Paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors

Abstract

The invention relates to the technical field of paliperidone metabolic efficiency evaluation, in particular to a paliperidone metabolic efficiency evaluation system based on a CYP2D6 gene and environmental factors. According to the system, the obtained historical data information is preprocessed, so that more effective data information can be obtained, and a model can be trained more effectively by using the data information; meanwhile, a dynamic metabolism prediction module formed by weighting output of each network in a multi-modal feature fusion module by utilizing a gating attention mechanism is trained based on the preprocessed data, so that a gene-environment-PK / PD multi-dimensional model is obtained, various factors of the environment and the gene can be quantified, and multi-modal data are fused; the dynamic evaluation of the metabolic efficiency is further realized; according to the method, the current data information is scored by using the gene-environment-PK / PD multi-dimensional model, and the score is compared with the preset score, so that whether the model evaluation is qualified or not can be quickly judged, and the model evaluation is adjusted when the model evaluation is unqualified, thereby further improving the evaluation accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of paliperidone metabolism efficiency evaluation, and particularly relates to a paliperidone metabolism efficiency evaluation system based on CYP2D6 genes and environmental factors. BACKGROUND

[0002] The prior art mainly evaluates paliperidone metabolism efficiency in the following ways: first, a genotyping static model: based on CYP2D6 genotypes (such as *3, *4, *10 mutations), patients are divided into fast, medium and slow metabolizers; second, therapeutic drug monitoring (TDM): blood drug concentrations are detected by liquid chromatography-mass spectrometry, and the dose is adjusted by combining a fixed formula; third, an empirical dose algorithm: based on patient weight, age and other parameters, an initial dose is recommended (such as oral preparation 4-12 mg / day); fourth, a machine learning model: a CYP2D6 metabolism classifier based on a random forest algorithm, but its input data only includes genotypes and basic clinical parameters (such as age, gender), without integrating dynamic environmental exposure data, resulting in limited adaptability of the model in real-world scenarios. Therefore, the prior art has the following defects: first, significant deviation between genotype prediction and actual metabolism (such as environmental factors that can induce changes in CYP2D6 enzyme activity); second, lack of environmental factors: existing models do not integrate environmental influences such as PM2.5 exposure and dietary habits, resulting in dose recommendations deviating from individual needs; finally, static data lag: TDM relies on laboratory testing and cannot respond in real time to dynamic changes in patient physiological status.

[0003] Chinese Patent Publication No. CN118039071A discloses a health evaluation method, device, equipment and storage medium based on a metabolism model, which comprises: acquiring basic information and metabolism data of sample personnel, constructing an energy metabolism model based on the basic information and metabolism data of the sample personnel; responding to and analyzing a health evaluation request to obtain user information to be evaluated and an evaluation project to be evaluated; finding the corresponding energy metabolism model according to the evaluation project to be evaluated; based on the user information to be evaluated, evaluating the health status of the current user to be evaluated according to the energy metabolism model corresponding to the evaluation project to be evaluated, and outputting the evaluation result of the evaluation project to be evaluated.

[0004] As can be seen, the prior art has the following problems: the evaluation process does not consider environmental and genetic factors, so there is a lack of dynamic evaluation models that integrate multiple data, and the score results of the evaluation model are not compared with the preset score, and the score results are not adjusted when the comparison result is unqualified, thereby reducing the accuracy of the evaluation. SUMMARY

[0005] To this end, the present application provides a paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors, so as to overcome the problem in the prior art that the evaluation process fails to consider various factors such as environment and gene, thus lacking a dynamic evaluation model of various data fusion, and failing to compare the score result of the evaluation model with a preset score and adjust it when the comparison result is unqualified, thereby reducing the accuracy of evaluation.

