Yak breeding excellent gene traceability positioning method and system based on deep learning

Abstract

The invention relates to the technical field of livestock breeding gene traceability positioning data analysis, and discloses a yak breeding excellent gene traceability positioning method and system based on deep learning, and the method comprises the steps: obtaining a six-dimensional environment stress factor time sequence data set and a five-dimensional breeding physiological stress response index set; a breeding hormone response model is constructed, and a prediction result is generated; dynamically adjusting an injection parameter set of gonadotropin release hormone and cloprostenol in the estrus synchronization treatment scheme, and forming and executing an individualized hormone regulation instruction; in a preset time range, performing artificial insemination by measuring blood flow parameters of the uterine artery of the yak in combination with the prediction result; a calf biological sample is collected, genome desoxyribonucleic acid is extracted, candidate gene locus typing detection is carried out, and an excellent gene traceability positioning analysis report is output. According to the invention, the integration of environmental adaptability, stress state monitoring and breeding technology precision cooperation is realized, and the breeding efficiency of yaks in a remote grazing mode is effectively optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of livestock breeding gene traceability positioning data analysis, more specifically, it relates to a yak breeding excellent gene traceability positioning method and system based on deep learning. BACKGROUND

[0002] With the continuous development of plateau animal husbandry, yak, as the core economic livestock species in the Qinghai-Tibet Plateau and surrounding alpine pastoral areas, its reproductive efficiency and production performance directly affect the income increase of regional farmers and herdsmen and the sustainable development of industry. However, the traditional extensive grazing management mode has led to long reproductive cycle, low estrus rate, poor calf survival rate and other problems for a long time.

[0003] In recent years, some regions have tried to improve the nutritional status of female yaks through off-site straw borrowing mode. However, environmental changes can trigger stress response, leading to dysfunction of hypothalamic-pituitary-gonadal axis, and reducing the physiological response efficiency of exogenous reproductive hormones, which is manifested as a decrease in the synchronization rate of estrus and a fluctuation in the pregnancy rate of artificial insemination, making it difficult to achieve the expected efficiency target.

[0004] The existing technology generally has the problems of heavy hormone intervention and light environmental adaptation, heavy technology stacking and light system coordination, and lacks dynamic evaluation and closed-loop control mechanism for environmental stress and physiological state of cows.

[0005] Therefore, how to integrate environmental adaptability, stress dynamic monitoring and precise coordination of reproductive technology has become a technical problem to be solved for optimizing the reproductive efficiency and economic benefit of yak under the off-site borrowing mode. SUMMARY

[0006] The present application provides a yak breeding excellent gene traceability positioning method and system based on deep learning, which solves the technical problem of how to integrate environmental adaptability, stress dynamic monitoring and precise coordination of reproductive technology in the prior art, and optimizes the reproductive efficiency and economic benefit of yak under the off-site borrowing mode.

