JP2017051118A - Apparatus and method for generating characteristic estimation model and apparatus and method for estimating characteristics of analysis object - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an apparatus for generating a characteristic estimation model capable of exhaustively extracting explanatory variables for target variables and also provide an apparatus for estimating characteristics, the apparatus automatically estimating the growth situation of a crop using the generated model.SOLUTION: The present invention provides an apparatus 100 for generating a characteristic estimation model, the apparatus including an analysis data-updating unit 150 that generates data obtained by excluding genetic information used for generating a characteristic estimation model from analysis data and provides the data as update data. The generation of a characteristic estimation model is repeated up to the number of repetitions defined by using the update data. For each generation of the model, a characteristic-related gene extraction unit 140 extracts a gene related to the characteristics and records the gene. A characteristic estimation apparatus 300 uses a model generated in the apparatus 100 for generating a characteristic estimation model to quantitatively estimate a characteristic variable from an input state variable. A state diagnosis unit 320 diagnoses the growth information of a crop using a characteristic variable estimated by the characteristic estimation unit 310.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an analysis target characteristic estimation model generation apparatus and method, and an analysis target characteristic estimation apparatus and method, which generate an analysis target characteristic estimation model based on an analysis target characteristic variable and state information. In particular, the present invention relates to a characteristic estimation model generation apparatus and method suitable for estimating a trait that is a characteristic of a crop from gene expression information of a crop such as rice, and a characteristic estimation apparatus and method to be analyzed.

In recent years, the performance of processing apparatuses such as PCs has been greatly improved. For this reason, technological development for modeling a complex system such as a phenomenon occurring in the natural world by statistical analysis using a large amount of data has been active. A technique often used as such a statistical analysis technique is a regression analysis having a regularization term such as LASSO regression.
However, regression analysis with regularization terms is all about generating the best model from the given data. For this reason, there is a problem that the results are not unique, and it is not possible to exhaustively extract explanatory variables for the objective variable when generating the model.

Moreover, agriculture is one of the fields in which modeling as described above is desired.
In recent years, a technique for performing cultivation management by quantitatively measuring the degree of growth of crops, performing growth diagnosis and soil diagnosis has been developed. Specifically, cultivation and harvesting of crops are carried out scientifically and systematically by quantitatively measuring or predicting the degree of growth of crops, etc., and determining the exact fertilization time and amount.

Fertilization is a particularly important factor in the cultivation and harvesting of scientific and planned crops. The timing of fertilization has a major impact on crop yield and quality control.
In general, increasing the amount of fertilizer can increase the yield, but increases the cost. Furthermore, excessive fertilization will accumulate a large amount of protein, that is, nitrogen, in the harvest target site, and the quality will deteriorate. In addition, the fertilizer contains phosphoric acid. Due to excessive fertilization, for example, the phosphoric acid saturated in the paddy field flows out into rivers, seas, etc., leading to serious water pollution.
Therefore, fertilization with an appropriate amount and timing according to the growing situation of the crop is important for increasing the yield and quality of the crop.

Therefore, appropriate management of fertilization is important for the cultivation and harvesting of scientific and planned crops. For that purpose, it is important to grasp the growing situation of the crop more accurately and in a short time.
Conventionally, the growth status of a crop has been judged from information obtained from the appearance of the crop, such as the color of leaves. However, since the appearance of crops is affected by various factors such as temperature, sunshine duration, and water quality, it is difficult to quantitatively judge the crop growth status from the appearance.

  In order to cope with such a problem, a method for determining a drought stress situation, which is an index of whether or not the amount of water is sufficient, is measured by measuring the amount of chlorophyll and the amount of delayed luminescence in a crop. 1 is disclosed.

  However, the chlorophyll value is susceptible to disturbances, so there are large individual differences. For this reason, there is a problem that the determination result of the growth state becomes unstable by using the amount of chlorophyll, and a method for quantitatively determining the growth state without using the amount of chlorophyll has been desired.

  As a conventional technique for quantitatively determining the growth status without using chlorophyll content, there is no technique for generating an estimation model for determining whether a gene responds to the presence or absence of fertilization by analyzing gene expression information in maize. It is disclosed in Patent Document 1.

  By implementing the technique of Non-Patent Document 1, it is possible to determine whether or not the gene information expressed in the measurement target corn responds to the presence or absence of fertilization. However, in order to generate an estimation model, it is necessary to conduct experiments for narrowing down genes that respond to the presence or absence of fertilization, which is very complicated. Also, according to this technique, it is possible to create a model for estimating the nitrogen content that expresses how much nitrogen is accumulated in the leaves. However, there was a problem that agricultural traits other than nitrogen content could not be comprehensively estimated.

  Furthermore, the gene analysis as described above is often performed by an engineer having specialized knowledge, and complicated labor and time are required for determining the growth status. Therefore, a technique for automatically determining the growth status in a short time is desired.