[0006] To achieve the above-mentioned object, the present application provides a paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors, comprising:

[0007] a data monitoring unit configured to acquire current data information, wherein the data information comprises gene data, environmental exposure data and physiological monitoring data;

[0008] a historical data acquisition unit configured to acquire historical data information;

[0009] a data preprocessing unit connected with the historical data acquisition unit, configured to screen the historical data information by using a mutual information method, and fuse the screened data information with the gene data and the environmental exposure data by using a self-encoder;

[0010] a virtual data generation unit connected with the data preprocessing unit, configured to generate virtual patient data based on the preprocessed data information;

[0011] a model construction unit connected with the data preprocessing unit, configured to construct a multi-modal feature fusion module, and weight the outputs of each network in the multi-modal feature fusion module by using a gated attention mechanism to form a dynamic metabolism prediction module, wherein the multi-modal feature fusion module is a Transformer-Encoder model and a time convolution network after inputting the historical data information;

[0012] a model training unit connected with the virtual data generation unit and the model construction unit respectively, configured to train the dynamic metabolism prediction module based on the virtual patient data to obtain a gene-environment-PK / PD multidimensional model;

[0013] a metabolism evaluation unit connected with the data monitoring unit and the model training unit respectively, configured to score the metabolic efficiency based on the gene-environment-PK / PD multidimensional model according to the current data information to obtain an actual score;

[0014] an analysis unit connected with the metabolism evaluation unit, configured to determine whether the evaluation is qualified based on the difference between the actual score and a preset score, and generate a corresponding processing mode based on the unqualified reason;

[0015] An adjusting unit, connected with the analyzing unit, is configured to adjust based on the received processing mode.

[0016] Further, the model constructing unit is further configured to fuse the features according to the formula: fused features = attention weight * gene environment features + (1-attention weight) * HRV features, and weight the gene environment features output by the Transformer-Encoder model and the HRV features output by the time convolution network, where the attention weight = sigma(W_a·[gene environment features||HRV features]).

[0017] Further, the virtual data generating unit is further configured to synthesize the gene data and the environmental exposure data based on a CYP2D6 enzyme kinetics equation, and superimpose noise on the HRV signals of healthy people in a public data set to generate the virtual patient data.

[0018] Further, the model training unit is further configured to train the dynamic metabolism prediction module based on the virtual patient data, and fine-tune the model based on a transfer learning method to obtain a gene-environment-PK / PD multidimensional model.

[0019] Further, the model training unit is further configured to determine whether the results output by the model are consistent based on a cross-validation method and an adversarial verification analysis model multiple times.

[0020] Further, the analyzing unit is further configured to determine whether the evaluation is qualified based on the difference between the actual score and the preset score, and determine the reason for the unqualified evaluation based on the average value of the difference between a plurality of historical scores and corresponding preset scores or the average value of the difference between each score obtained based on the cross-validation and the preset score.

[0021] Further, the analyzing unit is further configured to generate a corresponding processing mode based on the comparison result of the average value of the difference between a plurality of historical scores and corresponding preset scores and a preset historical average value, including determining the reason for the unqualified evaluation based on the average value of the difference between each score and the preset score obtained based on the cross-validation, or issuing a notification of unqualified iteration number of the trained model.

[0022] Further, the analyzing unit is further configured to generate a corresponding processing mode based on the comparison result of the average value of the difference between each score and the preset score obtained based on the cross-validation and a preset verification average value, including adjusting the screening criteria during data preprocessing based on the difference between the average value and the preset verification average value, or adjusting the attention weight based on the average value of the variance of each dimension data.

[0023] Further, the analysis unit is further used to increase the screening criterion in data preprocessing based on the difference between the average value and the preset verification average value, and the difference is proportional to the increase amplitude of the screening criterion.

[0024] Further, the analysis unit is further used to increase the attention weight based on the average value of the variance of each dimension data, and the average value is proportional to the increase amplitude of the attention weight.