[0007] The present application provides a yak breeding excellent gene traceability positioning method and system based on deep learning, which includes: In a first aspect, the yak breeding excellent gene traceability positioning method based on deep learning includes: Real-time acquisition of multiple yak breeding environment data and multiple reproductive physiological stress response indexes, respectively forming six-dimensional environmental stress factor time series data set and five-dimensional reproductive physiological stress response index set; Specifically, an environmental perception sensor array is deployed in the yak grazing area, which includes a temperature sensor, a humidity sensor, an illumination intensity sensor, a wind speed sensor, an ammonia concentration sensor, and a noise sensor. Real-time time series data of temperature, humidity, illumination intensity, wind speed, ammonia concentration, and noise decibel value are collected to form a six-dimensional environmental stress factor time series dataset. Meanwhile, a subcutaneous physiological monitoring device and a non-contact infrared thermal imager are used to obtain individual yak reproductive physiological stress response indicators, including cortisol concentration, heart rate variability, respiratory rate, rectal temperature, and body surface heat distribution map, to construct a five-dimensional reproductive physiological stress response indicator set. The six-dimensional environmental stress factor time series dataset and the five-dimensional reproductive physiological stress response indicator set are input into a reproductive hormone response model to generate a prediction result of the reproductive hormone response efficiency of the individual yak. The reproductive hormone response model is a multi-modal deep neural network model using a dual-branch convolutional long short-term memory network architecture. The environmental branch extracts local features and models time series dependence, while the physiological branch analyzes spatial correlation and predicts dynamic trends for multiple physiological indicators. The outputs of the two branches are fused through an attention weight layer for cross-modal feature interaction, and finally the prediction result of the reproductive hormone response efficiency is output. Based on the prediction result, the injection parameters of gonadotropin-releasing hormone and cloprostenol in the synchronized estrus treatment plan are dynamically adjusted, and individualized hormone control instructions are executed. The injection parameters include injection dose, injection interval time, and injection site depth. The dynamic adjustment is based on the prediction result of the reproductive hormone response efficiency within a preset parameter adjustment range, and individualized hormone control instructions are executed through intelligent injection equipment. Within a preset time range after executing the control instructions, the optimal insemination time window is obtained by measuring the uterine artery blood flow parameters of the yak and combining the prediction result, and artificial insemination is performed. The uterine artery blood flow parameters include blood flow velocity and pulsatility index. The optimal insemination time window is obtained by weighting the prediction result of the reproductive hormone response efficiency and the blood flow parameters, ensuring that artificial insemination is performed under optimal physiological conditions. After artificial insemination is performed, a yak calf biological sample is obtained and genomic deoxyribonucleic acid is extracted. High-throughput single nucleotide polymorphism typing technology is used to genotype and detect the pre-set candidate gene sites to obtain a genotyping result. The biological sample is an ear tissue sample collected using sterile biopsy forceps and immediately immersed in a lysis buffer containing protease K; the genomic deoxyribonucleic acid is extracted using a magnetic bead method; the candidate gene sites include gene sites related to growth, immunity and stress response, and specifically include growth hormone gene exons, insulin-like growth factor gene promoter regions, interleukin gene introns, heat shock protein gene exons and other gene sites related to growth rate, feed conversion efficiency and disease resistance; The typing detection result is associated with the yak individual identification, the father information, the environmental stress factor data and the reproductive physiological index and is stored to construct a gene traceability positioning database; The gene traceability positioning database is stored in a heterogeneous graph structure, the node types include calf nodes, cow nodes, father nodes, gene site nodes and environmental factor nodes, the edge types include parent-child relationship edges, genotype association edges and environmental exposure edges, and the multi-dimensional data is associated and quickly searched; The graph neural network is used for relationship reasoning on the gene traceability positioning database, the interaction mode of the key gene combination and the environmental stress factor is identified, and an excellent gene traceability positioning analysis report is output; the identification criterion of the interaction mode is that if the appearance frequency of a certain gene combination in a sample interval in which each item of environmental stress factor data is higher than 75% of the historical data is greater than or equal to 80%, and the corresponding calf daily gain is greater than or equal to 1.2 kg, the gene combination is determined as an excellent gene combination; The graph neural network includes three layers of graph convolution layers, the range threshold of aggregating neighbor node features of each layer is set, the mean aggregation is used as the aggregation function, and the rectified linear unit is used as the activation function; the final output layer uses a soft maximum function to generate the probability distribution of the key gene combination; based on the reasoning result of the graph neural network, the excellent gene combination is determined according to the appearance frequency of the gene combination under specific environmental conditions and the corresponding production performance; Further, the specific configuration of the environmental perception sensor array is that the temperature sensor adopts a platinum resistance thermometer, the measurement range is -40 DEG C to 80 DEG C, and the accuracy is ± 0.5 DEG C; the humidity sensor adopts a capacitive polymer humidity sensitive element, the measurement range is 5% to 95% relative humidity, and the accuracy is ± 3%; the light intensity sensor adopts a silicon photodiode, the spectral response range is 400 nm to 1100 nm, and the range is 0 to 200,000 lux; the wind speed sensor adopts a three-cup anemometer, the starting wind speed is not greater than 0.5 m / s, the measurement range is 0.5 m / s to 30 m / s; the ammonia concentration sensor adopts an electrochemical principle, the detection range is 0 to 100 ppm, and the resolution is 0.1 ppm; the noise sensor adopts an electret condenser microphone, the frequency response range is 20 Hz to 12,000 Hz, and the measurement range is 30 dB to 130 dB; Further, the subcutaneous implantable physiological monitoring device comprises a miniature cortisol biosensor, a heart rate variability acquisition electrode and a wireless radio frequency transmission unit, wherein the miniature cortisol biosensor adopts a molecularly imprinted polymer film as a recognition layer, with a detection lower limit of 0.1 ng / ml and a response time of 30 seconds; the heart rate variability acquisition electrode adopts a silver-silver chloride material, with a sampling frequency of 250 Hz; the wireless radio frequency transmission unit has a working frequency band of 433 MHz, with a transmission distance of not less than 50 meters; the non-contact infrared thermal imager has a spatial resolution of 0.5 milliradians, a temperature resolution of 0.05℃, a frame frequency of 30 Hz, and a detection wave band of 8 μm to 14 μm; Further, the environment branch of the multi-modal deep neural network model comprises three one-dimensional convolution layers, with convolution kernel sizes of 3, 5 and 7 respectively, a step length of 1 and a same padding mode; each layer is connected with a batch normalization layer and a rectified linear unit activation function, and then connected with two long short-term memory units, with a hidden unit number of 128; the physiological branch comprises two three-dimensional convolution layers, with a first layer having a convolution kernel size of 3×3×3 and a second layer having a convolution kernel size of 5×5×5, a step length of 1 and a same padding mode; each layer is connected with a batch normalization layer and a rectified linear unit activation function, and then connected with a bidirectional long short-term memory unit, with a hidden unit number of 256; the attention weight fusion layer adopts an additive attention mechanism, with a query vector generated from the final hidden state of the environment branch, a key vector and a value vector generated from the final hidden state of the physiological branch, and an output being a fusion feature vector after weighted summation, with a dimension of 512; Further, the injection dose of the gonadotropin-releasing hormone is adjusted in a range of 50 μg to 200 μg, the injection interval time is adjusted in a range of 72 hours to 120 hours, and the injection site depth is adjusted in a range of 0.5 cm subcutaneously to 1.5 cm intramuscularly; the injection dose of the cloprostenol is adjusted in a range of 0.1 mg to 0.5 mg, the injection interval time is adjusted in a range of 48 hours to 96 hours, and the injection site depth is adjusted in a range of 0.3 cm subcutaneously to 1 cm intramuscularly; the dose adjustment step length is 10 μg or 0.05 mg, the time adjustment step length is 12 hours, and the depth adjustment step length is 0.1 cm; Further, the uterine artery blood flow detection adopts a Doppler ultrasound blood flow detector, with a probe frequency of 7.5 MHz, a sampling volume of 1.5 mm and an angle correction range of ±60°; the determination threshold of uterine artery blood flow velocity is ≥30 cm / s, and the determination threshold of pulsatility index is ≤1.2; the calculation formula of the optimal insemination time window is: insemination time = reproductive hormone response efficiency prediction result × 24 hours + weighted score value × 6 hours, wherein the weighted score value = (blood flow velocity / 40 cm / s) + (1.2-pulsatility index) / 2.4; Further, the high-throughput single nucleotide polymorphism typing adopts a gene chip platform, the chip probe density is 1 million per square centimeter, and the candidate gene sites cover no less than 500; the candidate gene sites include the 17th nucleotide of the 3rd exon of the growth hormone gene, the 234th nucleotide of the promoter region of the insulin-like growth factor 1 gene, the 98th nucleotide of the 5th intron of the interleukin 6 gene, and the 45th nucleotide of the 3rd exon of the heat shock protein 70 gene; the purity of the extracted genomic DNA is greater than or equal to 1.8, and the concentration is greater than or equal to 50 ng / μl; Further, the identification criterion of the interaction mode is that if the occurrence frequency of a certain gene combination in a high value interval of the environmental stress factor data is greater than or equal to 80% and the corresponding daily weight gain of the calf is greater than or equal to 1.2 kg, the gene combination is determined as an excellent gene combination; the high value interval is defined as a sample interval in which each environmental stress factor data is higher than 75% of the historical data; the preset threshold of the occurrence frequency is 80%; and the preset threshold of the daily weight gain quality is 1.2 kg.