JP 2013-183702 A

Yang XS1, Wu J, Ziegler TE, Yang X, Zayed A, Rajani MS, Zhou D, Basra AS, Schachtman DP, Peng M, Armstrong CL, Caldo RA, Morrell JA, Lacy M, Staub JM (2011) Gene expression biomarkers provide sensitive indicators of in planta nitrogen status in maize.Plant Physiology December 2011, Vol. 157, pp. 1841-1852

  In view of the above circumstances, in the present invention, in model generation for estimating the characteristics of an analysis target, a model generation apparatus capable of exhaustively extracting explanatory variables for objective variables, and the growth status of crops automatically using the generated model An object of the present invention is to obtain an apparatus for estimating the above.

(Configuration 1)
A characteristic estimation model generation device that generates a model that estimates the characteristic variable from the state variable that represents the state of the analysis target and the characteristic variable that represents the characteristic of the analysis target.
A data output unit that receives the state variable to be analyzed and the characteristic variable to be analyzed and outputs as analysis data;
Of the analysis data, the regression variable representing the relationship between the objective variable and the explanatory variable is obtained by performing a regression analysis having a regularization term using the characteristic variable as an objective variable and the state variable as an explanatory variable. A regression analysis unit to generate,
Using the regression model and the analysis data, cross-validation is performed up to a preset number of verifications, and a model having an optimal regularization term among the regression models is generated as a characteristic estimation model generation. And
Data corresponding to the explanatory variable selected in the characteristic estimation model, data excluded from the analysis data, is generated as update data, and is sent to the data output unit as analysis data at the next generation of the characteristic estimation model. An analysis data update unit to output,
With
A characteristic estimation model generation apparatus that repeats updating of the analysis data and generation of a characteristic estimation model using the update data up to a preset number of repetitions.

(Configuration 2)
About the regression model generated every time cross-validation is performed in the characteristic estimation model generation unit, the explanatory variables selected in all models are characteristic-related state information that is state information related to the characteristic to be analyzed The characteristic estimation model production | generation apparatus of the structure 1 further provided with the characteristic relevant state information extraction part extracted as follows.

(Configuration 3)
The characteristic estimation model generation device according to Configuration 1 or 2, wherein the regression analysis is LASSO regression, Ridge regression, or Elastic-net.

(Configuration 4)
The characteristic estimation model generation device according to any one of configurations 1 to 3, wherein the analysis target is a living organism, the state variable is gene expression information, and the characteristic variable is trait information.

(Configuration 5)
The characteristic estimation model generation device according to Configuration 4, wherein the organism is a crop.

(Configuration 6)
The characteristic estimation model generation device according to Configuration 5, wherein the crop is a gramineous crop.

(Configuration 7)
The characteristic estimation model generation device according to configuration 6, wherein the gramineous crop is rice.

(Configuration 8)
The trait information is the number of days after the transplantation date representing the number of days from the date of transplantation, the flowering date representing the number of days required from the sampling date to flowering, or the nitrogen content representing the nitrogen content per dry weight of the observation target. The characteristic estimation model generation apparatus according to any one of the above.

(Configuration 9)
A characteristic estimation device for estimating the characteristic of the analysis target,
A characteristic estimation that is an estimation result of the characteristic of the analysis target when the state variable of the analysis target is input to the characteristic estimation model generated by the characteristic estimation model generation device according to any one of configurations 1 to 8 A characteristic estimation apparatus comprising a characteristic estimation unit for outputting information.

(Configuration 10)
A characteristic estimation device for estimating the characteristic of the analysis target,
Characteristic estimation that is an estimation result of the characteristics of the analysis target by inputting gene expression information of the organism to the characteristic estimation model generated by the characteristic estimation model generation device according to any one of configurations 4 to 8 A characteristic estimation unit for outputting information;
The gene expression information of the organism is measured from the amount of mRNA transcribed from a gene corresponding to the property-related state information extracted by the property-related state information extraction unit.

(Configuration 11)
The characteristic estimation apparatus according to Configuration 10, wherein the expression level of the mRNA is measured by a quantitative RT-PCR method.

(Configuration 12)
The characteristic estimation apparatus according to Configuration 11, wherein the quantitative RT-PCR method is a real-time RT-PCR method.

(Configuration 13)
The characteristic estimation apparatus according to Configuration 11, wherein the quantitative RT-PCR method is a multiplex RT-PCR method.

(Configuration 14)
The characteristic estimation apparatus according to any one of configurations 10 to 13, further comprising a state diagnosis unit that diagnoses the state of the organism using the characteristic estimation information.

