Epidemic disease detection device and method

Abstract

The present invention provides an epidemic disease detection device, which is characterized in that it is used to detect diseased individuals from a population to be tested whose infection rate is below a predetermined infection threshold, and comprises: a sample vector setting unit for testing, which sets a sample vector for testing for n samples of the individuals to be tested obtained by sampling from the population to be tested; a mixing matrix generating unit, which generates a mixing matrix; a mixed load obtaining unit, which obtains the viral load of each mixed sample obtained by mixing the samples to be tested according to the mixing matrix as a mixed load; a conversion unit, which converts the mixing matrix and the mixed load respectively according to the conversion matrix; and a diseased individual determining unit, which uses a predetermined multiple orthogonal matching pursuit algorithm to process the converted mixing matrix and the viral load to obtain the position of non-zero elements in the sample vector to be tested, and determines the diseased individual from the individuals to be tested based on the position of the non-zero element.

Description

Technical Field

[0001] The present invention belongs to the field of sample detection, and in particular relates to an epidemic disease detection device and method. Background Art

[0002] When some epidemic diseases break out, in order to effectively detect the disease situation in certain areas, it is often necessary to conduct large-scale testing on the population, so as to identify the sick individuals in the population, which will help to subsequently perform treatment or isolation on these sick individuals.

[0003] However, large-scale population testing is not only costly, but also time-consuming and labor-intensive. Generally speaking, in order to ensure the accuracy of testing of sick individuals, there are two main methods for large-scale screening of sick individuals: one is large-scale sampling and testing one by one; the other is to group and mix the samples for testing. If the virus is present, these suspected samples are tested one by one. If the virus is not present, all samples in this group are judged to be negative.

[0004] In practical applications, the number of tests required by the above two methods is very large. Generally, O(n / s) tests are required to detect positive cases among n individuals. In the formula, s is the number of groups in the group sampling, and O(·) represents the order of magnitude of the independent variable. Summary of the invention

[0005] In order to solve the above problems, a device and method for detecting epidemic diseases with a very small number of detection times are provided. The present invention adopts the following technical solutions:

[0006] The present invention provides an epidemic disease detection device, which is characterized in that it is used to detect diseased individuals from a population to be tested whose infection rate is below a predetermined infection threshold, and comprises: a sample vector setting unit to be tested, for setting the viral loads of the n samples to be tested as a sample vector to be tested with a length of n for the samples to be tested of n individuals to be tested obtained by sampling from the population to be tested; a mixing matrix generating unit to generate a mixing matrix according to the number n of samples to be tested, a predetermined number of mixing groups, a maximum number of mixing allowed for each group, and a sampling probability; a mixed load acquiring unit to acquire the viral loads of each mixed sample obtained by mixing the samples to be tested according to the mixing matrix as a mixed load; a conversion unit to generate a conversion matrix according to the number of mixing groups and the sampling probability, and to convert the mixing matrix and the mixed load according to the conversion matrix to obtain corresponding converted mixing matrices and converted viral loads; and a diseased individual determining unit to process the converted mixing matrix and the converted viral load using a predetermined multiple orthogonal least squares algorithm to obtain the positions of non-zero elements in the sample vector to be tested, and to determine the diseased individual from the individuals to be tested based on the positions of the non-zero elements.

[0007] The epidemic disease detection device provided by the present invention may also have such a technical feature, wherein the mixing matrix generation unit has: a random matrix generation unit, which generates a random matrix according to the number of mixing groups m, the maximum mixing number N and the sampling probability p, and the random matrix is ​​an m×n random Bernoulli matrix Φ, and each element φ in the random matrix ij Obeying the independent and identical distribution of Bernoulli(p), the random matrix meets the following conditions: The random Bernoulli matrix Φ satisfies row sparsity ‖φ i ‖0∈(0,N], column sparsity ‖φ j ‖0=mp, any two rows are different, any two columns are different, where i represents the i-th row of the random matrix, and j represents the j-th column of the random matrix; the dilution matrix generation unit generates the corresponding dilution matrix according to the random matrix: The mixing matrix generation unit generates a mixing matrix A←DΦ according to the random matrix and the dilution matrix. The element A in the mixing matrix A ij =0 means that the jth sample to be tested does not participate in the mixing of the i-th group of mixed samples, otherwise it participates.

