constructing the t+1 generation cancer gene expression auxiliary dataset in, represents the u-th cancer gene expression auxiliary sample of the t+1th generation; represents the attribute feature set of the u-th auxiliary sample of the t+1th generation, represents the category label of the u-th auxiliary sample in the t+1th generation; U t+1 is the total number of auxiliary samples for cancer gene expression; Step 5.2: Merge QF g , P g , and through the non-dominated sorting and elite selection strategy, select the top N main individuals from the merged population to form the g+1 generation main population P g+1 , and record the main population of the frontier surface of the g+1th generation Step 5.3: According to PF g and S t Update the t+1th generation flag vector like Keep the w-th cancer gene expression sample (x w ,y w ), otherwise, delete the w-th cancer gene expression sample (x w ,y w ), thereby constructing the t+1 generation cancer gene expression auxiliary dataset use Randomly construct the g+1th generation auxiliary population Q g+1 .

5. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the cancer marker screening method according to any one of claims 1 to 4, and the processor is configured to execute the program stored in the memory.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the cancer marker screening method according to any one of claims 1 to 4 are performed.

Images