(Configuration 15)
A characteristic estimation model generation method for generating a model for estimating the characteristic variable from the state variable representing the state of the analysis target and the characteristic variable representing the characteristic of the analysis target by the state variable,
The state variable to be analyzed and the characteristic variable to be analyzed are data for analysis,
Of the analysis data, the regression variable representing the relationship between the objective variable and the explanatory variable is obtained by performing a regression analysis having a regularization term using the characteristic variable as an objective variable and the state variable as an explanatory variable. A regression analysis step to generate,
Using the regression model and the analysis data, cross-validation is performed up to a preset number of verifications, and a model having an optimal regularization term among the regression models is generated as a characteristic estimation model generation. Steps,
Data for analysis corresponding to the explanatory variable selected in the characteristic estimation model is generated as update data, data that is excluded from the analysis data, and is updated as analysis data for the next generation of the characteristic estimation model An update step;
With
A characteristic estimation model generation method characterized by repeating the updating of the analysis data and the generation of a characteristic estimation model using the update data up to a preset number of repetitions.

(Configuration 16)
About the regression model generated every time cross-validation is performed in the characteristic estimation model generation step, the explanatory variables selected in all models are characteristic-related state information that is state information related to the characteristic to be analyzed The characteristic estimation model generation method according to configuration 15, further comprising a characteristic-related state information extraction step that is extracted as follows.

(Configuration 17)
The characteristic estimation model generation method according to Configuration 15 or 16, wherein the regression analysis is LASSO regression, Ridge regression, or Elastic-net.

(Configuration 18)
The characteristic estimation model generation method according to any one of configurations 15 to 17, wherein the analysis target is a living organism, the state variable is gene expression information, and the characteristic variable is trait information.

(Configuration 19)
The characteristic estimation model generation method according to configuration 18, wherein the organism is a crop.

(Configuration 20)
The characteristic estimation model generation method according to Configuration 19, wherein the crop is a gramineous crop.

(Configuration 21)
The characteristic estimation model generation method according to configuration 20, wherein the gramineous crop is rice.

(Configuration 22)
The trait information is any one of configurations 19 to 21 which is the number of days after the transplantation date indicating the number of days from the date of transplantation, the flowering date indicating the number of days required for flowering, or the nitrogen content indicating the nitrogen content per dry weight of the observation target The characteristic estimation model generation method described.

(Configuration 23)
A property estimation method for estimating the property of the analysis target,
A characteristic estimation that is an estimation result of the characteristic of the analysis target by inputting the state variable of the analysis target into the characteristic estimation model generated by the characteristic estimation model generation method according to any one of the configurations 15 to 22 A characteristic estimation method comprising a characteristic estimation step in which information is output.

(Configuration 24)
A property estimation method for estimating the property of the analysis target,
A characteristic estimation, which is an estimation result of the characteristic of the analysis target, when gene expression information of the organism is input to the characteristic estimation model generated by the characteristic estimation model generation method according to any one of configurations 18 to 22. A characteristic estimation step in which information is output;
The gene estimation information of the organism is measured from the amount of mRNA transcribed from the gene corresponding to the property-related state information extracted in the property-related state information extraction step.

(Configuration 25)
25. The characteristic estimation method according to Configuration 24, wherein the mRNA expression level is measured by a quantitative RT-PCR method.

(Configuration 26)
The characteristic estimation method according to Configuration 25, wherein the quantitative RT-PCR method is a real-time RT-PCR method.

(Configuration 27)
The characteristic estimation method according to Configuration 25, wherein the quantitative RT-PCR method is a multiplex RT-PCR method.

(Configuration 28)
The characteristic estimation method according to any one of configurations 24 to 27, further comprising a state diagnosis step of diagnosing the state of the organism using the characteristic estimation information.

  According to the present invention, there is an effect that it is possible to obtain a characteristic estimation model generation apparatus capable of exhaustively extracting state variables related to characteristic variables to be analyzed. Furthermore, there is an effect that an apparatus for automatically estimating the growth state of a crop can be obtained using the generated model.

It is a block diagram which shows the characteristic estimation model production | generation apparatus in embodiment of this invention. It is a flowchart which shows operation | movement of the characteristic estimation model production | generation apparatus in embodiment of this invention. It is a block diagram which shows the characteristic estimation apparatus in embodiment of this invention. It is a flowchart which shows operation | movement of the characteristic estimation apparatus in embodiment of this invention. It is a figure showing the result of having estimated the characteristic using the characteristic estimation model created from the analytical data of 225, and the analytical data by making the horizontal axis the number of days after transplantation and the vertical axis the nitrogen content per dry weight. FIG. 6 shows the estimation of characteristics generated for each number of iterations, using rice as an analysis target, and 497 samples, 428 samples, and 224 sampled from actual fields for the number of days after transplanting, date of flowering, and nitrogen content, respectively. The accuracy of the model is evaluated by the mean square error. FIG. 7 shows the number of characteristic variables recorded in the characteristic related state information extraction unit 140 for each number of repetitions when the number of repetitions is 10 and the other conditions are the same as in FIG. FIG. 8 is a diagram showing how many times the same explanatory variable is selected when cross-validation is repeated 10 times for the number of days after transplantation. FIG. 9 is a diagram showing how many times the same explanatory variable is selected when the cross-validation is repeated 10 times for the flowering date. FIG. 10 is a diagram showing how many times the same explanatory variable is selected when the cross-validation is repeated 10 times for the nitrogen content.