[0008] Further, the heterogeneous graph structure of the gene traceability positioning database is taken as the input of the excellent gene positioning analysis model, the excellent gene positioning analysis model takes the excellent evaluation result of each gene combination as the output, the correlation between the excellent evaluation result and the actual production performance is taken as the prediction target, the difference between the excellent evaluation result and the prediction target is taken as the prediction error, and the square sum of the prediction error is taken as the training target; the excellent gene positioning analysis model is trained until the square sum of the prediction error converges. The output result of the excellent gene positioning analysis model is taken as the input of the report generation module, the report generation module takes the detailed analysis report of each gene combination as the output, the correlation between the detailed analysis report and the user demand matching degree is taken as the prediction target, the difference between the detailed analysis report and the prediction target is taken as the report generation error, and the square sum of the report generation error is taken as the training target to train until the square sum of the report generation error converges.

[0009] In a second aspect, the yak breeding excellent gene traceability positioning system based on deep learning comprises: The data acquisition module is configured to acquire a plurality of yak breeding environmental data and a plurality of reproductive physiological stress response indexes in real time, and form a six-dimensional environmental stress factor time series data set and a five-dimensional reproductive physiological stress response index set, respectively. The cultivation prediction module is connected with the data acquisition module, and is used for inputting a six-dimensional environmental stress factor time series data set and a five-dimensional reproductive physiological stress response index set into a reproductive hormone response model to generate a prediction result of reproductive hormone response efficiency of the yak individual; based on the prediction result, injection parameter sets of gonadotropin releasing hormone and cloprostenol in the synchronization estrus treatment scheme are dynamically adjusted to form and execute individualized hormone regulation instructions; The insemination optimization module is used for calculating an optimal insemination time window and executing artificial insemination by measuring yak uterine artery blood flow parameters within a preset time range after hormone regulation and in combination with the prediction result; The gene tracing module is used for obtaining a calf biological sample and extracting genomic deoxyribonucleic acid after execution of artificial insemination, performing typing detection on preset candidate gene sites by using a high-throughput single nucleotide polymorphism typing technology to obtain a typing detection result; the typing detection result is stored in association with yak individual identification, paternal information, environmental stress factor data and reproductive physiological indexes to construct a gene tracing positioning database; The graph reasoning module is used for performing relationship reasoning on the gene tracing positioning database by using a graph neural network, identifying interaction modes of key gene combinations and environmental stress factors, and outputting an excellent gene tracing positioning analysis report.

[0010] The present application has the advantages that the present application realizes systematic and intelligent management of the whole cycle of yak reproduction by constructing a complete technical closed loop from environment monitoring, physiological sensing, intelligent prediction, precise regulation to gene tracing.

[0011] Individualized precise intervention: the hormone response efficiency of individual female yaks is dynamically predicted by a multi-modal deep learning model, and individualized precise regulation of hormone injection parameters and insemination time is realized accordingly, which effectively improves the synchronization estrus rate and the conception rate.

[0012] Environmental gene association tracing: the complex association among genotypes, phenotypes and environmental factors is mined by using a graph neural network, which can accurately identify gene combinations that can still perform well under specific environmental stress, and provides precise data support and decision basis for molecular selection of yaks.

[0013] Improving industrial benefits: by improving reproductive efficiency, shortening the interval between births, and cultivating excellent offspring, the economic benefits of yak breeding are improved, and the sustainable development of plateau animal husbandry is promoted. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 is a deep learning-based yak cultivation excellent gene tracing positioning method process schematic diagram provided in the embodiments of the present application; Figure 2FIG. 1 is a schematic diagram of a deep learning-based yak breeding fine gene tracing positioning system module provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is merely meant to provide a better understanding of the subject matter described herein and can be changed in function and arrangement without departing from the scope of the present disclosure. Various examples can omit, substitute, or add various procedures or components as appropriate. Also, it should be understood that some features described with respect to one example can be combined in other examples.

[0016] A deep learning-based yak breeding fine gene tracing positioning method and system are disclosed in at least one embodiment of the present application, comprising: As shown in Figure 1 The deep learning-based yak breeding fine gene tracing positioning method comprises the following steps: Step 1: Real-time acquisition of multiple yak breeding environment data and multiple reproductive physiological stress response indicators, respectively forming a six-dimensional environment stress factor time series dataset and a five-dimensional reproductive physiological stress response indicator set; Specifically, an array of environment perception sensors is deployed in the yak grazing area to collect temperature, humidity, light intensity, wind speed, ammonia concentration, and noise decibel value in real time, forming a six-dimensional environment stress factor time series dataset. Synchronously, through subcutaneous implantable physiological monitoring devices (collecting cortisol concentration and heart rate variability) and non-contact infrared thermal imaging instruments (collecting body surface heat distribution map), combined with rectal temperature probes and respiration rate sensors, multiple reproductive physiological stress response indicators of yak individuals are obtained, and a five-dimensional reproductive physiological stress response indicator set is constructed. All sensor data are synchronized and aggregated through a 433MHz wireless radio frequency network with a unified timestamp, ensuring that the environment data and reproductive physiological stress response indicators are strictly aligned in the time dimension.

[0017] Step 2: Input the six-dimensional environment stress factor time series dataset and the five-dimensional reproductive physiological stress response indicator set into the reproductive hormone response model to generate a prediction result of the reproductive hormone response efficiency of the yak individual; based on the prediction result, dynamically adjust the injection parameter set of gonadotropin-releasing hormone and cloprostenol in the same estrus treatment scheme to form and execute individualized hormone regulation instructions; Specifically, the reproductive hormone response model is a multi-modal deep neural network model adopting a double-branch convolutional long short-term memory network architecture. In the training process, the model is trained using a historical data set containing a large amount of environmental time series data, physiological indicator data, and corresponding reproductive hormone response effectiveness label values actually measured by radioimmunoassay. The training goal is to minimize the mean square error between the model prediction value and the true label value.