[0025] Compared with the prior art, the beneficial effects of the present application are that the system can obtain more effective data information by preprocessing the obtained historical data information, so as to train the model more effectively using the data information; at the same time, based on the data trained by the preprocessing, a dynamic metabolism prediction module formed by weighting the output of each network in the multi-modal feature fusion module using the gating attention mechanism is used to obtain a gene-environment-PK / PD multi-dimensional model, which can quantify various factors of environment and genes, and fuse multi-modal data, further realizing dynamic evaluation of metabolic efficiency; using the gene-environment-PK / PD multi-dimensional model to score the current data information and comparing the score with a preset score can quickly determine whether the model evaluation is qualified, and adjust it when it is unqualified, thereby further improving the accuracy of evaluation.

[0026] Further, the present application captures the HRV time sequence characteristics based on TCN, combines the modeling of gene and environment association based on Transformer, realizes dynamic evaluation of metabolic efficiency through multi-modal data fusion, at the same time, through the gating attention mechanism, the contribution weight of gene and environment factors (such as gene weight > environment weight) can be adaptively adjusted, avoiding the static deviation of traditional weighted average method, so that the model combines various data to score more accurately.

[0027] Further, the present application synthesizes gene data and environment exposure data based on the CYP2D6 enzyme kinetics equation, and superimposes noise on the health person HRV signal in the public data set to form virtual patient data, which can avoid relying on real clinical data, at the same time, also makes the data more diverse, and more has reference value, so as to train the model more accurately.

[0028] Further, the present application trains the dynamic metabolism prediction module based on the virtual patient data, and fine-tunes the model based on the transfer learning method, which can make the evaluation of the trained model more accurate, thereby improving the model evaluation efficiency.

[0029] Further, the present application judges whether the model is qualified according to the verification result based on the cross-validation method and the adversarial verification analysis model multiple output results, and adjusts the model according to the unqualified reason in the subsequent, so as to make the evaluation result of the model more accurate.

[0030] Further, the present application determines whether the evaluation is qualified based on the difference between the actual score and the preset score, can quickly determine whether the evaluation of the model is qualified, thereby improving the efficiency of determining the evaluation result of the model.

[0031] Further, the present application generates a corresponding processing mode based on the comparison result of the average value of the difference between the historical score and the corresponding preset score and the preset historical average value, can more accurately determine whether the model has a problem leading to the appearance of the result based on the historical score, thereby more accurately determining the reason for the unqualified evaluation result based on the comparison result.

[0032] Further, the present application generates a corresponding processing mode based on the comparison result of the average value of the difference between the historical score and the corresponding preset score and the preset historical average value, can more accurately determine whether the model has a problem leading to the appearance of the result based on the historical score, thereby more accurately determining the reason for the unqualified evaluation result based on the comparison result.

[0033] Further, the present application increases the screening criteria during data preprocessing based on the difference between the average value and the preset verification average value, can filter out some interference data during subsequent preprocessing of the data, thereby more effectively training the model, thereby making the evaluation result of the model more accurate.

[0034] Further, the present application increases the attention weight based on the average value of the variance of each dimension data, can adjust the attention weight according to the dispersion degree of the data, can make the model pay more attention to the data with high dispersion degree during the training process, thereby more effectively training the model, and further improving the accuracy of the evaluation result of the model. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The structure schematic diagram of the paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors of the embodiment of the present application;

[0036] Figure 2 The mechanism diagram based on environmental factors of the embodiment of the present application;

[0037] Figure 3 The step flow chart of the paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors of the embodiment of the present application;

[0038] Figure 4 The overall architecture diagram of the embodiment of the present application based on data construction and model training;

[0039] Figure 5 The step flow chart of the embodiment of the present application for constructing a dynamic metabolic prediction module;

[0040] Figure 6 The flow chart of the step of determining based on the comparison result of the difference between the actual score and the preset score and the preset difference value. DETAILED DESCRIPTION

[0041] In order to make the objects and advantages of the present application clearer, the present application will be further described in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0042] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and are not used to limit the protection scope of the present application.