Embodiment FIG. 1 is a block diagram showing a configuration of a characteristic estimation model generation apparatus according to an embodiment of the present invention. The configuration of the characteristic estimation model generation device according to the embodiment will be described with reference to the drawings.

In the present embodiment, the analysis target is rice.
A state variable is rice gene expression information.
Characteristic variables represent trait information, which is information that expresses the nature and characteristics of rice. Specifically, the nitrogen content in leaves per dry weight of rice leaves, and how many days after the sampling date the ears will appear. It represents the flowering date and the number of days after planting that represents how many days have passed since the planting.

  In FIG. 1, a characteristic estimation model generation apparatus 100 includes a data output unit 110, a regression analysis unit 120, a characteristic estimation model generation unit 130, a characteristic related state information extraction unit 140, and an analysis data update unit 150. It is an apparatus for generating a characteristic variable estimation model to be analyzed from a characteristic variable to be analyzed and gene expression information.

  The data output unit 110 receives characteristic variables and state variables to be analyzed as data for creating a characteristic estimation model via an interface (not shown) or the like, and outputs them to the regression analysis unit 120 as analysis data. .

  The regression analysis unit 120 receives analysis data from the data output unit 110. Then, regression analysis having a regularization term is performed on the analysis data. Specifically, a regression model having a regularization term is generated using characteristic variables as objective variables (dependent variables) and state variables as explanatory variables in the analysis data.

  The regression analysis used in the regression analysis unit 120 will be described below.

  In the present embodiment, the regression analysis unit 120 uses LASSO regression (Least Absolute Shrinkage and Selection Operator). The calculation formula of the regression model by LASSO regression is expressed by the following formula 1.

ｙは目的変数、ｘは説明変数、βは回帰係数、γは正則化パラメータである。

y is an objective variable, x is an explanatory variable, β is a regression coefficient, and γ is a regularization parameter.

  A regression model is generated by substituting the input analytical data into Equation 1 and calculating a coefficient β that minimizes the value of S. That is, it solves the minimization problem for S. The minimization problem can be uniquely obtained by solving the convex optimization problem for S.

  The obtained optimal solution is a regression model represented by a function of γ, with characteristic variables as objective variables and state variables as explanatory variables.

  The characteristic estimation model generation unit 130 obtains an optimum value of γ in the regression model generated by the regression analysis unit, and generates a characteristic estimation model that is a relational expression between an arbitrary characteristic variable and a state variable.

  Hereinafter, a method for generating a characteristic estimation model in the characteristic estimation model generation unit 130 will be described.

  The characteristic estimation model generation unit 130 performs cross-validation in order to obtain an optimal value of γ in the regression model generated by the regression analysis unit 120.

  In this embodiment, k-fold cross-validation is used as a cross-validation technique. In k-fold cross-validation, a value of k representing the number of verifications is set in advance. Then, all analysis data is divided into k sample groups, one of which is used as test data, and the remaining k-1 is used as training data. That is, k combinations of test data and training data are created from the analysis data.

  Then, one training data is substituted into Equation 1, and a regression model as a function of γ is obtained and recorded. Test data is applied to the model, and the value of γ is sequentially substituted to find the mean square error for the objective variable. This is repeated for k combinations of test data and training data.

  Thus, k mean square errors are calculated for a certain value of γ. The average of these values is taken as the value of the mean square error at a certain γ. Further, a standard error at a certain γ is calculated.

  Then, the minimum value of the mean square error calculated in this way and γ at that time are obtained. Next, an optimal value of γ is selected based on one standard error rule. That is, the maximum γ existing within the standard error range for γ is selected as the optimum value of γ.

  Finally, a characteristic estimation model is generated and recorded by obtaining a regression model according to Equation 1 using the optimum value of γ thus obtained and analysis data.

  In the present embodiment, k = 10. The value of k may be set in advance or may be input via an interface (not shown).

  The characteristic-related state information extraction unit 140 extracts and records a state variable related to a certain characteristic using the plurality of regression models generated by the characteristic estimation model generation unit 130.

First, the regression model calculated by substituting the training data into Equation 1 in the characteristic estimation model generation unit 130 is recorded every time cross-validation is performed. That is, k regression models are recorded.
Then, the k regression models recorded are compared. The state variables selected in all k models, that is, explanatory variables satisfying β ≠ 0 are extracted as those that are highly likely to be state variables related to a certain characteristic.
The state variables extracted in this way are recorded as characteristic related state information.

  The analysis data update unit 150 generates update data, which is analysis data used when generating the next characteristic estimation model. Then, the update data is output to the data output unit 110, and the analysis data is updated.