[0018] Model application and regulation: real-time six-dimensional environmental data and five-dimensional reproductive physiological stress response indicators are input into the trained model to generate a predicted result of the reproductive hormone response effectiveness of the individual female yak (between 0 and 1). Based on this prediction value, the central controller dynamically adjusts the injection dose, interval time, and site depth of gonadotropin-releasing hormone (GnRH) and cloprostenol within a preset parameter range. The adjustment instructions are issued to a handheld smart injector through an encrypted Bluetooth protocol, and the injector has a dose feedback and execution confirmation mechanism to ensure accurate execution of the instructions.

[0019] Step 3: Within a preset time range after hormone regulation, the optimal insemination time window is calculated based on the measured uterine artery blood flow parameters and the predicted result, and artificial insemination is performed; Specifically, within a preset time (such as 24 hours) after hormone regulation, the Doppler ultrasound blood flow detector is used to measure the uterine artery blood flow velocity and pulsatility index of the female yak. The measured values and the predicted result of the reproductive hormone response effectiveness are used to calculate the optimal insemination time window using the following formula: insemination time = predicted value x 24 hours + weighted score value x 6 hours; weighted score value = (blood flow velocity / 40 cm / s) + (1.2-pulsatility index) / 2.4; the calculation result is pushed to the technical personnel through the mobile terminal to guide the artificial insemination operation.

[0020] Step 4: After artificial insemination is performed, biological samples of the calf are obtained, and genomic deoxyribonucleic acid is extracted. High-throughput single nucleotide polymorphism typing technology is used to perform typing detection on the preset candidate gene sites to obtain a typing detection result. The typing detection result is associated with the yak individual identification, paternal information, environmental stressor historical data, and reproductive physiological indicators, and is stored to construct a gene traceability positioning database. Specifically, within 72 hours after the birth of the calf, an ear tissue sample is collected aseptically, and high-quality genomic DNA is extracted using a magnetic bead method. The pre-set candidate gene sites (such as key sites of genes such as GH, IGF-1, IL-6, and HSP70) are typed and detected using a high-throughput SNP gene chip. The production performance data (such as daily weight gain) are obtained by periodically recording the standardized weight of the calf, and are associated with the genotype data through a unique identifier. The typing detection results, production performance data, and the mother buffalo individual identifier, the father frozen semen information, the historical environmental stress factor data, and the reproductive physiological indicators are cleaned, standardized, and associatedly stored to construct a structured gene traceability positioning database.

[0021] Step 5: Relationship reasoning on the gene traceability positioning database is performed through a graph neural network, the interaction mode of key gene combinations and environmental stress factors is identified, and an excellent gene traceability positioning analysis report is output.

[0022] Specifically, relationship reasoning on the gene traceability positioning database is performed through a graph neural network. Graph construction: the database is converted into a heterogeneous graph, the nodes include calves, cows, fathers, gene sites, environmental factors, etc., and the edges represent the parent-child, genotype, environmental exposure, and other relationships therebetween.

[0023] Model training and reasoning: the graph neural network includes multiple layers of graph convolution layers, and a mean aggregation function is used. The model takes the gene combination and the corresponding standardized recorded production performance (such as daily weight gain) under the environment as training data, and is trained to identify gene combinations that have a high frequency of occurrence in a specific environment and excellent production performance in a high value interval.

[0024] Identification criteria, for example: if the frequency of occurrence of a certain gene combination in a sample interval in which each environmental stress factor data is higher than 75% of the historical data is ≥80% and the corresponding offspring calf daily weight gain is ≥1.2 kg, it is determined to be an excellent gene combination.

[0025] Report generation: the system automatically generates an excellent gene traceability positioning analysis report, which includes a list of excellent gene combinations, the best fitted environmental interval, the expected production performance range, and the recommended mating strategy, and is output through a printing device and a network interface.

[0026] In a preferred embodiment of the present application, the environmental perception sensor array comprises a temperature sensor, a humidity sensor, an illumination intensity sensor, a wind speed sensor, an ammonia concentration sensor, and a noise sensor. The temperature sensor is a platinum resistance thermometer with a measurement range of negative forty degrees Celsius to positive eighty degrees Celsius and an accuracy of plus or minus zero point five degrees Celsius, mounted one meter and five centimeters above the ground and fixed on a support to avoid direct contact with the ground affecting the measurement results. The humidity sensor is a capacitive polymer humidity sensitive element with a measurement range of 5% to 95% relative humidity and an accuracy of plus or minus 3%, mounted side by side with the temperature sensor on the same support to ensure consistency of data collection. The illumination intensity sensor is a silicon photodiode with a spectral response range of four hundred nanometers to one thousand one hundred nanometers and a range of zero to two hundred thousand lux, mounted on the top of the support to receive the maximum light area. The wind speed sensor is a three-cup anemometer with a starting wind speed not greater than zero point five meters per second and a measurement range of zero point five meters per second to thirty meters per second, mounted at the highest point of the support to reduce the impact of obstacles on wind speed measurement. The ammonia concentration sensor uses electrochemical principle with a detection range of zero to one hundred ppm and a resolution of zero point one ppm, mounted one point two meters above the ground to be close to the breathing height of the cow. The noise sensor uses an electret condenser microphone with a frequency response range of twenty hertz to twelve thousand hertz and a measurement range of thirty decibels to one hundred thirty decibels, mounted in the middle of the support and kept a certain distance from other sensors to avoid mutual interference. All sensors synchronize time stamp through a four hundred thirty-three megahertz wireless radio frequency network and form a six-dimensional environmental stress factor time series data stream after convergence, with the wireless radio frequency module installed inside each sensor and communicating with the central controller through an antenna.

