Ultralow-temperature preservation method for populus albus callus based on machine learning and molecular dynamics simulation

Abstract

The invention belongs to the technical field of machine learning, and provides a poplar callus ultralow-temperature preservation method based on machine learning and molecular dynamics simulation, and the main scheme is as follows: constructing a poplar stress-resistant phenotype prediction model based on machine learning, screening out poplar with stress-resistant phenotype based on the poplar stress-resistant phenotype prediction model, and storing the poplar callus ultralow-temperature preservation method based on machine learning and molecular dynamics simulation. The tissue is used as a populus albus callus tissue material; the method comprises the following steps: determining pre-culture medium components of a populus albus callus material based on molecular dynamics simulation, and performing pre-culture treatment on the populus albus callus material by using a determined pre-culture medium; a cryoprotectant is screened based on molecular dynamics simulation, and the screened cryoprotectant is used for conducting freezing treatment on the populus albus callus material subjected to pre-culture treatment; and determining a thawing mode based on molecular dynamics simulation, and unfreezing and re-culturing the frozen populus albus callus material by using the determined thawing mode.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method for cryopreservation of Populus alba callus based on machine learning and molecular dynamics simulation. Background Technology

[0002] Silver poplar is a deciduous tree belonging to the Salicaceae family. It has advantages such as cold resistance, deep root system, strong wind resistance, wide soil adaptability, and its wood can be used for construction and papermaking. It is very suitable for large-scale planting and helps to prevent desertification, regulate climate, and conserve water and soil. In addition, silver poplar can also create economic value.

[0003] Molecular dynamics (MD) is a computational technique based on Newtonian mechanics. It uses numerical integration to solve for the trajectories of atoms or molecules in a potential energy field, simulating the dynamic evolution of a system under specific conditions. Its core idea is to treat a molecular system as a collection of interatomic interactions, calculating the forces (potential energy gradient) acting on each atom to update its position and velocity, thus tracking time-resolved molecular behavior. Commonly used force fields include AMBER, CHARMM, and OPLS. The technical process includes: First, system construction: constructing an initial configuration (protein, lipid membrane, CPA solution, etc.) from experimental structures (such as the PDB database) or theoretical models; solvation: adding water molecules around the biomolecule (such as TIP3P, SPC / E water model). Second, energy minimization: eliminating unreasonable atomic overlaps or stresses in the initial configuration. Third, equilibrium stage: running under isothermal-barometric (NPT) or isothermal-volume (NPT) ensembles to achieve thermodynamic equilibrium. Commonly used temperature control algorithms include Berendsen temperature control and Langevin dynamics. The fourth step, the production stage: recording atomic trajectories and analyzing properties such as structure, dynamics, and free energy (timescales are typically from nanoseconds to microseconds).

[0004] Cryopreservation refers to the long-term storage of plant cells or tissues in extreme low temperatures, such as liquid nitrogen (-196°C). Under such extreme low-temperature conditions, cell differentiation, metabolism, and physiological and biochemical reactions of the preserved germplasm resources almost completely cease, while cell viability and morphogenesis potential are preserved. Currently, cryopreservation has been successfully applied to various tissues and organs of plants, including seeds, stem segments and shoot tips, embryos and somatic embryos, callus tissue, and buds. Cryopreservation has the advantages of not requiring subculturing, no variation occurring in the preserved germplasm resources, and high genetic stability of the preserved materials. It is a safe and long-term preservation method for germplasm resources such as abnormal plant seeds and asexually propagated plants, and it is also the only effective, safe, stable, economical, and long-term preservation method for biotechnological materials.

[0005] Currently, various cryopreservation methods have been established based on ultra-low temperature preservation, such as two-step cooling, pre-culturing, embedding and drying, vitrification, and embedding and vitrification. The general ultra-low temperature preservation process involves first pre-culturing and loading the material appropriately, then treating the plant material with a cryoprotectant before rapidly immersing it in liquid nitrogen. The cryoprotectant alters the dynamic behavior of water by forming a hydrogen bond network. For example, the strong hydrogen bonding between the cryoprotectant and water can inhibit ice crystal nucleation, reduce the free water content in the solution, or slow down ice crystal growth kinetics by disrupting the tetrahedral structure of water, thereby preventing mechanical damage caused by ice crystal formation and effectively protecting the plant. However, cryoprotectants often have a certain degree of toxicity and may cause cell function damage.

[0006] Cryopreservation technology can suppress the physiological metabolic intensity of materials to the maximum extent under ultra-low temperature conditions, reduce the frequency of deterioration, and thus achieve the purpose of long-term preservation of germplasm. It has been proven to be a very effective long-term preservation method. Currently, methods for cryopreservation of callus tissue already exist. For example, Chinese Patent Application No. 202010396058.8 discloses a method for cryopreservation of garlic callus tissue, which discloses the key steps of cryopreservation of garlic callus tissue, including pre-culture, cryoprotectant treatment, rewarming, and recovery. Chinese Patent No. 202010291387.6 discloses a method for cryopreservation of embryogenic callus tissue from European spruce, which discloses the specific steps of cryopreservation of embryogenic callus tissue from European spruce, including preparation of suspension, pretreatment, cryopreservation, and thawing and recovery. Chinese Patent No. 201510086065.7 discloses a cryopreservation technology for callus tissue from Saposhnikovia divaricata, which discloses the establishment of an in vitro cryopreservation technology system for Saposhnikovia divaricata using callus tissue induced from Saposhnikovia divaricata seeds as material. The specific steps include callus induction, pre-culture, loading, dehydration, freezing, washing, re-culture, and regeneration to achieve cryopreservation of Saposhnikovia divaricata callus tissue.

[0007] The aforementioned prior patent applications all involve cryopreservation of plant tissues using callus tissue as the cryopreservation material, employing methods such as programmed cooling and vitrification. However, these methods cannot be directly applied to other plant species such as *Populus tomentosa*. Due to the differences in the biological characteristics of different plant species, their preservation effectiveness requires further verification. Blindly applying cryopreservation methods for other plants to *Populus tomentosa* callus tissue will result in poor preservation outcomes, such as low survival rates and poor growth after revival. To overcome these obstacles, a more in-depth and detailed study of cryopreservation strategies for *Populus tomentosa* is needed, based on its physiological characteristics, to develop a cryopreservation method suitable for *Populus tomentosa* callus tissue.

[0008] Currently, no cryopreservation method for Populus tomentosa callus has been reported or published. Therefore, it is necessary to develop a cryopreservation method for Populus tomentosa callus. Summary of the Invention

[0009] The purpose of this invention is to provide a method for cryopreservation of Populus alba callus based on machine learning and molecular dynamics simulation.

[0010] The technical solution adopted by this invention to solve its technical problem is as follows: A method for cryopreservation of Populus tomentosa callus based on machine learning and molecular dynamics simulations includes the following steps: A machine learning-based model for predicting the stress resistance phenotype of Populus tomentosa was constructed. Based on the model, Populus tomentosa with a stress resistance phenotype was selected and used as Populus tomentosa callus material. The pre-culture medium components of Populus tomentosa callus were determined based on molecular dynamics simulations, and the determined pre-culture medium was used to pre-culture the Populus tomentosa callus. Cryoprotectants were screened based on molecular dynamics simulations, and the screened cryoprotectants were used to freeze-treat pre-cultured Populus tomentosa callus material. The thawing method was determined based on molecular dynamics simulations, and the determined thawing method was used to thaw and re-culture the frozen Populus spp. callus material.

[0011] As a further optimization, before constructing the machine learning-based Populus tomentosa stress resistance phenotype prediction model, the following steps are also included: The first batch of sterile Populus spp. seedlings was obtained. The 3rd to 8th healthy leaves below the terminal bud of the branches were cut off. After the leaf edges were removed, the seedlings were inoculated into differentiation medium. The culture temperature was 23℃±2℃ and the humidity was 60%. The seedlings were cultured in the dark. After 15 days of culture, the first subculture of Populus spp. callus was obtained. Cut 2-5cm stem segments of first-generation Populus spp. callus, retaining 2-3 lateral buds, and inoculate them into the subculture medium. The culture temperature is 23℃±2℃, the humidity is 75%, the light intensity is 6000Lux, and the photoperiod is 16h. After 20 days of culture, the next generation of sterile seedlings is obtained, and the leaves grown from them are used to culture the second generation of Populus spp. callus. The above operation was repeated to obtain the 10th generation of Populus spp. callus. The gene sequence information of all Populus spp. sterile seedlings obtained from the subculture was obtained by high-throughput sequencing.