  In creating the update data, data relating to the explanatory variable whose coefficient β is not 0 among the characteristic estimation model generated by the characteristic estimation model generation unit 130 is excluded from the analysis data.

  That is, in the next generation of the characteristic estimation model, the state variables used for generating the previous characteristic estimation model are not used.

  By generating a new property estimation model using the update data created in this way, a property estimation model is generated using only state variables that were not used to generate the previous property estimation model. can do.

  In addition, by repeatedly updating analysis data and generating characteristic estimation models up to a preset number of iterations, it is possible to generate multiple characteristic estimation models using all state variables related to any characteristic variable. It becomes.

  In the present embodiment, the number of repetitions is set to 5. Further, the value of the number of repetitions may be set in advance or may be input via an interface (not shown) or the like.

  FIG. 2 is a flowchart showing a schematic operation of the characteristic estimation model generation apparatus 100 according to the present embodiment. The characteristic estimation model generation apparatus 100 operates as follows.

  First, a characteristic variable, a state variable, the number of verifications, and the number of repetitions are input by an input unit (not shown), and the characteristic variable and the state variable are set as analysis data (S201).

  Next, the analysis data is divided into k sample groups, the i-th sample group is used as test data, and the remaining sample groups are used as training data. The initial value of i is 1. Further, the division into the sample groups is performed so that the number of samples of each sample group becomes equal (S202).

  Using the training data, the regression analysis unit 120 obtains a regression model and records it together with the values of i and j (S203).

  Next, the characteristic estimation model generation unit 130 inputs test data to the regression model obtained in S203, and calculates a mean square error and a standard error at each γ (S204).

  Here, the value of i is checked. If the value of i is not equal to the number of verifications (NO in S205), 1 is added to the value of i, and the process returns to S202. When the value of i is equal to the number of verifications (YES in S205), the process proceeds to S206.

  When the value of i is equal to the number of verifications in S205, the characteristic estimation model generation unit 130 calculates an addition average for the mean square error value in each γ value, and calculates the optimum γ value based on one standard error rule. Determine (S206).

  Next, in the characteristic estimation model generation unit 130, the regression analysis unit 120 obtains and records a characteristic estimation model using all of the optimum γ value obtained in S206 and the analysis data (S207).

Of the plurality of regression models recorded in S203, all regression models recorded together with the current value of j are compared. An explanatory variable whose coefficient is not 0 in all the models compared is extracted and recorded as characteristic related state information (S208).
Since the characteristic-related state information recorded here is selected as a state variable representing the characteristic variable in all the models, there is a high possibility that the characteristic-related state information is a state variable related to a specific characteristic variable.

  After recording the characteristic-related state information in S208, the update data obtained by excluding the data related to the explanatory variable whose coefficient is not 0 from the analysis data in the characteristic estimation model recorded in S207 is sent to the analysis data update unit 150. (S209).

  Here, the value of j is checked. If the value of j is not equal to the number of repetitions (in the case of NO in S210), 1 is added to the value of j, and the analysis data is changed to the update data generated in S209. , I is set to an initial value (i = 1), and the process returns to S202. If the value of j is equal to the number of repetitions (YES in S210), the operation is terminated.

  As described above, the analysis data is updated, and the generation of the characteristic estimation model and the extraction of the characteristic related state information are repeated. A plurality of models to be estimated can be generated.

  FIG. 3 is a block diagram showing the configuration of the characteristic estimation apparatus according to the embodiment of the present invention. The configuration of the characteristic estimation apparatus according to the present embodiment will be described with reference to the drawings.

  In FIG. 3, a characteristic estimation apparatus 300 includes a characteristic estimation unit 310 and a state diagnosis unit 320, and outputs characteristic estimation information and a state diagnosis result from input state variables.

  In the characteristic estimation unit 310, estimation models related to various characteristics generated by the characteristic estimation model generation apparatus 100 are recorded, and state information to be analyzed is input. By substituting the input state information into the recorded characteristic estimation model, estimation information regarding various characteristics is output. Here, the estimation information represents the estimation value itself output from the characteristic estimation model.

  Note that the estimation information may be displayed on a display unit (not shown), the output value of the estimation information may be displayed as it is, or a threshold value is set and the concept is such that it is more or less than the threshold value. It may be displayed in a typical expression. Further, all of the estimation information may be output, or the user may specify the characteristic desired to be estimated at the start of the operation of the characteristic estimation apparatus 300 by an input unit (not shown).

  Here, the state information in the characteristic estimation apparatus 300 will be described.

In the present embodiment, the state information is rice gene expression information. The rice gene expression information specifically represents the expression level of mRNA corresponding to the genes constituting rice. The number of all genes constituting rice is estimated to exceed 32,000. Therefore, the state information in this example represents the expression level of mRNA corresponding to all these genes.
However, for state information, measuring and inputting the expression level of mRNA corresponding to all 32,000 genes described above is a matter of time and characteristic estimation from the viewpoint of determining the growth status of rice in a short time. From the viewpoint of calculation cost, it is not realistic. Further, not all mRNA expression information has information on growth diagnosis.
Therefore, the state variable used in the characteristic estimation apparatus 300 uses the state variable extracted by the characteristic related state information extraction unit 140.