[0027] In terms of individual physiological monitoring of cows, the subcutaneously implanted physiological monitoring device and the non-contact infrared thermal imager jointly constitute the collection system of the reproductive physiological stress response index set. The subcutaneously implanted physiological monitoring device includes a miniature cortisol biosensor, a heart rate variability collection electrode, and a wireless radio frequency transmission unit. The miniature cortisol biosensor uses a molecularly imprinted polymer film as the recognition layer, with a detection lower limit of 0.1 ng / mL, a response time of 30 seconds, and an implantation position of about 0.5 cm deep under the skin behind the cow's scapula and fixed by medical glue. The heart rate variability collection electrode uses silver chloride silver material, with a sampling frequency of 250 Hz. The electrode patch is respectively pasted on the left side of the cow's chest near the heart and under the right scapula to form a differential signal collection channel. The wireless radio frequency transmission unit has a working frequency of 433 MHz, a transmission distance of not less than 50 meters, and is installed subcutaneously on the back of the cow and connected to the sensor through a flexible lead. The non-contact infrared thermal imager has a spatial resolution of 0.5 milliradians, a temperature resolution of 0.05 degrees Celsius, a frame frequency of 30 Hz, and a detection waveband of 8-14 microns. It is installed on a fixed bracket two meters away from the cow and adjusted by a pan-tilt to cover the whole body of the cow. The rectal temperature is measured by a digital temperature probe with an accuracy of ±0.1 degrees Celsius, inserted 10 cm deep into the cow's rectum and fixed to the tail to prevent slipping out. The respiratory rate is detected by a flexible piezoelectric film sensor, which is attached to the left side of the cow's chest cavity outside through a zero-crossing algorithm to calculate the number of breaths per minute. The sensor is connected to the data acquisition terminal through a lead. All reproductive physiological stress response indicators are aligned with environmental data by time stamp to construct a five-dimensional reproductive physiological stress response index set.

[0028] The multi-modal deep neural network model is the core processing unit of the application, and its double-branch convolutional long short-term memory network architecture processes the environmental stress factor time series data stream and the set of reproductive physiological stress response indicators respectively. The environmental branch includes three one-dimensional convolutional layers, the convolution kernel sizes are three, five and seven respectively, the step length is one, and the padding mode is the same padding. Each layer is followed by a batch normalization layer and a rectified linear unit activation function, and then connected to two layers of long short-term memory units, with 128 hidden units. The input end is connected to the environmental perception sensor array through a data interface. The physiological branch includes two three-dimensional convolutional layers, the first layer has a size of three by three by three, and the second layer has a size of five by five by five. The step length is one, and the padding mode is the same padding. Each layer is followed by a batch normalization layer and a rectified linear unit activation function, and then connected to a layer of bidirectional long short-term memory unit, with 256 hidden units. The input end is connected to the physiological monitoring device through a data interface. The attention weight fusion layer adopts an additive attention mechanism, the query vector is generated from the final hidden state of the environmental branch, the key vector and the value vector are generated from the final hidden state of the physiological branch, and the output is a fused feature vector after weighted summation, with a dimension of 512. After the fused feature vector passes through two fully connected layers and a rectified linear unit activation, the prediction result of the single-value reproductive hormone response effectiveness is output, with a value range of zero to one. The output end is connected to the central controller through a data interface.

[0029] After receiving the prediction result of the reproductive hormone response effectiveness, the central controller dynamically adjusts the hormone injection parameters in the same period estrus processing scheme. The injection dose of gonadotropin-releasing hormone is adjusted in the range of 50 to 200 micrograms, the injection interval time is adjusted in the range of 72 to 120 hours, the injection site depth is adjusted in the range of 0.5 cm subcutaneous to 1.5 cm intramuscular, the dose adjustment step is 10 micrograms, the time adjustment step is 12 hours, and the depth adjustment step is 0.1 cm. The injection dose of cloprostenol is adjusted in the range of 0.1 to 0.5 mg, the injection interval time is adjusted in the range of 48 to 96 hours, the injection site depth is adjusted in the range of 0.3 cm subcutaneous to 1 cm intramuscular, the dose adjustment step is 0.05 mg, the time adjustment step is 12 hours, and the depth adjustment step is 0.1 cm. The adjustment parameters are generated by the central controller to issue instructions to the handheld intelligent injector through a wireless network, and the handheld intelligent injector receives the instructions through the Bluetooth module and controls the injection pump to complete the accurate injection.

[0030] The Doppler ultrasound blood flow detector measures the uterine artery blood flow velocity and pulsatility index of the cow 24 hours before artificial insemination. The probe frequency of the Doppler ultrasound blood flow detector is 7.5 MHz, the sampling volume is 1.5 mm, the angle correction range is ±60 degrees, the probe is fixed on the cow's abdomen by a hand-held support and the angle is adjusted to obtain the best signal. The determination threshold of uterine artery blood flow velocity is greater than or equal to 30 cm / s, and the determination threshold of pulsatility index is less than or equal to 1.2. The measurement data is transmitted to the central controller through the data line. The calculation formula of the optimal insemination time window is insemination time equal to the predicted result of reproductive hormone response efficiency multiplied by 24 hours plus the weighted score value of uterine artery blood flow velocity and pulsatility index multiplied by 6 hours, wherein the weighted score value is equal to the sum of blood flow velocity divided by 40 cm / s plus 1.2 minus pulsatility index divided by 2.4, and the calculation result is pushed to the on-site technical personnel through the mobile terminal. The artificial insemination execution instruction includes the exact time point, the recommended frozen semen batch number, the operator's ID and the backup scheme identifier, which are pushed to the on-site technical personnel through the mobile terminal.