[0008] The epidemic disease detection device provided by the present invention may also have such a technical feature, wherein the random Bernoulli matrix Φ is generated by rejection sampling or Gibbs sampling.

[0009] The epidemic disease detection device provided by the present invention may also have such a technical feature, wherein the conversion matrix satisfy:

[0010]

[0011] In the formula, m is the number of mixed groups, p is the sampling probability, and a j is the j-th row of the mixing matrix A, transforms the mixing matrix Ψ←TA, transforms the viral load y←Tz, where z is the mixed load.

[0012] The epidemic disease detection device provided by the present invention may also have such a technical feature, wherein the infection threshold is 0.1‰.

[0013] The present invention also provides a method for detecting an epidemic disease, which is characterized in that it is used to detect diseased individuals from a population to be tested whose infection rate is below a predetermined infection threshold, and includes: for n samples to be tested of the individuals to be tested obtained by sampling from the population to be tested, setting the viral loads of the n samples to be tested as a sample vector to be tested of length n; generating a mixing matrix according to the number n of samples to be tested, a predetermined number of mixing groups, a maximum number of mixing allowed for each group, and a sampling probability; obtaining the viral loads of each mixed sample obtained by mixing the samples to be tested according to the mixing matrix as a mixed load; generating a conversion matrix according to the number of mixing groups and the sampling probability, and converting the mixing matrix and the mixed load according to the conversion matrix to obtain corresponding converted mixing matrices and converted viral loads; and using a predetermined multiple orthogonal least squares algorithm to process the converted mixing matrix and the converted viral load to obtain the positions of non-zero elements in the sample vector to be tested, and determining diseased individuals from the individuals to be tested based on the positions of the non-zero elements.

[0014] Function and Effect of the Invention

[0015] According to the epidemic disease detection device and method of the present invention, since a mixing matrix is ​​generated according to the number of samples to be tested n, the predetermined number of mixing groups m, the maximum number of mixing allowed for each group, and the sampling probability, and the viral load of the mixed sample obtained by mixing the actual samples to be tested according to the mixing matrix is ​​obtained, the viral load and the mixing matrix are further converted and processed by the conversion matrix to obtain the sample vector to be tested indicating the viral load of each sample to be tested. Therefore, in this way, the viral load of each sample to be tested can be detected not directly by RT-PCR, but each sample to be tested can be mixed according to the predetermined mixing matrix and then the viral load in the mixture is detected, so the number of detections is reduced from n times to m times, thereby greatly reducing the detection cost consumed by large-scale virus sampling detection. In addition, unlike traditional mixed detection, the epidemic disease detection device and method of the present invention only needs to detect each mixed sample once to determine whether each sample to be tested is sick, while the traditional algorithm needs to sample and test all individuals participating in the group again after detecting that the mixture is positive. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a block diagram of an epidemic disease detection device according to an embodiment of the present invention;

[0017] Figure 2 is a schematic diagram of an algorithm flow chart for detection in an embodiment of the present invention; and

[0018] Figure 3 2 is a schematic diagram for comparing experimental results of an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the technical means, creative features, objectives and effects of the present invention easy to understand, the epidemic disease detection device and method of the present invention are specifically described below in conjunction with embodiments and drawings.

[0020] <Example>

[0021] The epidemic disease detection device of this embodiment is a computer for running a program corresponding to the epidemic disease detection method, which can detect the epidemic disease with a very high probability. Mixed samples can effectively identify all K diseased individuals among the n individuals to be tested.

[0022] Next, in this embodiment, taking the detection of the new coronavirus as an example, for a population of n individuals to be tested, nucleic acid samples are collected from each individual to be tested as samples to be tested (throat swab, nasal swab, sputum, etc.).

[0023] Figure 1 is a block diagram of an epidemic disease detection device according to an embodiment of the present invention, Figure 2 It is a schematic diagram of the algorithm flow of detection in an embodiment of the present invention.