  The gene expression level of rice as a characteristic estimation target can be measured using a quantitative RT-PCR method.

  As the quantitative RT-PCR method, an arbitrary method such as a real-time RT-PCR method or a multiplex RT-PCR method can be used in addition to a normal RT-PCR method. In addition to the RT-PCR method, other gene amplification methods such as the LAMP method may be used.

  The state diagnosis unit 320 receives the property estimation information output from the property estimation unit 310, diagnoses the current state of the analysis target according to the property estimation information, and outputs a state diagnosis result.

The condition diagnosis results are characteristic estimation information such as the number of days after transplanting that represents the number of days since rice planting, the flowering date that represents the number of days required from the sampling date to flowering, or the nitrogen content that represents the nitrogen content per dry weight of the observation target. Is used to represent the growth status of rice.
In addition to the growth situation, input ideal growth data in advance, calculate the deviation from the ideal growth data, and follow the behavior guidelines such as when and how much fertilizer should be given according to the degree of deviation May be shown on a display unit (not shown).

  FIG. 4 is a flowchart showing a schematic operation of the characteristic estimation apparatus 300 in the present embodiment. The characteristic estimation apparatus 300 operates as follows.

  First, rice to be estimated for characteristics is prepared, and the expression level of the gene extracted by the characteristic-related state information extraction unit 140 in the rice is measured. (S401).

  Next, the measurement result is substituted into the characteristic estimation model for the characteristic desired to be estimated, which is generated by the characteristic estimation model generation apparatus 100 (S402).

  The characteristic estimation result in the characteristic estimation model is output as characteristic estimation information (S403).

  Rice growth information is diagnosed and output using the characteristic estimation result in the characteristic estimation model (S404).

  As mentioned above, according to this Embodiment, there exist the following effects.

  The characteristic estimation model generation apparatus 100 is configured to generate a characteristic estimation model in which the characteristic variable is a dependent variable and the characteristic variable is an explanatory variable from the characteristic variable and characteristic variable to be analyzed in the characteristic estimation model generation unit 130. There is an effect that a model for quantitatively estimating an arbitrary characteristic variable can be generated.

  Furthermore, in the characteristic estimation model generation apparatus 100, the analysis data update unit 150 generates, as update data, data obtained by excluding gene information used for characteristic estimation model generation from the analysis data. Then, the generation of the characteristic estimation model is repeated up to a preset number of times using the update data, and the characteristic-related gene extraction unit 140 extracts and records the gene related to the characteristic every time the characteristic estimation model is generated. Therefore, there is an effect that gene information related to an arbitrary characteristic variable can be exhaustively extracted.

  In the characteristic estimation apparatus 300, the characteristic estimation unit 310 quantitatively estimates the characteristic variable from the input state variable using the model generated by the characteristic estimation model generation apparatus 100. Further, the state diagnosis unit 320 uses the characteristic variable estimated by the characteristic estimation unit 310 to diagnose crop growth information. Since the characteristic estimation apparatus 300 is configured as described above, there is an effect that the growth state of the analysis target can be automatically estimated without the intervention of a professional engineer. Furthermore, there is also an effect that the cultivation management by the growth diagnosis and the soil diagnosis is advanced by automatically estimating the growth state of the analysis target.

  In the present embodiment, rice has been described as an analysis target. However, the present invention is not limited to this, and any analysis target can be used as long as an estimation model can be generated by regression analysis having a regularization term.

  In addition, regarding the characteristic variables, the nitrogen content in the leaf per dry weight, the flowering date indicating how many days after the rice planting, and the number of days after rice planting indicating how many days have passed since the rice planting were explained. It may be a characteristic variable that constitutes an analysis target that can generate an estimation model by regression analysis having a regularization term, such as phosphorus content, ear size and weight.

  The characteristic estimation model generation device 100 and the characteristic estimation device 300 according to the present embodiment are only required to be able to perform numerical computation, input numerical values, and record and output the results. Specifically, a CPU, a memory, and a display unit A computer having an input / output interface or the like, or dedicated hardware can be used.

  In this embodiment, LASSO regression is used as a regression analysis method used in the regression analysis unit 120, but a regression analysis method having a regularization term such as Ridge regression or Elastic-Net may be used. In addition, in the present embodiment, K-fold cross-validation is used for the cross-validation method used in the characteristic estimation model generation unit 130 and the characteristic-related gene extraction unit 140, but leave-one-out cross-validation, etc. Other cross-validation methods may be used.

  In the present embodiment, model creation by LASSO regression and determination of optimal γ were performed using statistical analysis software “R”.