[0012] As a further optimization, the constructed machine learning-based Populus tomentosa stress resistance phenotype prediction model uses the gene sequence information of sterile Populus tomentosa seedlings as input files. By using a machine learning-based model to predict the stress resistance phenotype of Populus tomentosa, the influence of cell variations during subculture on the phenotype of Populus tomentosa seedlings was obtained, namely the significance mapping score of the subcultured sterile Populus tomentosa seedlings. The sterile Populus tomentosa seedlings with the highest significance mapping score were selected as Populus tomentosa seedlings with stress resistance phenotype, and their leaves were cut to cultivate Populus tomentosa callus tissue.

[0013] As a further optimization, the constructed machine learning-based Populus alba stress resistance phenotype prediction model includes: an input layer, a normal noise layer, a hidden layer, and an output layer. The input layer is a C×S matrix, where C represents the number of entities and S represents the number of markers or genomic data types, used to input the gene sequence information of aseptic seedlings of Populus tomentosa. The normal noise layer is used to inject subtle normal noise with zero mean and fixed standard deviation into the gene sequence information of sterile Populus tomentosa seedlings. In fact, introducing small noises into the input data helps to understand the underlying data patterns more deeply, thereby producing accurate and stable phenotypic predictions. The hidden layer includes a two-stream convolutional layer, a batch normalization layer, and a feature compression layer; The dual-stream convolutional layer comprises an upper stream and a lower stream. The upper stream is a first one-dimensional convolutional layer with a kernel size of 16 and 4 filters. This first one-dimensional convolutional layer is used for shallow feature extraction of the gene sequence information of aseptic Populus tomentosa seedlings, reducing data dimensionality, extracting local variations in base pairs in the gene sequence information, and capturing local associations of SNP sites in the Populus tomentosa gene. The lower stream comprises a combined second and third one-dimensional convolutional layer. The second one-dimensional convolutional layer has a kernel size of 16 and 4 filters, while the third one-dimensional convolutional layer has a kernel size of 16 and 20 filters. The second one-dimensional convolutional layer is used for shallow feature extraction of the gene sequence information of aseptic Populus tomentosa seedlings. Shallow feature extraction of sequence information reduces data dimensionality and extracts local sequence patterns from the gene sequence information of sterile Populus tomentosa seedlings. Specifically, it extracts short sequence fragments that recur in the genomic DNA and have specific stress resistance functions or structural features. The third one-dimensional convolutional layer is used to perform deep feature extraction of the gene sequence information of sterile Populus tomentosa seedlings, capturing long-range dependencies in the Populus tomentosa gene sequence (long-range dependencies refer to functional, structural, or evolutionary associations between nucleotide or amino acid sites that are far apart in the sequence. Such dependencies cannot be explained by simple combinations of locally adjacent sequences, but involve interactions across large distances.), and identifying the complex effects of SNPs in the Populus tomentosa genome. The batch normalization layer is used to accelerate training and reduce internal covariate bias. The feature compression layer includes a flattening layer and a fully connected layer. The flattening layer is used to convert multidimensional features into one-dimensional vectors, and the fully connected layer is used to integrate the extracted features and output phenotypic prediction values. The output layer is C×1 + significance mapping, which means outputting the phenotypic predicted value + significance mapping score. The phenotypic predicted value is used to assess whether there is a gene related to the stress-resistant phenotypic mutation. A positive phenotypic predicted value indicates that the model predicts that the sample has a gene related to the stress-resistant phenotypic mutation, and a negative phenotypic predicted value indicates that the model predicts that the sample does not have a gene related to the stress-resistant phenotypic mutation. The significance mapping score is used to assess the contribution of genomic markers to the phenotype of interest, representing the marker's association with a single phenotype. For each sample, the significance value ranges from 0 to 1. The higher the significance value, the higher the association between the sample's genome and a single phenotype.

[0014] As a further optimization, the determination of the pre-culture medium composition for Populus tomentosa callus material based on molecular dynamics simulation includes the following steps: Prepare the basic pre-medium culture medium with the following formula: WPM 2.41 g / L + Sucrose 30 g / L + 6-BA 0.4 mg / L + NAA 0.2 mg / L; Five hypertonic solutions were added to the basal pre-medium: 0.75 mol / L mannitol, 5% sucrose, 0.75 mol / L L-sorbitol, 8% sucrose, and 0.5 mg / L abscisic acid, respectively, to form five pre-mediums. The pH of the pre-mediums was adjusted to 5.80 ± 0.20. The solutions were: (a) 0.75 mol / L mannitol added to the basal pre-medium; (b) 5% sucrose added to the basal pre-medium; (c) 0.75 mol / L L-sorbitol added to the basal pre-medium; (d) 8% sucrose added to the basal pre-medium; and (e) 0.5 mg / L abscisic acid added to the basal pre-medium. Dehydrogenases were added to five groups of pre-culture media, and room-temperature molecular dynamics simulations were performed to simulate the pre-culture process. The stability of the dehydrogenases during this process was compared, and the components of the pre-culture media used for the Populus tomentosa callus were determined based on the stability of the dehydrogenases.

[0015] As a further optimization, when the determined pre-culture medium is a pre-culture medium with 8% sucrose added to the basic pre-culture medium, the pre-culture treatment of the *Populus silvereris* callus material using the determined pre-culture medium includes: The callus tissue of Populus spp. was cultured on the pre-culture medium for 3-5 days, and then directly used as a cryopreservation material after 3-5 days of culture. Before cryopreservation, 100 mg of Populus tomentosa callus tissue was cut and the viability of the callus tissue was determined using a plant root viability assay kit. The cell viability of the Populus tomentosa callus tissue after pre-culture treatment was 398.2514 μg TTC / (g·h), and the cell viability of the callus tissue before pre-culture treatment was 400.0262 μg TTC / (g·h).

[0016] As a further optimization, the screening of cryoprotectants based on molecular dynamics simulations includes the following steps: Seven cryoprotectant formulations that may be suitable for freezing silver poplar were selected: ① (10% DMSO + 0.5 mol / L sorbitol), ② (10% DMSO + 8% hydrolyzed milk protein), ③ (10% DMSO + 10% sucrose), ④ (10% DMSO + 10% glycerol), ⑤ (10% DMSO + 8% glucose + 10% polyethylene glycol), ⑥ (10% DMSO + 10% glycerol + 5% sucrose), and ⑦ (PVS2). Room temperature molecular dynamics simulations were performed on 7 groups of cryoprotectant formulations to analyze the formation of cryoprotectant-water hydrogen bond networks after mixing the cryoprotectant with water. Cryoprotectants are screened based on the formation of hydrogen bond networks between the cryoprotectant and water molecules, as well as the number of water molecules bound by the cryoprotectant.

[0017] As a further optimization, when the selected cryoprotectant is PVS2, the step of using the selected cryoprotectant to freeze the pre-cultured Populus alba callus material refers to: 100 mg of pre-cultured Populus spp. callus was placed in a 2 ml cryovial, and PVS2 cryoprotectant was added to the ice bath. After holding for 20 min, a gradient cooling treatment was performed: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After the above treatment, the cryovial was directly immersed in liquid nitrogen at -196℃ for ultra-low temperature cryopreservation.