[0031] Within 72 hours after the calf is born, an ear tissue sample is collected using sterile biopsy forceps to collect a tissue sample about 3 mm in diameter from the middle of the right ear, immediately immersed in a lysis buffer containing proteinase K, the sample container is sealed, labeled with a unique number, and sent to the laboratory through a conveyor belt. Genomic DNA extraction uses magnetic bead method, the extraction purity is greater than or equal to 1.8, the concentration is greater than or equal to 50 ng / μL, and the extraction process is completed in an automatic nucleic acid extractor and the processing steps of each sample are recorded by barcode scanning. High-throughput single nucleotide polymorphism typing uses a gene chip platform, the chip probe density is 1 million per square centimeter, covering not less than 500 candidate gene sites, including the 17th nucleotide of the third exon of the growth hormone gene, the 234th nucleotide of the promoter region of the insulin-like growth factor 1 gene, the 98th nucleotide of the fifth intron of the interleukin 6 gene, and the 45th nucleotide of the third exon of the heat shock protein 70 gene. The typing results are uploaded to the database through the data interface. The typing results are associated with the cow identification, the father batch number, the environmental mean value, and the physiological peak value to form a structured gene traceability positioning database.

[0032] When the graph neural network performs relationship reasoning on the gene tracing positioning database, the database records are converted into a heterogeneous graph structure, the node types include calf nodes, cow nodes, sire nodes, gene locus nodes, and environmental factor nodes, and the edge types include parent-child relationship edges, genotype association edges, and environmental exposure edges. The graph neural network includes three layers of graph convolution layers, the range of aggregating neighbor node features in each layer is one hop to three hops, the aggregation function adopts mean aggregation, and the activation function is a rectified linear unit. The input end is connected with the database through a data interface. The final output layer adopts a soft maximum function to generate a probability distribution of key gene combinations, and the identification criterion of the interaction mode is that if the appearance frequency of a certain gene combination in the high value interval of the environmental stress factor data is greater than or equal to 80% and the corresponding calf daily weight gain is more than 2,200 grams, the gene combination is determined as an excellent gene combination. The output end is connected with the report generation system through a data interface. The positioning report includes a gene combination list, an environmental adaptation interval, an expected daily weight gain range, and a recommended mating strategy, and a paper report is output through a printing device, and an electronic version of the report is sent to the email of the relevant personnel through a network interface.

[0033] In order to better enable the relevant personnel in the technical field to fully understand and implement the present application, the specific implementation principles of the present application are further described below in conjunction with a specific application scenario.

[0034] When deploying the environmental perception sensor array in the yak grazing area, first, the temperature sensor is fixed on the support, ensuring that its height from the ground is 1.5 meters. The temperature sensor adopts a platinum resistance thermometer, which measures the temperature change of the environment where the cow is located to obtain temperature data in real time. The humidity sensor is installed side by side with the temperature sensor on the same support, with a capacitive polymer humidity sensitive element as the core, to synchronously collect environmental humidity information. The light intensity sensor is installed at the top of the support, using a silicon photodiode to receive the maximum light area, thereby accurately recording light changes. The wind speed sensor is installed at the highest point of the support, designed with a three-cup anemometer to reduce obstacle interference and ensure the accuracy of the wind speed data. The ammonia concentration sensor is installed close to the breathing height of the cow, using electrochemical principle to detect ammonia concentration with a resolution of 0.1 ppm. The noise sensor is installed in the middle of the support, maintaining a certain distance from other sensors to avoid signal interference, and collecting environmental noise decibel values through an electret condenser microphone. After synchronizing the time stamp through a 433 MHz wireless radio frequency network, all sensors converge data to the central controller to form six-dimensional environmental stress factor time series data streams.

[0035] In terms of individual physiological monitoring of cows, the miniature cortisol biosensor of the subcutaneously implanted physiological monitoring device is implanted about 0.5 cm deep under the skin behind the scapula of the cow and fixed by medical glue. The sensor uses a molecularly imprinted polymer film as a recognition layer and can quickly respond to changes in the concentration of cortisol in the cow's body. The heart rate variability collection electrode uses silver chloride silver material and is pasted on the left side of the cow's chest near the heart and under the right scapula. The heart rate variability data is obtained through a differential signal acquisition channel. The wireless radio frequency transmission unit operates at a frequency of 433 MHz, with a transmission distance of not less than 50 meters, ensuring stable data transmission. The non-contact infrared thermal imager is installed on a fixed bracket two meters away from the cow, and the angle is adjusted by a pan-tilt to cover the whole body of the cow, generating a real-time body surface heat distribution map. The rectal temperature is measured by a digital temperature probe, which is inserted 10 cm deep into the cow's rectum and fixed to the tail to prevent it from slipping out. The respiratory rate is detected by a flexible piezoelectric film sensor, and the respiratory rate per minute is calculated by a zero-crossing point algorithm. All reproductive physiological stress response indicators are aligned with environmental data by timestamp to construct a five-dimensional reproductive physiological stress response indicator set.

[0036] The core processing unit of the multi-modal deep neural network model is divided into an environmental branch and a physiological branch. The environmental branch extracts local features of environmental stress factor time series data streams through three one-dimensional convolution layers, and models temporal dependencies through two long short-term memory units. The physiological branch analyzes the spatial correlation of multi-source physiological indicators through two three-dimensional convolution layers, and predicts dynamic trends through a one-layer bidirectional long short-term memory unit. The attention weight fusion layer adopts an additive attention mechanism to perform cross-modal feature interaction on the outputs of the environmental branch and the physiological branch to generate a fusion feature vector. After the fusion feature vector is activated by two fully connected layers, the output is a single-valued prediction result of the reproductive hormone response efficiency, with a value range of zero to one.

[0037] The central controller dynamically adjusts the hormone injection parameters in the synchronization estrus treatment scheme according to the prediction result of the reproductive hormone response efficiency. The injection dose of gonadotropin-releasing hormone is adjusted in the range of 50 to 200 micrograms, the injection interval time is adjusted in the range of 72 to 120 hours, and the injection site depth is adjusted in the range of 0.5 cm subcutaneously to 1.5 cm intramuscularly. The injection dose of cloprostenol is adjusted in the range of 0.1 to 0.5 mg, the injection interval time is adjusted in the range of 48 to 96 hours, and the injection site depth is adjusted in the range of 0.3 cm subcutaneously to 1 cm intramuscularly. The adjusted parameters are generated by the central controller and issued to the handheld intelligent injector through a wireless network to ensure accurate and controllable injection process.