[0024] like Figure 1 and Figure 2 As shown, the epidemic disease detection device 10 has a sample vector setting unit 11, a mixing matrix generating unit 12, a mixed load acquiring unit 13, a converting unit 14, a diseased individual determining unit 15 and a device controlling unit 16 for controlling the above-mentioned units.

[0025] The sample vector setting unit 11 is used to set the n samples to be tested (i.e. Figure 2 The viral load of the test sample (in the test sample) is set as a sample vector x with a length of n.

[0026] In this embodiment, the to-be-tested sample vector x is actually an unknown vector to be solved, and the value of x will be solved in the subsequent process to determine the value of each to-be-tested sample.

[0027] The mixing matrix generating unit 12 is used to generate a mixing matrix according to the number n of samples to be tested, the predetermined number m of mixing groups (m is much smaller than n), the maximum allowed number N of mixing per group, and the sampling probability p∈(0,1).

[0028] In this embodiment, the mixed matrix generation unit includes a random matrix generation unit, a dilution matrix generation unit, and a mixed matrix generation unit.

[0029] The random matrix generation unit is used to generate a random matrix. In this embodiment, the random matrix is ​​an m×n random Bernoulli matrix Φ, and each element φ in the random matrixij It must obey the independent and identical distribution of Bernoulli (p), and the random matrix must meet the predetermined conditions.

[0030] In this embodiment, the cyclic process of the random matrix generation unit generating the random matrix is ​​as follows:

[0031]

[0032] In addition, in this embodiment, since the probability of a random matrix that meets the requirements appearing is low, the random matrix generation unit of this embodiment uses rejection sampling or Gibbs sampling to generate a random Bernoulli matrix Φ that meets the conditions.

[0033] The dilution matrix generation unit generates the corresponding dilution matrix according to the random matrix: Where diag(·) represents a diagonal matrix.

[0034] The mixing matrix generating unit generates a mixing matrix A←DΦ according to the random matrix and the dilution matrix.

[0035] In this embodiment, since the viral loads of each group are not additive when mixed (i.e., z≠Φx), but are diluted according to the number of samples involved in the mixing (for example, if 10 samples to be tested in a group are mixed, the viral load of the obtained mixed sample should be the average of the viral loads of the 10 samples rather than the sum. "Viral load" can be understood as viral concentration), this embodiment converts the random matrix Φ to the mixing matrix A by left-multiplying the dilution matrix D.

[0036] Next, the n samples to be tested are mixed according to the mixing matrix A to obtain m groups of mixed samples. The element A in the mixing matrix A ij =0 means that the jth sample to be tested does not participate in the mixing of the ith group of mixed samples, otherwise it participates. The mixing process can be described in matrix form as z=Ax.

[0037] The mixed load acquisition unit 13 obtains the viral load of each mixed sample obtained by mixing the sample to be tested according to the mixing matrix as the mixed load.

[0038] In this embodiment, reverse transcription polymerase chain reaction (RT-PCR) is used to measure the cycle threshold value (CT value) of m groups of mixed samples, so that the viral load of each group of mixed samples is calculated according to the CT value as the mixed load, and the mixed load of the m groups of mixed samples is represented by a vector z with a length of m.

[0039] The conversion unit 14 generates a conversion matrix according to the number of mixing groups and the sampling probability, and converts the mixing matrix and the mixing load according to the conversion matrix to obtain the corresponding conversion mixing matrix and conversion viral load.

[0040] In this embodiment, the conversion matrix satisfy:

[0041]

[0042] In the formula, m is the number of mixed groups, p is the sampling probability, and a j is the j-th row of the mixing matrix A.

[0043] According to the above conversion matrix T, the mixing matrix and the mixed load are converted (left multiplied) respectively, that is, the conversion mixing matrix Ψ←TA and the conversion viral load y←Tz are obtained.

[0044] The above conversion mixing matrix Ψ and the conversion viral load y satisfy the relationship y=Ψx+v, where v is noise.

[0045] The diseased individual determination unit 15 uses a predetermined multiple orthogonal least squares algorithm (mOLS) to process the conversion mixing matrix Ψ and the conversion viral load y, so as to estimate the support set supp(x) of the sample vector x to be tested, that is, the position of the non-zero elements (that is, non-zero values) in the sample vector x to be tested. The sample to be tested corresponding to the position of the non-zero element is the positive sample. Therefore, the diseased individual determination unit 15 can determine the diseased individual from the individuals to be tested through the position of the non-zero element.