[0018] As a further optimization, the determination of the thawing method based on molecular dynamics simulation includes the following steps: Molecular dynamics was used to simulate the thawing process of different thawing methods. There were three different thawing methods: room temperature at 27°C, water bath at 30°C, and water bath at 40°C. Cryoprotectant and dehydrogenase were added to all three thawing methods; Comparing the stability of dehydrogenases under two thawing methods: direct immersion and gradient cooling. The direct input system was simulated by decreasing the temperature from 300K to 77.15K at a rate of 2K / ns. After the temperature dropped to 77.15K, the simulation continued for another 90ns to allow the direct input system to adapt to the temperature change from 300K to 77.15K. The total simulation time was set to 200ns. The gradient cooling process involves decreasing the temperature from 300 K to 273.15 K at a rate of 2 K / ns, holding the temperature at 273.15 K for 10 ns to allow the system to adapt to the temperature change. Then, the temperature is decreased from 273.15 K to 253.15 K at a rate of 2 K / ns. After reaching 233.15 K, the temperature is held for 20 ns to allow the system to adapt to the temperature change. Finally, the temperature is decreased from 253.15 K to 233.15 K at a rate of 2 K / ns, and held for 40 ns to allow the system to adapt to the temperature change. The system adapts to a temperature change from 253.15K to 233.15K; then, it continues to decrease from 233.15K to 193.15K at a rate of 2K / ns. After the temperature reaches 193.15K, it is held for 80ns to allow the gradient cooling system to adapt to the temperature change from 233.15K to 193.15K; then, it continues to decrease from 193.15K to 77.15K at a rate of 2K / ns. After the temperature reaches 77.15K, the simulation continues for 140ns to allow the gradient cooling system to adapt to the temperature change from 193.15K to 77.15K. This completes the entire simulation process, with the total simulation time set to 400ns. The thawing method of Populus tomentosa callus was screened based on the dehydrogenase stability of two cryopreservation methods.

[0019] As a further optimization, when the thawing method is determined to be a gradient cooling thawing method, the step of thawing and re-culturing the frozen Populus alba callus material using the determined thawing method refers to: Two weeks later, the frozen tubes stored in the liquid nitrogen biological container were removed and thawed in a 40°C water bath. After the tissues were thawed, the cryoprotectant was filtered out. The poplar callus tissues in each cryotube were washed 3-4 times with liquid induction medium. The liquid medium on the surface of the poplar callus tissues was blotted dry with filter paper, and the poplar callus tissues were placed in solid proliferation medium.

[0020] The beneficial effects of this invention are as follows: Existing cryopreservation methods for callus tissue cannot be directly applied to Populus tomentosa callus tissue. This invention uses machine learning and molecular dynamics simulation to screen key factors (callus material, pre-culture medium components, cryoprotectant, cryopreservation method, and thawing method) that affect the cryopreservation effect of Populus tomentosa callus tissue. Ultimately, the survival rate of Populus tomentosa callus tissue after cryopreservation reaches 90.16%, solving the problem of long-term cryopreservation of Populus tomentosa callus tissue and achieving long-term safe and stable preservation of Populus tomentosa callus tissue. Attached Figure Description

[0021] Figure 1This is a flowchart of the cryopreservation method for Populus tomentosa callus based on machine learning and molecular dynamics simulation in Embodiment 1 of the present invention. Figure 2 This is a diagram of the architecture of the machine learning-based Populus tomentosa stress resistance phenotype prediction model constructed in Embodiment 1 of the present invention. Figure 3 This is a comparison chart of the accuracy of prediction results for five prediction models with different convolutional kernel sizes in Embodiment 1 of the present invention; Figure 4 The steps and results for obtaining correlation parameters of the machine learning-based Populus tomentosa stress resistance phenotype prediction model constructed in Embodiment 1 of the present invention; Figure 5 (A) is a schematic diagram showing the change in the number of contacts of the five pre-culture medium components over time during the entire MD simulation process in Example 1 of the present invention. Figure 5 (B) shows the change of RMSD over time for the five pre-culture medium components in Example 1 of this invention throughout the entire MD simulation process; Figure 6 This is a snapshot of the molecular dynamics simulation in Embodiment 1 of the present invention, showing the permeation of each group of cryoprotectants in the aqueous solution. Figure 6 (A) shows the permeation of PVS2 in aqueous solution. Figure 6 (B) shows the formation of the hydrogen bond network in PVS2; Figure 7 (A) is a schematic diagram illustrating the change in the number of contacts over time during the entire MD simulation process for the two freezing methods, gradient and direct immersion in liquid nitrogen, in Embodiment 1 of the present invention. Figure 7 (B) is a schematic diagram showing the change of RMSD over time during the entire MD simulation process of the two freezing methods, gradient freezing and direct liquid nitrogen immersion, in Embodiment 1 of the present invention. Figure 8 (A) is a schematic diagram showing the change of the number of contacts over time during the entire MD simulation process for the three thawing methods in Embodiment 1 of the present invention. Figure 8 (B) is a schematic diagram illustrating the change of RMSD over time during the entire MD simulation process for the three thawing methods in Embodiment 1 of the present invention. Figure 9 This is the growth of the *Populus silvereris* callus tissue in step 1 of Example 2 after being inoculated into the proliferation medium and cultured for 2 weeks; Figure 10 The image shows the TTC staining of the Populus alba callus tissue in step 2 of Example 2. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Example 1

[0024] This embodiment provides a method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulations. The flowchart is shown below. Figure 1 The method includes the following steps: S1. Construct a machine learning-based model for predicting the stress resistance phenotype of Populus tomentosa, and select Populus tomentosa with a stress resistance phenotype based on the model, and use them as Populus tomentosa callus materials. S2. Based on molecular dynamics simulation, the pre-culture medium components of the Populus tomentosa callus material were determined, and the determined pre-culture medium was used to pre-culture the Populus tomentosa callus material. S3. Based on molecular dynamics simulation, cryoprotectants were screened, and the screened cryoprotectants were used to freeze the pre-cultured Populus spp. callus material. S4. Based on molecular dynamics simulation, determine the thawing method, and use the determined thawing method to thaw and re-culture the frozen Populus spp. callus material.

[0025] In practical applications, during the subculture of aseptic stem segments from *Populus tomentosa* seedlings to obtain the next generation of aseptic *Populus tomentosa* seedlings, cell mutations can occur, affecting the cryopreservation effect of *Populus tomentosa* callus tissue. Therefore, it is necessary to screen the callus material used. Thus, in this embodiment, a machine learning-based *Populus tomentosa* stress resistance phenotype prediction model can be constructed. Based on this model, *Populus tomentosa* seedlings with stress resistance phenotypes can be screened and used as callus material. The machine learning-based *Populus tomentosa* stress resistance phenotype prediction model can obtain the impact of cell mutations occurring during subculture on the phenotype of *Populus tomentosa* juveniles. This prediction model can learn complex nonlinear relationships from poplar genomic data, extract mutation information such as single nucleotide polymorphisms (SNPs), insertions / deletions, and copy number variations (CNVs); and through gene sequence... The process involves data acquisition and processing, gene filtering, and gene data dimensionality reduction to obtain the final input file for prediction. Based on phenotypic predicted values ​​and significance mapping scores (a set of genotype data is provided; quantitative phenotypic and marker importance are estimated by a user-selected trained model; for each sample, the significance mapping score ranges from 0 to 1, representing markers with low and high association with a single phenotype, respectively), the system identifies whether there are genes associated with stress-resistant phenotype mutations and the contribution of sample genomic markers to the development of a stress-resistant phenotype. This is used to predict whether stem height and diameter of Populus tomentosa saplings grow in a coordinated manner during their growth, i.e., whether they possess a stress-resistant phenotype. The distribution of stem height and diameter growth during sapling years reflects a tree's ability to respond to changing conditions and is crucial for ensuring the survival of natural and plantation forests. Then, based on the significance mapping values ​​of the prediction model, recommended Populus tomentosa saplings are determined, and their leaves are cut and inoculated into an induction medium to cultivate Populus tomentosa callus tissue.