[0038] Within twenty-four hours before artificial insemination, the Doppler ultrasound blood flow detector measures the uterine artery blood flow velocity and pulsatility index of the cow through a seven-point-five-megahertz probe. The probe is fixed on the abdomen of the cow and the angle is adjusted through the hand-held support to obtain the best signal. The determination threshold of the uterine artery blood flow velocity is greater than or equal to thirty centimeters per second, and the determination threshold of the pulsatility index is less than or equal to one point two. The measurement data is transmitted to the central controller through the data line, combined with the prediction result of the reproductive hormone response effectiveness, and the optimal insemination time window is calculated. The calculation formula of the optimal insemination time window is that the insemination time is equal to the prediction result of the reproductive hormone response effectiveness multiplied by twenty-four hours plus the weighted score value of the uterine artery blood flow velocity and the pulsatility index multiplied by six hours. The artificial insemination execution instruction includes the precise time point, the recommended frozen semen batch number, the operator ID and the backup scheme identifier, which are pushed to the on-site technical personnel through the mobile terminal.

[0039] Within seventy-two hours after the birth of the calf, a tissue sample with a diameter of about three millimeters is collected from the middle position of the right ear using sterile biopsy forceps and immediately immersed in a lysis buffer containing proteinase K. After the sample container is sealed, it is labeled with a unique number and sent to the laboratory through a conveyor belt. The genomic deoxyribonucleic acid is extracted using a magnetic bead method, and the extraction purity is greater than or equal to one point eight, and the concentration is greater than or equal to fifty nanograms per microliter. High-throughput single nucleotide polymorphism typing uses a gene chip platform, with a chip probe density of one million per square centimeter, covering no less than five hundred candidate gene sites. The candidate gene sites include the seventeenth nucleotide of the third exon of the growth hormone gene, the two hundred thirty-fourth nucleotide of the promoter region of the insulin-like growth factor one gene, the ninety-eighth nucleotide of the fifth intron of the interleukin six gene, and the forty-fifth nucleotide of the third exon of the heat shock protein seventy gene. The typing results are uploaded to the database through a data interface and are associated with the cow identifier, the father batch number, the environmental mean value, and the physiological peak value to form a structured gene traceability positioning database.

[0040] As shown in Figure 2 The yak breeding excellent gene traceability positioning system based on deep learning includes: A data acquisition module is used to acquire a plurality of yak breeding environment data in real time to form a six-dimensional environment stress factor time series data set; based on the six-dimensional environment stress factor time series data set, a plurality of reproductive physiological stress response indexes of the yak individual are synchronously acquired to construct a five-dimensional reproductive physiological stress response index set; A breeding prediction module is used to input the six-dimensional environment stress factor time series data set and the five-dimensional reproductive physiological stress response index set into a reproductive hormone response model to generate a prediction result of the reproductive hormone response effectiveness of the yak individual; based on the prediction result, the injection parameter set of the gonadotropin-releasing hormone and the cloprostenol in the synchronization estrus treatment scheme is dynamically adjusted to form and execute individualized hormone regulation instructions; An insemination optimization module is configured to calculate an optimal insemination time window and perform artificial insemination by measuring blood flow parameters of a yak uterine artery within a preset time range after hormone regulation and in combination with the prediction result; A gene tracing module is configured to obtain a biological sample of a calf and extract genomic deoxyribonucleic acid after artificial insemination is performed, perform typing detection on preset candidate gene sites by using a high-throughput single nucleotide polymorphism typing technology, obtain a typing detection result, store the typing detection result in association with yak individual identification, paternal information, environmental stress factor data and reproductive physiological indexes, and construct a gene tracing positioning database. A graph reasoning module is configured to perform relationship reasoning on the gene tracing positioning database by using a graph neural network, identify an interaction mode of a key gene combination and an environmental stress factor, and output an excellent gene tracing positioning analysis report.

[0041] The embodiments of the present application are described above, but the embodiments are not limited to the specific implementation described above, which is only illustrative and not restrictive. Those skilled in the art can make more forms of equivalent embodiments under the inspiration of the embodiments, which are all within the protection scope of the embodiments.

Claims

1. A method for tracing and positioning excellent genes of a yak based on deep learning, characterized in that, The method comprises the following steps: Real-time acquisition of multiple yak breeding environment data and multiple reproductive physiological stress response indicators to form six-dimensional environment stress factor time series data sets and five-dimensional reproductive physiological stress response indicator sets, respectively; Input the six-dimensional environment stress factor time series data set and the five-dimensional reproductive physiological stress response indicator set into the reproductive hormone response model to generate a prediction result of the reproductive hormone response efficiency of the yak individual; Based on the prediction result, dynamically adjust the multiple injection parameters of gonadotropin-releasing hormone and cloprostenol in the same estrus treatment scheme, and execute individualized hormone regulation instructions; Within a preset time range after executing the regulation instructions, determine the yak uterine artery blood flow parameters, combine the prediction result, obtain the optimal insemination time window, and execute artificial insemination; After artificial insemination is executed, obtain a yak calf biological sample and extract genomic deoxyribonucleic acid, use high-throughput single nucleotide polymorphism typing technology to perform typing detection on preset candidate gene sites, and obtain a typing detection result; Associate and store the typing detection result with the yak individual identification, paternal information, environment stress factor data and reproductive physiological indicators to construct a gene traceability positioning database; Through graph neural network, the relationship of the gene traceability positioning database is inferred to identify the interaction mode of the key gene combination and the environmental stress factor, and an excellent gene traceability positioning analysis report is output.