FIG. 5 shows the reproducibility of the model generated by the characteristic estimation model generation device 100 in the present embodiment when the analysis target is rice and the analysis data is 224 samples collected from an actual field. Is. The number of verifications is 10 and the number of repetitions is 1.
The upper part of FIG. 5 is a plot of 224 samples for analysis, with the horizontal axis representing the number of days after transplantation and the vertical axis representing the nitrogen content per dry weight, for the rice to be analyzed. The type of marker corresponds to the place where the rice was collected.
Next, a characteristic estimation model is created by the characteristic estimation model generation apparatus 100 in the present embodiment using the gene expression information in the analysis data as a state variable, the number of days after transplantation, and the nitrogen content per dry weight as a characteristic variable.
The lower diagram of FIG. 5 is a plot of the result of substituting the data for analysis into the characteristic estimation model for the number of days after transplantation and the nitrogen content per dry weight.

  FIG. 6 shows a characteristic estimation model according to the present embodiment, using rice as an analysis target, the days after transplanting, the days of flowering, and the nitrogen content, using 497 samples, 428 samples, and 224 collected from an actual field as analysis data. When the characteristic estimation model is generated by the generation device 100, the accuracy of the model is evaluated by the mean square error with each sample. The number of verifications is 10 and the number of repetitions is 5. In FIG. 6, 1stModel, 2ndModel,... Indicate the evaluation of the model generated for each repetition count.

FIG. 7 shows the number of characteristic variables, that is, the number of genes recorded in the characteristic-related state information extraction unit 140 for each number of repetitions when the number of repetitions is 10 and the other conditions are the same as in FIG. Is.
Taking the nitrogen content as an example, it can be seen that the genes that have a high possibility of being related to the nitrogen content are narrowed down to 582 from the 32,000 genes that constitute rice.

FIG. 8 shows the number of days after transplanting, FIG. 9 shows the flowering date, and FIG. 10 shows the stability distribution when the analysis target is the nitrogen content per dry weight.
The vertical axis of the stability distribution shows the number of times the explanatory variable is selected by comparing all the models when the generation of the model is repeated the number of verifications in the process of extracting the explanatory variable in the characteristic related state information extracting unit 140. That number.
The horizontal axis of the stability distribution is the number of repetitions.
It can be seen that as the number of iterations increases, the explanatory variables selected in all models decrease by the number of verifications.
In FIGS. 8, 9, and 10, the number of verifications is 10 times. The number of repetitions is 15, 25 and 15 respectively. The analysis data are rice 939 samples, 819 samples, and 302 samples collected from actual fields, respectively.

  Although the present invention has been described with reference to the embodiment, the present invention is not limited to the above-described embodiment. Various changes that can be understood by those skilled in the art can be made to the configuration and operation of the present invention without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 100 ... Characteristic estimation model generation apparatus 110 ... Data output part 120 ... Regression analysis part 130 ... Characteristic estimation model generation part 140 ... Characteristic related information extraction part 150 ... Analysis data update part 300 ... Characteristic estimation apparatus 310 ... Characteristic estimation part 320 ... Condition diagnosis section

Claims (28)