[0043] It should be noted that in the description of the present application, unless explicitly defined and limited, the terms "mounting", "connecting", "connecting" should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication inside two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0044] Please refer to Figure 1 As shown in the figure, it is a structure schematic diagram of the paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors according to an embodiment of the present application.

[0045] The system comprises a data monitoring unit, a historical data acquisition unit, a data preprocessing unit, a virtual data generation unit, a model construction unit, a model training unit, a metabolic evaluation unit, an analysis unit and an adjustment unit.

[0046] The data monitoring unit is used to acquire current data information, wherein the data information comprises gene data, environmental exposure data and physiological monitoring data;

[0047] The historical data acquisition unit is used to acquire historical data information.

[0048] The data preprocessing unit is connected with the historical data acquisition unit, and is used to screen the historical data information by using the mutual information method, and fuse the gene data and the environmental exposure data by using the self-encoder according to the screened data information.

[0049] The virtual data generation unit is connected with the data preprocessing unit, and is used to generate virtual patient data based on the preprocessed data information.

[0050] The model construction unit is connected with the data preprocessing unit, which is used to construct a multi-modal feature fusion module, and to form a dynamic metabolism prediction module by weighting the outputs of each network in the multi-modal feature fusion module using a gated attention mechanism, wherein the multi-modal feature fusion module is a Transformer-Encoder model and a time convolution network after inputting historical data information;

[0051] The model training unit is connected with the virtual data generation unit and the model construction unit respectively, which is used to train the dynamic metabolism prediction module based on the virtual patient data to obtain a gene-environment-PK / PD multidimensional model;

[0052] The metabolism evaluation unit is connected with the data monitoring unit and the model training unit respectively, which is used to score the metabolism efficiency according to the current data information based on the gene-environment-PK / PD multidimensional model to obtain an actual score;

[0053] The analysis unit is connected with the metabolism evaluation unit, which is used to determine whether the evaluation is qualified based on the difference between the actual score and a preset score, and to generate a corresponding processing mode based on the unqualified reason;

[0054] The adjustment unit is connected with the analysis unit, which is used to adjust based on the received processing mode.

[0055] Specifically, in the embodiment, the gene data is CYP2D6 genotype (*1 / *1, *1 / *10, *10 / *10, etc.), which is detected by qPCR or NGS and coded as enzyme activity score (based on known metabolic phenotype of PharmGKB database).

[0056] Specifically, in the embodiment, as shown in Figure 2 It is an environmental factor action mechanism diagram based on the embodiment of the application. The environmental exposure data includes PM2.5 exposure amount (μg / m 3 h), which is collected in real time by a portable air quality sensor, and the daily average value; smoking index (cigarettes / day x years of smoking), which is reported by the patient himself, combined with the calibration of the exhaled CO2 detector (such as ≥6ppm for smokers); caffeine intake (mg / day), which is calculated by a diet record APP (based on the FDA standard, 1 cup of coffee ≈95mg); alcohol intake (g / day), which is converted into a standard drinking unit (1 unit ≈14g of ethanol).

[0057] Specifically, in the embodiment, the physiological monitoring data includes HRV time domain indicators, respectively SDNN (normal value >= 50ms), RMSSD (reflecting parasympathetic nerve activity); HRV frequency domain indicators, respectively LF (0.04-0.15Hz, sympathetic nerve activity), HF (0.15-0.4Hz, parasympathetic nerve activity), LF / HF ratio (sympathetic-vagal balance); the collection device is a medical grade wearable device.

[0058] Referring to Figure 3 As shown in the figure, it is a step flow chart of the paliperidone metabolism efficiency evaluation system based on CYP2D6 gene and environmental factors implemented by the embodiment of the application.