[0026] It should be noted that in the selection of callus material from *Populus silvereris*, the first batch of aseptic seedlings came from the State Key Laboratory of Forest Genetics and Breeding, Chinese Academy of Forestry. These were materials obtained from the original explant stem segments after subculture induction and amplification (30-day cycle). The 3rd to 8th healthy leaves below the terminal bud of each branch were cut, and after removing the leaf edges, they were inoculated onto differentiation medium (1 / 2 MS 2.47 g / L + IBA 0.5 mg / L + Sucrose 15 g / L + 6 g / L). Agar (pH=6.00±0.20), cultured at 23℃±2℃ and 60% humidity in the dark for 15 days to obtain the first subculture of Populus tomentosa callus. Stem segments of 2-5cm were cut, retaining 2-3 lateral buds, and inoculated into the subculture medium. The culture temperature was 23℃±2℃, humidity was 75%, light intensity was 6000Lux, and photoperiod was 16h. After 20 days of culture, the next generation of sterile seedlings was obtained. The leaves grown from these seedlings were used to culture the second subculture of Populus tomentosa callus. The above operation was repeated to obtain the tenth subculture of Populus tomentosa callus. The gene sequence information of all the sterile Populus tomentosa seedlings obtained by subculture was obtained by high-throughput sequencing.

[0027] Therefore, the machine learning-based poplar stress resistance phenotype prediction model constructed in this embodiment has the following model architecture: Figure 2 As shown, using the gene sequence information of sterile Populus tomentosa seedlings as input, the constructed prediction model was used to obtain the effect of cell mutations during subculture on the phenotype of Populus tomentosa larvae. Populus tomentosa seedlings with stress-resistant phenotypes were screened. Based on the significance mapping results of the model on Populus tomentosa seedlings, recommended Populus tomentosa larvae were determined. The predicted phenotype values ​​of sterile Populus tomentosa seedlings from the 1st to 10th subcultures were: 0.30 / 0.32 / 0.43 / 0.11 / 0.03 / -0.06 / 0.06 / 0.09 / -0.11 / 0.13, respectively, and the significance mapping scores were: 0.13 / 0.15 / 0.36 / 0.08 / 0.01 / 0.01 / 0.02 / 0.03 / 0.01 / 0.03, therefore, the aseptic seedlings of Populus tomentosa from the third subculture were selected, and their leaves were cut to cultivate Populus tomentosa callus tissue.

[0028] In practical applications, high-throughput sequencing can be used to obtain genetic data (SNPs or genotypes) of sterile Populus tomentosa seedlings from different subculture cycles as input files (subcultures 1-10, 15 bottles per subculture, resulting in a total of 150 Populus tomentosa seedlings, with a sample size of 150). The prediction model is built on a deep convolutional neural network (CNN) architecture, consisting of an input layer, a normal noise layer, a hidden layer, and an output layer.

[0029] Here, a deep convolutional neural network (CNN) is used to capture the complex relationship between genotype and phenotype, solving the problem that traditional linear models (such as Bayesian regression and elastic networks) cannot handle non-additive effects. Early stopping monitors the validation set loss, terminating training early when performance no longer improves to prevent overfitting. Learning rate decay dynamically adjusts the learning rate to balance training efficiency and model convergence. Using Populus alba genotype data as input, given the numerous noise and outliers hidden in genomic data, injecting normal noise into the hidden layers during training can effectively enhance the model's resistance to overfitting and its generalization ability. Figure 2 The image demonstrates the specific impact of different noise ratios on the image, enhancing the model's robustness to data perturbations and uncertainties. Then, the hidden layer employs the Parametric ReLU (PReLU) activation function. PReLU allows gradient propagation in negative intervals, mitigating the "neuron death" problem of traditional ReLU and improving model training stability. One-dimensional convolutional layers can capture key patterns in genomic sequences through local feature extraction. Batch normalization accelerates training convergence and reduces internal covariate shifts. Flattening layers convert multi-dimensional features into one-dimensional vectors, facilitating processing by fully connected layers. The fully connected layers integrate the extracted features and output phenotypic predictions.

[0030] In practical applications, the hidden layer can include a two-stream convolutional layer, a batch normalization layer, and a feature compression layer. The two-stream convolutional layer can include an upper stream and a lower stream. The upper stream is a first-dimensional convolutional layer with a kernel size of 16 and 4 filters. This first-dimensional convolutional layer is used for shallow feature extraction, reducing data dimensionality and extracting local variations in base pairs in the *Populus alba* gene data. It is suitable for capturing local associations of *Populus alba* gene SNP sites. The lower stream uses shallow feature extraction in the early pre-training stage and then transfers to deep feature extraction in the later stage. It includes a combined second-dimensional and third-dimensional convolutional layer. The kernel size of the second-dimensional convolutional layer is 16. The model employs a dual-stream convolutional layer design: it includes a dimensionality reduction function tailored to genomic data (simplifying computational requirements without affecting prediction accuracy); and it combines shallow and deep feature extraction. The fine-tuned pattern captures the complex association between genotype and phenotype in the model, ensuring its predictive ability from genotype to phenotype. The second-dimensional convolutional layer is for shallow feature extraction, reducing data dimensionality and extracting local sequence patterns in the *Populus tomentosa* gene sequence, specifically short sequence fragments that recur in the genomic DNA and possess specific stress-resistance functions or structural features. The third-dimensional convolutional layer is for deep feature extraction, capturing long-range dependencies in the *Populus tomentosa* gene sequence and identifying complex effects (additive, dominant, epistatic effects) of SNPs in the *Populus tomentosa* genome. These effects collectively influence phenotypic variation in *Populus tomentosa*, extracting high-dimensional phenotypes, making it suitable for genome-wide interaction studies. The batch normalization layer is used to accelerate training and reduce internal covariate bias. The feature compression layer includes a flattening layer and a fully connected layer. The flattening layer is used to convert multidimensional features into one-dimensional vectors, and the fully connected layer is used to integrate the extracted features and output phenotypic prediction values. The output layer is a C×1 + significance map, i.e., outputting the phenotypic predicted value plus the significance map score. The phenotypic predicted value is used to assess whether the sample contains a gene associated with a stress-resistant phenotypic mutation. A positive phenotypic predicted value indicates that the model predicts the sample contains such a gene; the closer the phenotypic predicted value is to 1, the higher the probability that the sample contains such a gene. A negative phenotypic predicted value indicates that the model predicts the sample does not contain such a gene. The significance map score is used to assess the contribution of genomic markers to the phenotype of interest, representing the marker's association with a single phenotype. A set of genotype data is provided, and the quantitative phenotypic and marker importance is estimated by a user-selected trained model. For each sample, the significance value ranges from 0 to 1; a higher significance value indicates a higher association between the sample's genome and a single phenotype.

[0031] Specifically, for a two-stream convolutional layer, the genotype is single-click encoded, converting the genotype (e.g., AA, AT, TT) into a binary vector (e.g., [1, 0, 0], [0, 1, 0], [0, 0, 1]), with missing values ​​filled with a vector of all zeros or the mean. The lower stream consists of a one-dimensional convolutional layer (second one-dimensional convolutional layer) Conv1D(4, 16) combined with a one-dimensional convolutional layer (third one-dimensional convolutional layer) Conv1D(20, 16), while the upper stream consists of a single one-dimensional convolutional layer (first one-dimensional convolutional layer) Conv1D(4, 16). The kernel size of Conv1D(20, 16) is... A kernel size of 16 (defining the convolution window length, i.e., the number of time steps of the input sequence covered by one convolution) and a filter count of 20 (controlling the number of channels output by the convolutional layer) will generate 20 types of Populus alba genome data features; Conv1D(4, 16), with a kernel size of 16 and a filter count of 4, will generate 4 types of Populus alba genome data features.