[0046] In this embodiment, the multiple orthogonal least squares algorithm is specifically to continuously reduce the residual energy ‖r by iteration. k ‖2(residual vector r k Initialized to the transformed viral load y, the superscript k indicates the kth iteration).

[0047] Specifically, in the kth iteration, the mOLS algorithm measures the unselected columns ψ of the matrix (here, the transformation mixing matrix Ψ) i ,i∈Ω\T k-1 Projection to the selected column ψ i ,i∈T k-1 In the orthogonal complement space of (Ω={1,2,…,n} is the full set), and l2 normalize each column after projection, and select the L normalized columns with the highest correlation with the residual, that is:

[0048]

[0049] in, is the projection operator, is the orthogonal complementary projection, ψ iis the i-th column of Ψ, and <,> represents the vector inner product. Then, the selected columns are added to the estimated support set, that is:

[0050] T k =T k-1 ∪I k

[0051] And perform least squares projection on the measurement vector according to the new estimated support set to obtain a new estimate:

[0052]

[0053] With the new residual:

[0054]

[0055] Repeat the iteration until convergence, and finally select the cK column with the largest coefficient to re-project the least squares to obtain the final estimate of the sample vector x to be tested. and its estimated support set (i.e. the estimated support set supp(x)).

[0056] In this embodiment, when satisfying And the signal-to-noise ratio (SNR) satisfies In the case of , the epidemic disease detection device 10 can recover a set containing supp(x) from y with a probability approaching 1, where c is an algorithm hyperparameter and K is the sparsity (i.e., the number of positive samples). In addition, the signal-to-noise ratio of 28dB in this embodiment takes into account the worst case. In practical applications, such a high signal-to-noise ratio is not required. The experimental results (see below) show that the algorithm is still effective under a lower signal-to-noise ratio.

[0057] Figure 3 2 is a schematic diagram for comparing experimental results of an embodiment of the present invention.

[0058] During the experiment, the infection rate in the population to be tested was 0.1‰. In order to characterize the sparsity, the Poisson distribution parameter λ=0.1‰ was used to simulate the disease conditions of 1,000 individuals to be tested, and each person's throat swab sample was divided into two parts and randomly mixed into 100 mixed samples for nucleic acid virus testing.

[0059] Assuming that the simulated virus load follows a log-normal distribution, After being processed by the above-mentioned epidemic disease detection device 10, the precision and recall are as follows: Figure 3As shown in the figure, the precision rate represents the proportion of real patients among the samples tested positive, while the recall rate represents the proportion of samples identified as positive among all patients. The higher the recall rate, the more effective the large-scale screening is. The parameter +Bin indicates whether the test result is an accurate viral load. If it does not contain the accurate viral load, only whether it is positive, it is RGB+Bin. The parameter c is a hyperparameter of mOLS. The larger the c, the larger the set of suspected positive cases, which reduces the precision and increases the recall rate. Figure 3 The experimental results in show that the epidemic disease detection device 10 of this embodiment not only has good detection performance, but also can reduce the sampling amount.

[0060] Example Function and Effect

[0061] According to the epidemic disease detection device and method provided by this embodiment, since a mixing matrix is ​​generated according to the number of samples to be tested n, the predetermined number of mixing groups m, the maximum number of mixing allowed for each group, and the sampling probability, and the viral load of the mixed sample obtained by mixing the actual samples to be tested according to the mixing matrix is ​​obtained, and the viral load and the mixing matrix are further converted and processed by the conversion matrix to obtain the sample vector to be tested indicating the viral load of each sample to be tested, therefore, in this way, the viral load of each sample to be tested can be detected directly by RT-PCR, but each sample to be tested can be mixed according to the predetermined mixing matrix and then the viral load in the mixture is detected, so the number of detections is reduced from n times to m times, thereby greatly reducing the detection cost consumed by large-scale virus sampling and detection. In addition, unlike traditional mixed detection, the epidemic disease detection device and method of this embodiment only needs to detect each mixed sample once to determine whether each sample to be tested is sick, while the traditional algorithm needs to sample and test all individuals participating in the group again after detecting that the mixture is positive.