[0059] The implementation process of the paliperidone metabolism efficiency evaluation system based on CYP2D6 gene and environmental factors includes:

[0060] S1, obtaining current data information through a data monitoring unit, wherein the data information includes genetic data, environmental exposure data and physiological monitoring data;

[0061] S2, obtaining historical data information through a historical data acquisition unit;

[0062] S3, screening historical data information by a data preprocessing unit connected with the historical data acquisition unit using the mutual information method, and fusing the genetic data and the environmental exposure data using the self-encoder with the screened data information;

[0063] S4, generating virtual patient data based on the preprocessed data information through a virtual data generation unit connected with the data preprocessing unit;

[0064] S5, constructing a multi-modal feature fusion module through a model construction unit connected with the data preprocessing unit, and weighting the outputs of each network in the multi-modal feature fusion module using a gated attention mechanism to form a dynamic metabolism prediction module, wherein the multi-modal feature fusion module is a Transformer-Encoder model and a time convolution network after inputting historical data information;

[0065] S6, training the dynamic metabolism prediction module based on the virtual patient data through a model training unit connected with the virtual data generation unit and the model construction unit respectively, to obtain a gene-environment-PK / PD multi-dimensional model, as Figure 4 As shown in the figure, it is a general architecture diagram of the model constructed and trained based on data by the embodiment of the application;

[0066] S7, scoring metabolic efficiency according to current data information based on the gene-environment-PK / PD multidimensional model by a metabolic evaluation unit connected with the data monitoring unit and the model training unit respectively, to obtain an actual score;

[0067] S8, determining whether the evaluation is qualified based on a difference between the actual score and a preset score by an analysis unit connected with the metabolic evaluation unit, and generating a corresponding processing mode based on a reason for disqualification;

[0068] S9, adjusting based on the received processing mode by an adjustment unit connected with the analysis unit.

[0069] Please refer to Figure 5 The model construction unit is also used to fuse features according to attention weight x gene environment features + (1-attention weight) x HRV features based on the gated attention mechanism, and weight the gene environment features output by the Transformer-Encoder model and the HRV features output by the time convolution network, where attention weight = sigma (W_a·[gene environment features||HRV features]).

[0070] Specifically, in this embodiment, the gene environment features output by the Transformer-Encoder and the HRV time sequence features output by the time convolution network (TCN) are fused, and the outputs of the two networks are weighted by the gated attention mechanism, where the output layer is a fully connected network predicting metabolic efficiency score, and the activation function of the network is Swish.

[0071] Specifically, the virtual data generation unit is also used to synthesize the gene data and the environmental exposure data based on the CYP2D6 enzyme kinetics equation, and superimpose noise on the HRV signal of healthy people in the public data set to generate the virtual patient data.

[0072] Specifically, in this embodiment, the gene data and the environmental exposure data are synthesized based on the CYP2D6 enzyme kinetics equation (Michaelis-Menten model), where the enzyme kinetics parameters Km=15μM, V_max=0.8nmol / min / mg, and an environmental attenuation factor f(PM2.5, smoking index)=0.7 is introduced, where the CYP2D6 enzyme kinetics equation is mainly used to quantify the metabolic capacity of the enzyme to drugs, and its formula is as follows:

[0073] Metabolic rate V = (V_max * [S]) / (K_m + [S]) * gene activity score * f(PM2.5, smoking index)

[0074] In the formula, f is an environmental factor attenuation function, PM2.5 >= 75 mu g / m 3 When the attenuation is 30%.

[0075] Specifically, in the embodiment, the HRV signal of a healthy person in the PhysioNet public data set is used, and noise is superimposed to simulate patient variation, thereby ensuring the diversity of the data.

[0076] Specifically, the model training unit in the embodiment of the application is also used to train the dynamic metabolism prediction module based on the virtual patient data, and fine-tune the model based on a transfer learning method to obtain a gene-environment-PK / PD multidimensional model.

[0077] Specifically, in the embodiment, the TCN-Transformer model is trained on synthetic data, wherein the loss function is MAE, and is set to time series smoothing regularization; and a small amount of real world data (such as paliperidone pharmacokinetics, PK data in PharmGKB) is used to fine-tune the model.