[0032] In shallow feature extraction, fewer filters (e.g., 4) are typically used. These layers are responsible for capturing basic, low-level features (such as edges, simple waveforms, and basic trends), and the number of input channels may also be relatively small. In deep feature extraction, more filters (e.g., 20) are typically used. These layers are responsible for combining the low-level features extracted by previous layers to form more complex and abstract high-level feature representations. If previous layers have already output multiple feature maps, more filters are needed to process this combined information. It is worth noting that using larger filter values ​​increases model complexity and improves feature extraction capabilities, but may also increase the risk of overfitting. Similarly, larger kernel sizes can capture a wider range of features, but the computational cost increases, while smaller kernel values ​​focus on local features. A kernel that is too small (e.g., 3) may result in too narrow an extraction range and an inability to form complete local patterns; a kernel that is too large (e.g., 32) will introduce too much redundant information, reduce local discriminative power, and increase training instability. Through multiple sets of experiments, a kernel size of 4 achieves a good balance between local structure perception capability and feature sparsity. We conducted a systematic comparison of different kernel sizes (3, 5, 8, 16, 20). The results show that a kernel size of 16 achieves higher accuracy and stronger generalization ability in phenotypic prediction tasks. The accuracy of prediction results for different kernel sizes is as follows: Figure 3 As shown.

[0033] Specifically, Conv1D(4, 16) is a shallow feature extraction method that reduces data dimensionality and extracts local variations in base pairs from the Populus tomentosa gene data. It is suitable for capturing local associations of SNP sites in the Populus tomentosa gene. Conv1D(20, 16) is a deep feature extraction method that can identify the complex effects (additive, dominant, and epistatic effects) of SNPs in the Populus tomentosa genome. These effects collectively influence the phenotypic variation of Populus tomentosa, extracting high-latitude phenotypes. It is suitable for genome-wide interaction studies of Populus tomentosa and for multi-gene analysis of stress resistance traits. The downflow of this model uses Conv1D(4, 16) for pre-training in the early stage and then transfers to Conv1D(20, 16) for fine-tuning in the later stage. This mode can precisely capture the complex association between Populus tomentosa genotype and phenotype, ensuring the predictive ability of this gene prediction model from Populus tomentosa genotype.

[0034] Compared to four other classic models (Bayesian Ridge Regression (BRR), Elastic Net, Support Vector Regression (SVR), and DualCNN), this gene prediction model has three typical advantages in performance: it has a higher model correlation parameter (PEARSON_CORRELATION_COEFFICIENT) of 0.79. The steps and results are as follows... Figure 4 As shown, the correlation parameter for Support Vector Regression is 0.09, for Elastic Net it is 0.06, for BRR it is -0.25, and for DualCNN it is 0.03. Strong features are learned from the *Populus tomentosa* genome data, thus avoiding the tedious steps of manual feature generation. A normal noise layer is introduced to enhance predictive ability under inherent uncertainty or perturbation. Batch normalization layers and early stopping techniques are added to enhance generalization ability and predictive stability on test data.

[0035] It should be noted that the input layer of the prediction model in this embodiment is a C×S matrix, where C represents the number of entities and S represents the number of markers or genomic data types, used to input the gene sequence information of the sterile Populus tomentosa seedlings. Following the input layer is a normal noise layer, which is used to inject subtle normal noise with zero mean and a fixed standard deviation into the gene sequence information of the sterile Populus tomentosa seedlings.

[0036] For a given input, the output of the normal noise layer can be expressed as: ; here, Where N represents a normal distribution with mean 0 and variance 0. That is, for each input data point, the gene prediction model assigns a normally distributed Gaussian random sampling noise layer with a mean of 0 and a standard deviation of 0. Here, the introduction of the normal noise layer not only enhances the robustness of the model, making it less susceptible to small perturbations in the input data, but also simulates the diversity of the training data to some extent, which helps to avoid overfitting.

[0037] In this embodiment, the PReLU activation function is used to address the vanishing gradient problem and improve the stability of model training. In the *Populus alba* genome data, some genes may have a negative impact on phenotype. Compared to the traditional ReLU activation function, PReLU can better handle negative inputs, improving model stability and training efficiency. It can be expressed as: ; Here, α is a learnable parameter (usually set to 0.25 initially), which is automatically optimized during training through backpropagation.

[0038] Here, batch normalization layers are used to accelerate training and reduce internal covariate bias. Batch normalization reduces internal covariate bias, speeding up model training. By normalizing the input data, the model can converge faster and reduce the risk of overfitting. Each feature is standardized to have a mean of 0 and a variance of 1, using the following formula: , and The formulas representing the mean and variance of a feature are: , , : Number of samples in a small batch Input data (which can be the input of a fully connected layer or a feature map of a convolutional layer). It is a small constant (e.g., 10) -5 (), used to prevent the denominator from being zero.

[0039] This embodiment also uses early stopping and learning rate decay to optimize the model training process and prevent overfitting. Early stopping: Training is terminated early when the model's performance on the validation set no longer improves, saving computational resources. Learning rate decay: The learning rate is dynamically adjusted to ensure the model maintains optimal learning speed during training.

[0040] After selecting Populus tomentosa with a stress-resistant phenotype, callus tissue material can be cultured. In this embodiment, the culture process of Populus tomentosa callus tissue material is as follows: Leaves (approximately 0.5cm × 0.5cm in size) from sterile Populus tomentosa seedlings in the third subculture were cut as explants and inoculated into induction medium (WPM 2.41g / L + Sucrose 30g / L + 6-BA 0.4mg / L + NAA 0.2mg / L + Agar 6g / L + Ca(NO3)2 0.56g / L, pH adjusted to 5.80±0.20) to obtain Populus tomentosa callus tissue. Ten culture flasks were prepared, and 2-3 small cut leaves were inoculated into each flask. After inoculation, the flasks were placed in the dark for culture, and the formation of Populus tomentosa callus tissue was observed regularly (1-2 times per week). Healthy Populus sylvestris callus tissue was transferred to proliferation medium (WPM 2.41 g / L + Sucrose 30 g / L + 6-BA 0.08 mg / L + NAA 0.04 mg / L + Agar 6 g / L + Ca(NO3)2 0.56 g / L, pH adjusted to 5.80 ± 0.20) to induce differentiation and obtain a large number of healthy Populus sylvestris callus tissue.

[0041] It should be noted that, in this embodiment, as a woody plant, the callus of Populus tomentosa has the characteristics of thick cell walls and a large amount of stored substances in the cells, requiring a longer dehydration time or a specific concentration of osmotic agent. Therefore, when pre-culturing the callus of Populus tomentosa, a hypertonic solution needs to be added to the basic pre-culture medium formula. Through gradual stress adaptation, the cell's tolerance to subsequent cryoprotectants and extreme low temperatures is enhanced, which is a key step in improving the cryopreservation survival rate.

[0042] Five hypertonic solutions were added to the basal pre-medium: 0.75 mol / L mannitol, 5% sucrose, 0.75 mol / L L-sorbitol, 8% sucrose, and 0.5 mg / L abscisic acid, respectively, to form five pre-mediums. The pH of the media was adjusted to 5.80 ± 0.20, and the solutions were: (a) 0.75 mol / L mannitol added to the basal pre-medium; (b) 5% sucrose added to the basal pre-medium; (c) 0.75 mol / L L-sorbitol added to the basal pre-medium; (d) 8% sucrose added to the basal pre-medium; and (e) 0.5 mg / L abscisic acid added to the basal pre-medium. Dehydrogenase was then added to the five pre-mediums, and a room-temperature MD simulation (200 ns) was performed to simulate the pre-culture process. The stability of the dehydrogenase during this process was compared, and the composition of the pre-medium used for the *Populus silvereris* callus was determined based on the stability of the dehydrogenase.

[0043] like Figure 5As shown in (A), the number of residue contacts in the five pre-medium systems did not decrease as the simulation progressed; instead, they all showed an initial increase followed by stabilization. However, overall, the number of residue contacts in the five systems was: (d) > (a) > (c) > (b) > (e). The dehydrogenase showed higher conformational conservation in the pre-medium of (d), indicating that the dehydrogenase was more stable when 8% sucrose was added to the basal medium. Figure 5 (B) A volatility analysis was conducted, which showed the change of RMSD over time. The average RMSD of the five systems were as follows: 0.123±0.005 / 0.152±0.022 / 0.104±0.013 / 0.134±0.006 / 0.174±0.021. It can be seen that the two systems in groups (a) and (d) tended to be stable throughout the process, while the three systems in groups (b), (c) and (e) had larger fluctuations and were unstable. The RMSD of both groups (a) and (d) were relatively stable with little difference. However, the number of residue contacts in group (d) was significantly higher than that in group (a). Based on the RMSD and residue contact number results, the pre-culture composition should be group (d): WPM 2.41 g / L + Sucrose 30 g / L + 6-BA 0.4 mg / L + NAA 0.2 mg / L + 8% sucrose, and the pH of the culture medium should be adjusted to 5.80 ± 0.20.