[0062] In addition, in the above embodiments, only the detection of the new coronavirus is cited as an example. However, in actual applications, as long as certain diseases have the possibility of large-scale epidemics and there is a need for large-scale disease detection, and at the same time, virus carriers are relatively rare in the entire test population, effective detection can be performed using the epidemic disease detection device and method of this embodiment, that is, all infected people in a large sample can be detected with far fewer tests than traditional methods.

[0063] In the embodiment, before restoring the support set supp(x) of the sample vector x to be tested, the mixing matrix and the viral load are respectively transformed by the transformation matrix, so the column mean and column variance of the mixing matrix can be adjusted, thereby further improving the recovery effect.

[0064] The above embodiments are only used to illustrate specific implementation modes of the present invention, and the present invention is not limited to the description scope of the above embodiments.

[0065] For example, in the above embodiment, the diseased individual determination unit first recovers the sample vector x to be tested, and then determines the diseased individual by the position of the non-zero elements in the sample vector x to be tested. As an alternative, the diseased individual determination unit can also directly determine the support set supp(x) of the sample vector x to be tested, that is, the position of the non-zero elements in x, through a multiple orthogonal least squares algorithm, and thus directly determine the diseased individual based on the position.

Claims

1. An epidemic disease detection device, characterized in that: Used to detect sick individuals from a population with an infection rate below a predetermined infection threshold, including: A sample vector setting unit for testing sets the viral loads of n samples of individuals to be tested obtained by sampling from the population to be tested as a sample vector to be tested with a length of n; A mixing matrix generating unit generates a mixing matrix according to the number n of samples to be tested, the predetermined number of mixing groups, the maximum number of mixing allowed for each group, and the sampling probability; A mixed load acquisition unit, which acquires the viral load of each mixed sample obtained by mixing the samples to be tested according to the mixing matrix as a mixed load; a conversion unit, generating a conversion matrix according to the number of mixing groups and the sampling probability, and converting the mixing matrix and the mixing load according to the conversion matrix to obtain a corresponding conversion mixing matrix and a conversion viral load; and a diseased individual determination unit, which processes the conversion mixing matrix Ψ and the conversion viral load y using a predetermined multi-orthogonal least squares algorithm to obtain the position of the non-zero element in the sample vector x to be tested, and determines the diseased individual from the individuals to be tested based on the position of the non-zero element, Wherein, the mixing matrix generating unit comprises: A random matrix generating unit generates a random matrix according to the mixing group number m, the maximum mixing number N and the sampling probability p, The random matrix is ​​an m×n random Bernoulli matrix Φ, and each element φ in the random matrix ij Obeying Bernoulli(p) independent and identical distribution, The random matrix meets the following conditions: for The random Bernoulli matrix Φ satisfies row sparsity ‖φ i ‖0∈(0,N], column sparsity ‖φ j ‖0=mp, any two rows are different, any two columns are different, Wherein, i represents the i-th row of the random matrix, and j represents the j-th column of the random matrix; A dilution matrix generating unit generates a corresponding dilution matrix according to the random matrix: A mixing matrix generating unit generates the mixing matrix A←DΦ according to the random matrix and the dilution matrix, The element A in the mixing matrix A ij =0 means that the jth sample to be tested does not participate in the mixing of the i-th group of mixed samples, otherwise it participates. The random Bernoulli matrix Φ is generated by rejection sampling or Gibbs sampling. The transformation matrix satisfy: Where m is the number of mixed groups, p is the sampling probability, and a j is the j-th row of the mixing matrix A, The conversion mixing matrix Ψ←TA, the conversion viral load y←Tz, z is the mixed load, The multiple orthogonal least squares algorithm is specifically to continuously reduce the residual energy ‖r by iteration. k ‖2, where the residual vector r k Initialized to the transformed viral load y, the superscript k indicates the kth iteration, At the kth iteration, the mOLS algorithm measures the matrix, here the unselected columns ψ of the transformation mixing matrix Ψ i ,i∈Ω\T k-1 Projection to the selected column ψ i ,i∈T k-1 In the orthogonal complement space of Ω={1,2,…,n} is the full set, and each column after projection is l2-normalized, and the L normalized columns with the highest correlation with the residual are selected, that is: in, is the projection operator, is the orthogonal complementary projection, ψ i is the i-th column of Ψ, <,> represents the vector inner product; Afterwards, the selected columns are added to the estimated support set, namely: T k =T k-1 ∪I k And perform least squares projection on the measurement vector according to the new estimated support set to obtain a new estimate: And the new residual: Repeat the iteration until convergence, and finally select the cK column with the largest coefficient to re-project the least squares to obtain the final estimate of the sample vector x to be tested. and its estimated support set 2. The epidemic disease detection device according to claim 1, characterized in that: in, The infection threshold is 0.1‰.