[0078] Please refer to Figure 6 The actual score and the preset score are compared, and the difference value L is obtained, as shown in the step flow chart of the actual score and the preset score in the embodiment of the application. The analysis unit in the embodiment of the application is also used to determine whether the evaluation is qualified based on the difference value between the actual score and the preset score, and determine the reason for the evaluation being unqualified based on the average value of the difference value between a plurality of historical scores and corresponding preset scores or the average value of the difference value between each score and the preset score obtained based on cross-validation.

[0079] Specifically, in the embodiment, the difference value L0 can be divided into a first preset difference value L1 and a second preset difference value L2, and the first preset difference value L1 = 4 and the second preset difference value L2 = 9 in the difference value standard, and it should be noted that the values of L1 and L2 can also be determined based on the paliperidone metabolism efficiency requirement in other embodiments; the comparison process of the difference value L and L1 and L2 is as follows:

[0080] If the difference value L is less than or equal to the first preset difference value L1, it is determined that the evaluation of the model is qualified;

[0081] If the difference value L is greater than the first preset difference value L1 and less than the second preset difference value L2, it is not determined whether other factors cause the result at this time, and the average value P of the difference value between a plurality of historical scores and corresponding preset scores is used to determine whether the evaluation is qualified.

[0082] If the difference value L is greater than or equal to the second preset difference value L2, it is determined that the evaluation of the model is unqualified, and the average value Q of the difference between each score obtained based on cross-validation and the preset score is used to determine the reason for the unqualified evaluation.

[0083] Specifically, the analysis unit in the embodiment of the present application is also used to generate a corresponding processing mode based on the comparison result of the average value of the difference between a plurality of historical scores and corresponding preset scores and a preset historical average value, including determining the reason for the unqualified evaluation based on the average value of the difference between each score obtained based on cross-validation and the preset score, or issuing a notification of the unqualified number of iterations of the training model.

[0084] Specifically, in the present embodiment, the preset historical average value P0=4, and the comparison process of the average value P of the difference between a plurality of historical scores and corresponding preset scores and the preset historical average value P0 is specifically as follows:

[0085] If the average value P is less than or equal to the preset historical average value P0, it indicates that the model does not have evaluation errors, and it is determined that there is a problem with the data used in the present evaluation process, resulting in inaccurate evaluation results, and the average value Q of the difference between each score obtained based on cross-validation and the preset score is used to determine the reason for the unqualified evaluation;

[0086] If the average value P is greater than the preset historical average value P0, it is determined that the model is unqualified, and a notification of the unqualified number of iterations of the training model is issued.

[0087] Specifically, the analysis unit in the embodiment of the present application is also used to generate a corresponding processing mode based on the comparison result of the average value of the difference between each score obtained based on cross-validation and the preset score and a preset verification average value, including adjusting the screening criteria during data preprocessing based on the difference between the average value and the preset verification average value, or adjusting the attention weight based on the average value of the variance of each dimension data.

[0088] Specifically, in the present embodiment, the preset verification average value Q0=3, and the comparison process of the average value Q of the difference between each score obtained based on cross-validation and the preset score and the preset verification average value Q0 is specifically as follows:

[0089] If the average value Q is less than or equal to the preset verification average value Q0, it indicates that there is a problem with the screening criteria in the data preprocessing, and the screening criteria during data preprocessing is adjusted based on the difference R between the average value and the preset verification average value;

[0090] If the average value Q is greater than the preset verification average value Q0, it indicates that the degree of dispersion in the collected data is large, and the attention weight is adjusted based on the average value T of the variance of each dimension data.

[0091] Specifically, the analysis unit in the embodiment of the present application is also used to increase the screening criterion in data preprocessing based on the difference between the average value and the preset verification average value, and the increase of the difference and the screening criterion is directly proportional.

[0092] Specifically, in the embodiment, the preset difference R0=1, and the comparison process between the difference R between the average value and the preset verification average value and the preset difference R0 is specifically as follows:

[0093] If the difference R is less than or equal to the preset difference R0, the screening criterion is adjusted to 1.2 times of the original screening criterion.