[0044] Therefore, in this embodiment, determining the pre-culture medium composition of the *Populus silvereris* callus material based on molecular dynamics simulation may include the following steps: Prepare the basic pre-medium culture medium with the following formula: WPM 2.41 g / L + Sucrose 30 g / L + 6-BA 0.4 mg / L + NAA 0.2 mg / L; Five hypertonic solutions were added to the basal pre-medium: 0.75 mol / L mannitol, 5% sucrose, 0.75 mol / L L-sorbitol, 8% sucrose, and 0.5 mg / L abscisic acid, respectively, to form five pre-mediums. The pH of the pre-mediums was adjusted to 5.80 ± 0.20. The solutions were: (a) 0.75 mol / L mannitol added to the basal pre-medium; (b) 5% sucrose added to the basal pre-medium; (c) 0.75 mol / L L-sorbitol added to the basal pre-medium; (d) 8% sucrose added to the basal pre-medium; and (e) 0.5 mg / L abscisic acid added to the basal pre-medium. Dehydrogenases were added to five groups of pre-culture media, and room-temperature molecular dynamics simulations were performed to simulate the pre-culture process. The stability of the dehydrogenases during this process was compared, and the components of the pre-culture media used for the Populus tomentosa callus were determined based on the stability of the dehydrogenases.

[0045] Among them, the callus tissue of *Populus silvereris* is an important research object in plant cell engineering. Its metabolic activity is closely related to the presence of dehydrogenases. Dehydrogenases are key enzymes in the cellular respiratory chain (such as complexes I and II on the inner mitochondrial membrane), catalyzing the dehydrogenation reaction of substrates (such as succinate and lactate) and transferring electrons to coenzymes (NAD). + The dehydrogenase generates ATP precursors (NADH / FADH2). The stability of the dehydrogenase is a prerequisite for the dehydrogenase to perform its biological functions, and the stability of the dehydrogenase is crucial throughout the cryopreservation process.

[0046] In molecular dynamics simulations, the enzyme contact number and root mean square deviation (RMSD) are core indicators for assessing structural stability. They reveal the dynamic changes in enzyme conformation and the mechanism of function maintenance from different dimensions. Contact usually refers to the spatial proximity of two atoms or residues in an enzyme molecule within a certain distance threshold (such as 0.45 nm), reflecting non-covalent interactions such as hydrogen bonds, hydrophobic interactions, and salt bridges. A high contact value indicates an active state, while a low contact value indicates an inactive state. RMSD reflects overall drift, and small conformational fluctuations indicate that the enzyme is in a stable state.

[0047] It should be noted that when the determined pre-culture medium is a pre-culture medium containing 8% sucrose added to the basic pre-culture medium, the pre-culture treatment of the *Populus silvereris* callus material using the determined pre-culture medium may include: The callus tissue of Populus spp. was cultured on the pre-culture medium for 3-5 days, and then directly used as a cryopreservation material after 3-5 days of culture. 100 mg of *Populus sylvestris* callus tissue was collected before cryopreservation. The viability of the callus tissue was determined using a plant root viability assay kit. The cell viability of the callus tissue after pre-culture treatment was 398.2514 μg TTC / (g·h), while that of the callus tissue before pre-culture treatment was 400.0262 μg TTC / (g·h). The small difference in cell viability before and after pre-culture treatment demonstrates the rationality of the pre-culture treatment. The high cell viability of the pre-cultured callus tissue is suitable for subsequent cryopreservation experiments.

[0048] It is important to note that the molecular interactions between biomolecules, cryoprotectants, and water fundamentally determine the effectiveness of cryopreservation. Determining the molecular structures of cryoprotectants and water in the mixture is a crucial step in understanding the cryopreservation mechanism. A series of molecular dynamics (MD) simulations show that in mixtures of water and cryoprotectants, the presence of cryoprotectants significantly affects the structure of water, especially for polyhydroxy cryoprotectants (such as alcohols and sugars). With increasing solute concentration, hydrogen bonding interactions between cryoprotectants and water lead to a reduction in volumetric water and the formation of a hydration shell in the cryoprotectant. This binds water molecules around the cryoprotectant, creating a locally ordered structure, significantly reducing the free water available for ice crystal formation, and improving cryopreservation efficiency. With increasing cryoprotectant concentration, the structure and hydrogen bond network of the cryoprotectant-water mixture exhibit heterogeneity. Therefore, MD simulations can help elucidate the dynamic formation process and spatial distribution characteristics of hydrogen bond networks in cryoprotectant formulations, providing atomic-level insights for the rational design of cryoprotectants. Furthermore, the formation of hydrogen bond networks in the cryoprotectant-water mixture can provide a preliminary assessment of the cryopreservation effect of the cryoprotectant.

[0049] Traditional experimental methods (such as thermal analysis and spectroscopy) are limited by macroscopic observation scales, making it difficult to capture the dynamic details of the interaction between cryoprotectants and biomolecules / water. The dynamic evolution of hydrogen bond networks (as CPA concentration increases, the hydrogen bond network in the water-CPA mixture shifts from "water-dominated" to "CPA-dominated," with an intermediate stage exhibiting a "double osmosis" phenomenon, i.e., coexistence of water and CPA clusters) is something that traditional experiments cannot directly observe. MD simulations, however, can observe this crucial process of hydrogen bond network dynamic evolution in real time, overcoming the spatiotemporal limitations of experimental observation. In the screening of cryoprotectants, MD simulations, as a computational method based on physical principles, have demonstrated unique advantages in the rational screening of cryoprotectants (such as PVS2, DMSO, and ethylene glycol) in recent years. Compared to traditional experimental screening, high-throughput techniques (HTS), and other computational models (such as QSAR and machine learning), MD significantly improves the efficiency and scientific rigor of protective agent development through the precise capture of atomic-level dynamic behavior. This includes: accurate visualization of molecular interaction mechanisms; dynamic tracking of phase transitions and glass transitions; extreme condition simulation capabilities; and multi-scale and multi-physics coupled analysis. Furthermore, while traditional experimental trial-and-error methods require substantial experimental resources, MD simulations can screen potential candidates, narrowing the experimental scope. MD can rapidly evaluate the protective potential of thousands of candidate molecules through "computational pre-screening," reducing the blind spots of wet experiments.

[0050] In summary, the MD simulation method can visualize the cryopreservation effect of cryoprotectants and predict novel formulations and toxicity neutralizers. These advantages make MD simulation an indispensable tool in the research and development of cryoprotectants, especially in solving bottleneck problems such as vitrification toxicity and ice crystal inhibition mechanisms.

[0051] Therefore, in this embodiment, the screening of cryoprotectants based on molecular dynamics simulations may include the following steps: Seven cryoprotectant formulations that may be suitable for freezing silver poplar were selected: ① (10% DMSO + 0.5 mol / L sorbitol), ② (10% DMSO + 8% hydrolyzed milk protein), ③ (10% DMSO + 10% sucrose), ④ (10% DMSO + 10% glycerol), ⑤ (10% DMSO + 8% glucose + 10% polyethylene glycol), ⑥ (10% DMSO + 10% glycerol + 5% sucrose), and ⑦ (PVS2). Seven cryoprotectant formulations were subjected to room temperature molecular dynamics simulations (100 ns) to analyze the formation of cryoprotectant-water hydrogen bond networks after mixing the cryoprotectant with water. Cryoprotectants are screened based on the formation of hydrogen bond networks between the cryoprotectant and water molecules, as well as the number of water molecules bound by the cryoprotectant.