3. A method for detecting an epidemic disease, characterized in that: Used to detect sick individuals from a population with an infection rate below a predetermined infection threshold, including: For n samples of individuals to be tested obtained by sampling from the population to be tested, the viral loads of the n samples to be tested are set as a sample vector to be tested with a length of n; Generate a mixing matrix according to the number n of samples to be tested, the predetermined number of mixing groups, the maximum number of mixing allowed for each group, and the sampling probability; Acquiring the viral load of each mixed sample obtained by mixing the samples to be tested according to the mixing matrix as a mixed load; Generate a conversion matrix according to the number of mixing groups and the sampling probability, and convert the mixing matrix and the mixing load according to the conversion matrix to obtain a corresponding conversion mixing matrix and conversion viral load; and The conversion mixing matrix Ψ and the conversion viral load y are processed by a predetermined multi-orthogonal least squares algorithm to obtain the position of the non-zero element in the sample vector x to be tested, and the diseased individual is determined from the individuals to be tested based on the position of the non-zero element, Wherein, the mixing matrix generating unit comprises: A random matrix generating unit generates a random matrix according to the mixing group number m, the maximum mixing number N and the sampling probability p, The random matrix is ​​an m×n random Bernoulli matrix Φ, and each element φ in the random matrix ij Obeying Bernoulli(p) independent and identical distribution, The random matrix meets the following conditions: for The random Bernoulli matrix Φ satisfies row sparsity ‖φ i ‖0∈(0,N], column sparsity ‖φ j ‖0=mp, any two rows are different, any two columns are different, Wherein, i represents the i-th row of the random matrix, and j represents the j-th column of the random matrix; A dilution matrix generating unit generates a corresponding dilution matrix according to the random matrix: A mixing matrix generating unit generates the mixing matrix A←DΦ according to the random matrix and the dilution matrix, The element A in the mixing matrix A ij =0 means that the jth sample to be tested does not participate in the mixing of the i-th group of mixed samples, otherwise it participates. The random Bernoulli matrix Φ is generated by rejection sampling or Gibbs sampling. The transformation matrix satisfy: Where m is the number of mixed groups, p is the sampling probability, and a j is the j-th row of the mixing matrix A, The conversion mixing matrix Ψ←TA, the conversion viral load y←Tz, z is the mixed load, The multiple orthogonal least squares algorithm is specifically to continuously reduce the residual energy ‖r by iteration. k ‖2, where the residual vector r k Initialized to the transformed viral load y, the superscript k indicates the kth iteration, At the kth iteration, the mOLS algorithm measures the matrix, here the unselected columns ψ of the transformation mixing matrix Ψ i ,i∈Ω\T k-1 Projection to the selected column ψ i ,i∈T k-1 In the orthogonal complement space of Ω={1,2,…,n} is the full set, and each column after projection is l2-normalized, and the L normalized columns with the highest correlation with the residual are selected, that is: in, is the projection operator, is the orthogonal complementary projection, ψ i is the i-th column of Ψ, <,> represents the vector inner product; Afterwards, the selected columns are added to the estimated support set, namely: T k =T k-1 ∪I k And perform least squares projection on the measurement vector according to the new estimated support set to obtain a new estimate: And the new residual: Repeat the iteration until convergence, and finally select the cK column with the largest coefficient to re-project the least squares to obtain the final estimate of the sample vector x to be tested. and its estimated support set

Images