A characteristic estimation model generation device that generates a model that estimates the characteristic variable from the state variable that represents the state of the analysis target and the characteristic variable that represents the characteristic of the analysis target.
A data output unit that receives the state variable to be analyzed and the characteristic variable to be analyzed and outputs as analysis data;
Of the analysis data, the regression variable representing the relationship between the objective variable and the explanatory variable is obtained by performing a regression analysis having a regularization term using the characteristic variable as an objective variable and the state variable as an explanatory variable. A regression analysis unit to generate,
Using the regression model and the analysis data, cross-validation is performed up to a preset number of verifications, and a model having an optimal regularization term among the regression models is generated as a characteristic estimation model generation. And
Data corresponding to the explanatory variable selected in the characteristic estimation model, data excluded from the analysis data, is generated as update data, and is sent to the data output unit as analysis data at the next generation of the characteristic estimation model. An analysis data update unit to output,
With
A characteristic estimation model generation apparatus that repeats updating of the analysis data and generation of a characteristic estimation model using the update data up to a preset number of repetitions.   About the regression model generated every time cross-validation is performed in the characteristic estimation model generation unit, the explanatory variables selected in all models are characteristic-related state information that is state information related to the characteristic to be analyzed The characteristic estimation model production | generation apparatus of Claim 1 further equipped with the characteristic relevant state information extraction part extracted as.   The characteristic estimation model generation device according to claim 1, wherein the regression analysis is LASSO regression, Ridge regression, or Elastic-net.   The characteristic estimation model generation apparatus according to claim 1, wherein the analysis target is a living organism, the state variable is gene expression information, and the characteristic variable is trait information.   The characteristic estimation model generation device according to claim 4, wherein the organism is a crop.   The characteristic estimation model generation device according to claim 5, wherein the crop is a gramineous crop.   The characteristic estimation model generation apparatus according to claim 6, wherein the gramineous crop is rice.   The trait information is the number of days after the date of transplantation representing the number of days from the date of transplantation, the date of flowering representing the number of days required for flowering, or the nitrogen content representing the nitrogen content per dry weight of the observation target. The characteristic estimation model generation device described in 1. A characteristic estimation device for estimating the characteristic of the analysis target,
A characteristic that is an estimation result of the characteristic of the analysis target when the state variable of the analysis target is input to the characteristic estimation model generated by the characteristic estimation model generation device according to claim 1. A characteristic estimation apparatus, comprising: a characteristic estimation unit that outputs estimation information. A characteristic estimation device for estimating the characteristic of the analysis target,
A characteristic that is an estimation result of the characteristic of the analysis target by inputting gene expression information of the organism into the characteristic estimation model generated by the characteristic estimation model generation device according to claim 4. A characteristic estimation unit that outputs estimation information;
The gene expression information of the organism is measured from the amount of mRNA transcribed from a gene corresponding to the property-related state information extracted by the property-related state information extraction unit.   The characteristic estimation apparatus according to claim 10, wherein the expression level of the mRNA is measured by a quantitative RT-PCR method.   The characteristic estimation apparatus according to claim 11, wherein the quantitative RT-PCR method is a real-time RT-PCR method.   The characteristic estimation apparatus according to claim 11, wherein the quantitative RT-PCR method is a multiplex RT-PCR method.   The characteristic estimation apparatus according to claim 10, further comprising a state diagnosis unit that diagnoses the state of the living organism using the characteristic estimation information. A characteristic estimation model generation method for generating a model for estimating the characteristic variable from the state variable representing the state of the analysis target and the characteristic variable representing the characteristic of the analysis target by the state variable,
The state variable to be analyzed and the characteristic variable to be analyzed are data for analysis,
Of the analysis data, the regression variable representing the relationship between the objective variable and the explanatory variable is obtained by performing a regression analysis having a regularization term using the characteristic variable as an objective variable and the state variable as an explanatory variable. A regression analysis step to generate,
Using the regression model and the analysis data, cross-validation is performed up to a preset number of verifications, and a model having an optimal regularization term among the regression models is generated as a characteristic estimation model generation. Steps,
Data for analysis corresponding to the explanatory variable selected in the characteristic estimation model is generated as update data, data that is excluded from the analysis data, and is updated as analysis data for the next generation of the characteristic estimation model An update step;
With
A characteristic estimation model generation method characterized by repeating the updating of the analysis data and the generation of a characteristic estimation model using the update data up to a preset number of repetitions.   About the regression model generated every time cross-validation is performed in the characteristic estimation model generation step, the explanatory variables selected in all the models are characteristic related state information that is state information related to the characteristics to be analyzed. The characteristic estimation model generation method according to claim 15, further comprising a characteristic-related state information extraction step that is extracted as follows.   The characteristic estimation model generation method according to claim 15 or 16, wherein the regression analysis is LASSO regression, Ridge regression, or Elastic-net.   The characteristic estimation model generation method according to claim 15, wherein the analysis target is an organism, the state variable is gene expression information, and the characteristic variable is trait information.   The characteristic estimation model generation method according to claim 18, wherein the organism is a crop.   The characteristic estimation model generation method according to claim 19, wherein the crop is a gramineous crop.   21. The characteristic estimation model generation method according to claim 20, wherein the gramineous crop is rice.   The trait information is the number of days after the transplantation date representing the number of days from the date of transplantation, the flowering date representing the number of days required for flowering, or the nitrogen content representing the nitrogen content per dry weight of the observation target. The characteristic estimation model generation method described in 1. A property estimation method for estimating the property of the analysis target,
A characteristic, which is an estimation result of the characteristic of the analysis target, by inputting the state variable of the analysis target to the characteristic estimation model generated by the characteristic estimation model generation method according to any one of claims 15 to 22. A characteristic estimation method comprising a characteristic estimation step in which estimation information is output. A property estimation method for estimating the property of the analysis target,
23. A characteristic that is an estimation result of the characteristic of the analysis target by inputting gene expression information of the organism to the characteristic estimation model generated by the characteristic estimation model generation method according to claim 18. A characteristic estimation step in which estimation information is output;
The gene estimation information of the organism is measured from the amount of mRNA transcribed from the gene corresponding to the property-related state information extracted in the property-related state information extraction step.   The characteristic estimation method according to claim 24, wherein the expression level of the mRNA is measured by a quantitative RT-PCR method.   The characteristic estimation method according to claim 25, wherein the quantitative RT-PCR method is a real-time RT-PCR method.   The characteristic estimation method according to claim 25, wherein the quantitative RT-PCR method is a multiplex RT-PCR method.   The characteristic estimation method according to claim 24, further comprising a state diagnosis step of diagnosing the state of the organism using the characteristic estimation information.

Images