[0052] The better the hydrogen bond network between the cryoprotectant and water molecules, the more water molecules the cryoprotectant binds (the higher the proportion of bound water), and the better the cryoprotectant's freezing effect.

[0053] In this embodiment, all cryoprotectant formulations were subjected to room temperature (300K, with the Berendsen hot bath used to rapidly reach 300K during the NVT equilibration phase) and MD simulation (100ns) to observe the formation kinetics of the cryoprotectant-water hydrogen bond network after mixing the cryoprotectant with water. Based on the formation of the hydrogen bond network between the cryoprotectant and water molecules and the number of water molecules bound by the cryoprotectant, the ability of the cryoprotectant to inhibit ice crystal formation during the subsequent glass transition process can be determined.

[0054] Specifically, the TIP3P model (accurately reproducing the hydrogen bond network and diffusion characteristics of liquid water) and the OPLS-AA force field (the OPLS-AA force field is applicable to the charge distribution and torsional potential energy of organic molecules such as DMSO and sugar alcohols) are used. The TIP3P model is a three-point rigid water model developed by Jorgensen et al. in 1983, designed to balance computational efficiency and accuracy, and is mainly used for solvation (adding water molecules around biomolecules) simulations. The TIP3P model has a rigid triangular geometry, an OH bond length of 0.09572 nm, and a HOH bond angle of 104.52° (consistent with experimental values). The total charge of the TIP3P model is neutrally distributed: oxygen atom charge -0.834e, each hydrogen atom charge +0.417e. The optimized charge distribution of TIP3P enables it to accurately simulate the hydrogen bond strength between water molecules and polar protein groups (such as arginine guanidino groups) (error <5%). TIP3P can accurately reproduce the hydrogen bond network and diffusion properties of liquid water, and with its computational efficiency and biomolecular compatibility, it has become the "benchmark water model" in the field of biological simulation. The proportion of bound water to free water in the mixture of cryoprotectant and water molecules after 200 ns was statistically analyzed using the GROMACS hbond module, and the formation of the hydrogen bond network was observed through time-dependent kinetic snapshots.

[0055] Based on the cryoprotectant formulations of groups ① to ⑦, the required box size for the simulation system is 1000 nm. 3 In a closed environment, according to the formula: number of molecules = concentration (mol / L) × volume (L) × Avogadro's constant (NA = 6.022 × 10²³ mol / L), the molecular number is calculated as follows: −1 The required number of molecules for each reagent in groups ① to ⑦ was calculated. For example, group ① required 771 DMSO molecules and 301 sorbitol molecules, with the remainder being water molecules. After creating the boxes, the reagents and water molecules used in the cryoprotectant formulation were added sequentially. Molecular stacking was achieved using GROMACS' built-in tools to avoid spatial overlap. Energy optimization and environmental optimization of the mixture system employed a dual optimization approach: the steepest descent method and the conjugate gradient method. By leveraging the advantages of both methods at different stages, optimization efficiency and robustness were significantly improved. Kinetic simulations were then performed to obtain time-dependent kinetic snapshots, observing the formation of hydrogen bond networks between the cryoprotectant and water molecules. The ratio of bound water to all water molecules was calculated, with the hydrogen bond criterion referencing the official GROMACS standard. Specifically, the geometric definition was donor-acceptor distance < 0.35 nm and bond angle > 150°. The cryopreservation effect of the cryoprotectant was preliminarily assessed based on the formation of hydrogen bond networks in the cryoprotectant-water mixtures in each group.

[0056] like Figure 6 (A) and Figure 6As shown in (B), compared to other components, the cryoprotectants in groups ①, ③, and ⑦ exhibit better penetration in aqueous solutions and better aggregation. Notably, in group ⑦, the cryoprotectant aggregates are the largest and contain water molecules. These aggregates form a double-layered hydrogen bond network with water molecules, binding them around the cryoprotectant and creating a locally ordered structure. This significantly reduces the free water available for ice crystal formation, improving cryopreservation. In contrast, in groups ① and ③, the cryoprotectant aggregates do not contain water; they only form an outer hydrogen bond network with water molecules. In groups ④, ⑤, and ⑥, the cryoprotectants are relatively dispersed in aqueous solutions; the resulting cryoprotectant aggregates are smaller and also only form an outer hydrogen bond network with water molecules. However, in group ②, the cryoprotectant penetration is poor, with a large amount failing to penetrate into water molecules, and the aggregates formed after penetration are the smallest. The bound water percentages of the cryoprotectant formulations in groups ① through ⑦ are 15.3%, 6.3%, 15.8%, 10.2%, 9.1%, 10.8%, and 21.3%, respectively. Therefore, it is speculated that group ⑦ has the best cryopreservation effect, groups ① and ③ have good cryopreservation effects (with groups ① and ③ having similar effects), groups ④, ⑤, and ⑥ having the next best effects (group ⑤ has the lowest bound water percentage, suggesting the worst cryopreservation effect, and groups ④ and ⑥ have similar effects), and group ② has the worst cryopreservation effect. Therefore, group ⑦ (PVS2) is selected as the cryoprotectant for *Populus alba* callus.

[0057] When the selected cryoprotectant is PVS2, the step of using the selected cryoprotectant to freeze the pre-cultured Populus spp. callus material refers to: 100 mg of pre-cultured Populus spp. callus was placed in a 2 ml cryovial, and PVS2 cryoprotectant was added to the ice bath. After holding for 20 min, a gradient cooling treatment was performed: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After the above treatment, the cryovial was directly immersed in liquid nitrogen at -196℃ for ultra-low temperature cryopreservation.

[0058] It should be noted that after adding cryoprotectant and dehydrogenase, the stability of dehydrogenase was compared between the two freezing methods of direct addition and gradient cooling. The direct addition freezing method involved cooling from room temperature (27℃) to 300K at a rate of 2K / ns to 77.15K (-196℃). After the temperature dropped to 77.15K, the simulation continued for a period of time to allow the direct addition system to adapt to the temperature change from 300K to 77.15K. The total duration of the entire MD simulation process was 200ns. The gradient cooling method involves decreasing the temperature from 300 K to 273.15 K (0℃) at a rate of 2 K / ns, holding the temperature at 273.15 K for 10 ns to allow the system to adapt to the temperature change from 300 K to 273.15 K; then decreasing the temperature from 273.15 K to 253.15 K (-20℃) at a rate of 2 K / ns, holding the temperature at 20 ns to allow the system to adapt to the temperature change from 273.15 K to 253.15 K; then decreasing the temperature from 253.15 K to 233.15 K (-40℃) at a rate of 2 K / ns, holding the temperature at 233.15 K for 40 ns to allow the system to adapt to the temperature change from 0℃ to 273.15 K. The temperature was varied from 253.15 K to 233.15 K; then, it was continued to decrease from 233.15 K to 193.15 K (-80℃) at a rate of 2 K / ns. After the temperature dropped to 193.15 K, it was held for 80 ns to allow the gradient cooling system to adapt to the temperature change from 233.15 K to 193.15 K; then, it was decreased from 193.15 K to 77.15 K (-196℃) at a rate of 2 K / ns. After the temperature dropped to 77.15 K, the simulation continued for a period of time to allow the gradient cooling system to adapt to the temperature change from 193.15 K to 77.15 K. The total duration of the entire MD simulation process was 400 ns. The dehydrogenase stability of the two cryopreservation methods was used to screen the cryopreservation method for Populus tomentosa callus.

[0059] like Figure 7 As shown in (A), the number of residue contacts in both systems under the two cryopreservation methods of direct immersion in liquid nitrogen and gradient cooling did not show a decreasing trend as the simulation progressed. Instead, they both exhibited a state of first increasing and then stabilizing. The number of residue contacts in the gradient cooling system was significantly higher than that in the direct immersion in liquid nitrogen system. Furthermore, in the gradient cooling system, a clear segmented increasing trend in the number of residue contacts can be observed, corresponding to the cooling process from -80℃ to -196℃. This indicates that the dehydrogenase can better adapt to the effects of temperature changes when using the gradient cooling method. The dehydrogenase exhibits higher conformational conservation during the gradient cooling process, and the dehydrogenase is more stable. Figure 7(B) A volatility analysis was performed, showing the change of RMSD over time. The average RMSD values ​​for the two systems were 0.061±0.012 and 0.058±0.012, respectively. It can be seen that the RMSD of both systems decreased after the temperature decreased, but quickly stabilized. Although the RMSD of both cryopreservation methods showed a rapid initial decrease followed by stabilization, the contact value of the system using the gradient cooling cryopreservation method was significantly higher than that of the system directly immersed in liquid nitrogen. Based on the RMSD and residue contact number results, the cryopreservation method should employ gradient cooling: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After this treatment, the cryopreservation tubes should be directly immersed in liquid nitrogen (-196℃) for cryopreservation.