2. The method of claim 1, wherein, The collection of the environmental data is achieved by deploying an environmental perception sensor array, and the sensor array includes a temperature sensor, a humidity sensor, an illumination intensity sensor, a wind speed sensor, an ammonia concentration sensor and a noise sensor; The acquisition of the reproductive physiological stress response indicators is achieved by a subcutaneous implantable physiological monitoring device and a non-contact infrared thermal imager; The reproductive hormone response model is a multi-modal deep neural network model, which adopts a double-branch architecture to process environment time series data and reproductive physiological stress response indicators, respectively, and finally outputs a prediction result of reproductive hormone response efficiency through attention mechanism to fuse double-branch features; The injection parameter set includes injection dose, injection interval time and injection site depth, and the dynamic adjustment is based on the prediction result of the reproductive hormone response efficiency within a preset parameter adjustment range.

3. The method according to claim 1, wherein, The uterine artery blood flow parameters include blood flow velocity and pulsatility index; the optimal insemination time window is obtained by weighted calculation of the prediction result of the reproductive hormone response efficiency and the blood flow parameters; The biological sample is an ear tissue sample; The candidate gene sites include gene sites related to growth, immunity and stress response.

4. The method according to claim 1, wherein, The interaction mode of the key gene combination and the environmental stress factor is identified. Based on the inference result of the graph neural network, the frequency of the gene combination under specific environmental conditions and its corresponding production performance are determined to be an excellent gene combination.

5. The method according to claim 1, wherein, By training the excellent gene positioning analysis model constituted by the graph neural network, the error between the predicted excellent result and the actual production performance is minimized to generate an excellent gene traceability positioning analysis report that matches the demand.

6. The method according to claim 1, wherein, The six-dimensional environmental stress factor time series data set includes time series change data of temperature, humidity, light intensity, wind speed, ammonia concentration, and noise decibel value; and the five-dimensional reproductive physiological stress response index set includes cortisol concentration, heart rate variability, respiratory rate, rectal temperature, and body surface heat distribution map. The reproductive hormone response model adopts a double-branch convolutional long short-term memory network architecture, wherein the environmental branch performs local feature extraction and time series dependence modeling on the time series data, the physiological branch performs spatial correlation analysis and dynamic trend prediction on the multi-source physiological indicators, and the outputs of the two branches are fused through an attention weight fusion layer for cross-modal feature interaction.

7. The method according to claim 3, wherein the method is based on deep learning. The ear tissue sample is collected using sterile biopsy forceps and immediately immersed in a lysis buffer containing protease K; The genomic deoxyribonucleic acid is extracted using a magnetic bead method; The candidate gene sites include growth hormone gene exons, insulin-like growth factor gene promoter regions, interleukin gene introns, and heat shock protein gene exons, which are related to growth rate, feed conversion efficiency, and disease resistance; The gene tracing positioning database adopts a heterogeneous graph structure for storage, and the node types include calf nodes, cow nodes, sire nodes, gene site nodes, and environmental factor nodes, and the edge types include parent-child relationship edges, genotype association edges, and environmental exposure edges.

8. The method according to claim 5, wherein, The graph neural network includes three layers of graph convolution layers, and the range threshold of aggregating neighbor node features is set for each layer. The aggregation function uses mean aggregation, and the activation function is a rectified linear unit. The final output layer uses a soft-max function to generate a probability distribution of key gene combinations. The identification criterion of the interaction mode is that if the frequency of a certain gene combination in the high value interval of the environmental stress factor data is greater than or equal to a preset threshold and the corresponding yak daily weight gain quality exceeds a preset threshold, the gene combination is determined as an excellent gene combination.

9. The method according to claim 5, wherein, The heterogeneous graph structure of the gene tracing positioning database is used as the input of the excellent gene positioning analysis model, the excellent gene positioning analysis model takes the excellent evaluation result of each gene combination as the output, takes the actual production performance as the prediction target, takes the difference between the excellent evaluation result and the prediction target as the prediction error, and takes the minimization of the sum of squares of the prediction error as the training target; the excellent gene positioning analysis model is trained until the sum of squares of the prediction error converges, and the training is stopped; The output result of the excellent gene positioning analysis model is used as the input of the report generation step, the report generation step takes the detailed analysis report of each gene combination as the output, takes the user demand matching degree as the prediction target, takes the difference between the detailed analysis report and the prediction target as the report generation error, and takes the minimization of the sum of squares of the report generation error as the training target to train until the sum of squares of the report generation error converges, and the training is stopped.

10. A system for tracing and positioning excellent genes of a yak based on deep learning, for performing the method for tracing and positioning excellent genes of a yak based on deep learning according to any one of claims 1 to 9, characterized in that, The data acquisition module is configured to acquire multiple yak breeding environment data and multiple reproductive physiological stress response indicators in real time, and form a six-dimensional environmental stress factor time series data set and a five-dimensional reproductive physiological stress response index set, respectively. The data acquisition module is configured to acquire multiple yak breeding environment data and multiple reproductive physiological stress response indicators in real time, and form a six-dimensional environmental stress factor time series data set and a five-dimensional reproductive physiological stress response index set, respectively. The cultivation prediction module is configured to input the six-dimensional environmental stress factor time series data set and the five-dimensional reproduction physiological stress response index set into a reproduction hormone response model to generate a prediction result of reproduction hormone response efficiency of the yak individual; based on the prediction result, injection parameters of gonadotropin releasing hormone and cloprostenol in the synchronization estrus treatment scheme are dynamically adjusted to form and execute individualized hormone regulation instructions; The insemination optimization module is configured to, within a preset time range after hormone regulation, calculate an optimal insemination time window and execute artificial insemination by measuring yak uterine artery blood flow parameters and combining the prediction result; The gene tracing module is configured to, after the artificial insemination is executed, obtain a biological sample of a calf and extract genomic deoxyribonucleic acid, perform typing detection on preset candidate gene sites by using a high-throughput single nucleotide polymorphism typing technology to obtain a typing detection result, and store the typing detection result in association with yak individual identification, paternal information, environmental stress factor data and reproduction physiological indexes to construct a gene tracing positioning database; The graph reasoning module is configured to perform relationship reasoning on the gene tracing positioning database by using a graph neural network, identify an interaction mode of a key gene combination and an environmental stress factor, and output an excellent gene tracing positioning analysis report.

Images