[0094] If the difference R is greater than the preset difference R0, the screening criterion is adjusted to 1.5 times of the original screening criterion.

[0095] Specifically, the analysis unit in the embodiment of the present application is also used to increase the attention weight based on the average value of the variance of each dimension data, and the increase of the average value and the attention weight is directly proportional.

[0096] Specifically, in the embodiment, the preset average value T0=1.8, and the comparison process between the average value T of the variance of each dimension data and the preset average value T0 is specifically as follows:

[0097] If the average value T is less than or equal to the preset average value T0, the attention weight is adjusted to 1.4 times of the original attention weight.

[0098] If the average value T is greater than the preset average value T0, the attention weight is adjusted to 1.9 times of the original attention weight.

[0099] In order to make the purpose and advantages of the present application more clear and obvious, the present application is further described below in combination with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the present application.

[0100] Embodiment 1

[0101] After the data information of the patient is input into the gene-environment-PK / PD multidimensional model, the actual score output is 67, and the preset score of the metabolic efficiency of the patient is 55, at this time, it is determined that the evaluation of the model is unqualified, then, in the 10-fold cross-validation, the difference between each actual score and the preset score is calculated respectively, and the difference is 3, 5, 6, 2, 1, 4, 6, 3, 4, 2, the average value of these differences is 3.6, which indicates that the dispersion degree in the collected data is large, and then the attention weight is adjusted based on the average value of the variance of different kinds of data.

[0102] Embodiment 2

[0103] The process of scoring the metabolic efficiency of the simulated patient using the model is as follows:

[0104] First, simulate patient parameters: genotype: CYP2D6*10 / *10 (enzyme activity score = 0.5); environmental exposure: PM2.5 daily average = 85 μg / m 3 (past 7-day sliding average), smoking index = 20 cigarettes / day x 10 years; HRV indicator: LF / HF = 3.5 (high sympathetic nervous activity); clinical data: liver function Child-Pugh A, BMI = 24.

[0105] Second, model processing flow: first step, data input: genotype, PM2.5, smoking index, HRV, etc. data input system; second step, metabolic score calculation: genetic contribution: 0.5 (based on PharmGKB metabolic phenotype), environmental attenuation factor: PM2.5>75 μg / m 3 , then metabolic efficiency x 0.7; smoking index>150, then metabolic efficiency x 0.8; final score: 0.5 x 0.7 x 0.8 = 0.28 (28 / 100), the actual score output is 28.

[0106] Finally, dose recommendation: the basic dose (oral) is 6 mg / day; the corrected dose: 6 mg x (1 / 0.28) ≈ 21.4 mg / day.

[0107] System prompt: need to be combined with TDM verification, and suggest reducing PM2.5 exposure.

[0108] Technical effect verification: traditional genotyping (slow metabolizer) recommended dose: ≤4 mg / day, which may lead to insufficient efficacy. The system recommended dose: 21.4 mg / day, which is more in line with the actual metabolic efficiency (simulated blood drug concentration compliance rate increased by 40%).

[0109] So far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the drawings, but those skilled in the art will readily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without departing from the principles of the present application, and the technical solutions after such changes or replacements will fall within the protection scope of the present application.

[0110] The above is only the preferred embodiment of the present application and is not intended to limit the present application; for those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included within the protection scope of the present application.

Claims

1. A paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors, characterized by, The method comprises the following steps: a data monitoring unit is used to obtain current data information, wherein the data information comprises genetic data, environmental exposure data and physiological monitoring data; a historical data acquisition unit is used to obtain historical data information; a data preprocessing unit connected with the historical data acquisition unit is used to screen the historical data information by using the mutual information method, and fuse the screened data information with the genetic data and the environmental exposure data by using a self-encoder; a virtual data generation unit connected with the data preprocessing unit is used to generate virtual patient data based on the preprocessed data information; a model construction unit connected with the data preprocessing unit is used to construct a multi-modal feature fusion module, and weight the outputs of each network in the multi-modal feature fusion module by using a gated attention mechanism to form a dynamic metabolism prediction module, wherein the multi-modal feature fusion module is a Transformer-Encoder model and a time convolution network after inputting the historical data information; a model training unit connected with the virtual data generation unit and the model construction unit is used to train the dynamic metabolism prediction module based on the virtual patient data to obtain a gene-environment-PK / PD multidimensional model; a metabolism evaluation unit connected with the data monitoring unit and the model training unit is used to score the metabolic efficiency based on the gene-environment-PK / PD multidimensional model according to the current data information to obtain an actual score; an analysis unit connected with the metabolism evaluation unit is used to determine whether the evaluation is qualified based on the difference between the actual score and a preset score, and generate a corresponding processing mode based on the reason for unqualification; an adjustment unit connected with the analysis unit is used to adjust based on the received processing mode.

2. The paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors according to claim 1, characterized by, The model construction unit is also used to weight the gene environment feature output by the Transformer-Encoder model and the HRV feature output by the time convolution network based on the gated attention mechanism according to the fusion feature = attention weight x gene environment feature + (1-attention weight) x HRV feature, wherein the attention weight = σ(W_a·[gene environment feature||HRV feature]). 3.The system for evaluating paliperidone metabolic efficiency based on CYP2D6 gene and environmental factors according to claim 1, wherein, The virtual data generation unit is also used to synthesize the genetic data and the environmental exposure data based on the CYP2D6 enzyme kinetics equation, and to superimpose noise on the HRV signal of healthy people in a public data set to generate the virtual patient data.

4. The paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors according to claim 1, characterized by, The model training unit is also used to train the dynamic metabolism prediction module based on the virtual patient data, and fine-tune the model based on a transfer learning method to obtain a gene-environment-PK / PD multidimensional model.

5. The paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors according to claim 1, characterized by, The model training unit is also used to determine whether the output results of the model are consistent by using a cross-validation method and an adversarial verification analysis model multiple times.

6. The paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors according to claim 1, characterized by, The analysis unit is further configured to determine whether the evaluation is qualified based on a difference between the actual score and the preset score, and determine a reason for the evaluation being unqualified based on an average of differences between a plurality of historical scores and corresponding preset scores or an average of differences between each score obtained based on cross-validation and the preset score.

7. The paliperidone metabolic efficiency evaluation system based on CYP2D6 gene and environmental factors according to claim 6, characterized by, The analysis unit is further configured to generate a corresponding processing manner based on a comparison result between the average of differences between the plurality of historical scores and the corresponding preset scores and a preset historical average, including determining a reason for the evaluation being unqualified based on the average of differences between each score obtained based on cross-validation and the preset score, or issuing a notification that an iteration number of the trained model is unqualified. 8.The system for evaluating paliperidone metabolic efficiency based on CYP2D6 gene and environmental factors according to claim 6, wherein, The analysis unit is further configured to generate a corresponding processing manner based on a comparison result between the average of differences between each score obtained based on cross-validation and the preset score and a preset verification average, including adjusting a screening criterion during data preprocessing based on a difference between the average and the preset verification average, or adjusting an attention weight based on an average of variances of the dimensional data. 9.The system for evaluating paliperidone metabolic efficiency based on CYP2D6 gene and environmental factors according to claim 8, wherein, The analysis unit is further configured to increase the screening criterion during data preprocessing based on the difference between the average and the preset verification average, and a positive correlation between the difference and an increase amplitude of the screening criterion. 10.The system for evaluating paliperidone metabolic efficiency based on CYP2D6 gene and environmental factors according to claim 8, wherein, The analysis unit is further configured to increase the attention weight based on the average of variances of the dimensional data, and a positive correlation between the average and an increase amplitude of the attention weight.

Images