[0060] It should be noted that during the MD simulation of thawing, different thawing methods (27℃ room temperature, 30℃ water bath, and 40℃ water bath) were used. For all three thawing methods, the cryoprotectant and dehydrogenase were added, and the initial temperature was 77.15K. The temperature was increased to 300K, 303K, and 313K at a rate of 2 K / ns, with a total simulation time of 200ns. The thawing method of Populus tomentosa callus was screened based on the stability of the dehydrogenase among the three thawing methods.

[0061] like Figure 8 As shown in (A), the number of residue contacts in the three thawing systems did not decrease as the simulation progressed; instead, they all showed an initial increase followed by stabilization. However, overall, the residue contact value of the 40°C water bath thawing system was the highest, and significantly higher than that of the 27°C room temperature and 30°C water bath systems. This indicates that the dehydrogenase is more stable when thawing in a 40°C water bath. Figure 8 (B) A volatility analysis was performed, showing the changes in RMSD over time. The average RMSD values ​​for the three systems (27℃ room temperature, 30℃ water bath, and 40℃ water bath) were 0.172±0.025, 0.161±0.022, and 0.153±0.017, respectively. It can be seen that all three systems exhibited an initial increase followed by stabilization, stabilizing around 40 ns. The system thawing at 40℃ showed the smallest volatility and was the most stable. In contrast, the other two systems exhibited larger volatility. Based on the RMSD and residue contact number results, a gradient cooling method was adopted for thawing.

[0062] Therefore, in this embodiment, determining the thawing method based on molecular dynamics simulation may include the following steps: Molecular dynamics was used to simulate the thawing process of different thawing methods. There were three different thawing methods: room temperature at 27°C, water bath at 30°C, and water bath at 40°C. Cryoprotectant and dehydrogenase were added to all three thawing methods; Compare the dehydrogenase stability of three thawing methods; All three thawing methods involved adding cryoprotectant and dehydrogenase, starting at 77.15K, and increasing to 300K, 303K, and 313K at a rate of 2 K / ns, with a total simulation time of 200ns. The thawing method of Populus tomentosa callus was screened based on the stability of dehydrogenase among three thawing methods.

[0063] It should be noted that when the thawing method is determined to be a 40℃ water bath thawing method, the phrase "thawing and reculturing the frozen Populus spp. callus material using the determined thawing method" refers to: Two weeks later, the frozen tubes stored in the liquid nitrogen biological container were removed and thawed in a 40°C water bath. After the tissues were thawed, the cryoprotectant was filtered out. The poplar callus tissues in each cryotube were washed 3-4 times with liquid induction medium. The liquid medium on the surface of the poplar callus tissues was blotted dry with filter paper, and the poplar callus tissues were placed in solid proliferation medium.

[0064] Example 2 Based on Example 1, this example verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed according to the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: See Figure 9 Leaves (approximately 0.5cm × 0.5cm) from sterile Populus tomentosa seedlings in their third subculture were used as explants. These explants were inoculated into an induction medium (WPM 2.41g / L + Sucrose 30g / L + 6-BA 0.4mg / L + NAA 0.2mg / L + Agar 6g / L + Ca(NO3)2 0.56g / L, pH adjusted to 5.80±0.20) to obtain Populus tomentosa callus. Ten culture flasks were prepared, with 2-3 excised leaflets inoculated into each flask. After inoculation, the flasks were placed in the dark for culture, and the formation of Populus tomentosa callus was observed regularly (1-2 times per week). Healthy Populus sylvestris callus tissue was transferred to proliferation medium (WPM 2.41 g / L + Sucrose 30 g / L + 6-BA 0.08 mg / L + NAA 0.04 mg / L + Agar 6 g / L + Ca(NO3)2 0.56 g / L, pH adjusted to 5.80 ± 0.20) to induce differentiation and obtain a large number of healthy Populus sylvestris callus tissue.

[0065] 2. Pre-culture treatment: See Figure 10The purpose of pre-culture is to improve the frost resistance of Populus tomentosa callus and reduce or avoid frost damage. Common pre-culture methods include hypertonic treatment and cold acclimatization. In this example, a basal pre-culture medium (WPM 2.41 g / L + Sucrose 30 g / L + 6-BA 0.4 mg / L + NAA 0.2 mg / L) with 8% sucrose added and the pH of the medium adjusted to 5.80 ± 0.20 was used. Populus tomentosa callus was cultured on WPM basal medium for 3-5 days to reduce free water in the callus cells and increase soluble sugars. After culturing for 3-5 days, the callus tissue was directly used as cryopreservation material. Before freezing, 100 mg of Populus tomentosa callus tissue was cut off, and the callus viability was determined using a plant root viability assay kit (TTC method). The cell viability of the Populus tomentosa callus tissue after pre-culture was 398.2514 μg TTC / (g·h), while that before pre-culture was 400.0262 μg TTC / (g·h). The difference in cell viability before and after pre-culture was small, proving the rationality of the pre-culture treatment. The high cell viability of the Populus tomentosa callus tissue after pre-culture makes it suitable for subsequent cryopreservation experiments.

[0066] 3. Freezing treatment: 100 mg of pre-cultured Populus spp. callus was placed in a 2 ml cryovial. Group 7 (PVS2) cryoprotectant was added to the ice bath and kept for 20 min. Then, a gradient cooling treatment was performed: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After the above treatment, the cryovial was directly immersed in liquid nitrogen (-196℃) for ultra-low temperature cryopreservation.

[0067] 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: Two weeks later, the cryopreservation tubes stored in the liquid nitrogen biological container were removed and thawed in a 40°C water bath. After the tissues thawed, the cryoprotectant was filtered off, and the *Populus silvereris* callus tissue in each cryopreservation tube was washed 3-4 times with liquid induction medium. The liquid medium on the surface of the *Populus silvereris* callus tissue was blotted dry with filter paper, and then the *Populus silvereris* callus tissue was placed in solid proliferation medium. Under these conditions, the cryopreserved *Populus silvereris* callus tissue was cultured for 30 days after thawing, and the recovery rate of the *Populus silvereris* callus tissue in Example 1 reached 90.16%.

[0068] Example 3 This embodiment verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed based on the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: Same as in Example 2; 2. Pre-culture treatment: Same as in Example 2; 3. Freezing treatment: 100 mg of pre-cultured Populus spp. callus tissue was placed in a 2 ml cryovial. Group ① (10% DMSO + 0.5 mol / L sorbitol) cryoprotectant was added to the ice bath and kept for 20 min. Then, a gradient cooling treatment was performed: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After the above treatment, the cryovial was directly immersed in liquid nitrogen (-196℃) for ultra-low temperature cryopreservation.

[0069] 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: The method was the same as in Example 2; the recovery rate of the callus tissue from *Populus silvereris* was 65.61% after thawing and culturing for 30 days.

[0070] Example 4 This embodiment verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed based on the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: Same as in Example 2; 2. Pre-culture treatment: Same as in Example 2; 3. Freezing treatment: 100 mg of pre-cultured Populus spp. callus tissue was placed in a 2 ml cryovial. Group ③ (10% DMSO + 10% sucrose) cryoprotectant was added to the ice bath and kept for 20 min. Then, a gradient cooling treatment was performed: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After the above treatment, the cryovial was directly immersed in liquid nitrogen (-196℃) for ultra-low temperature cryopreservation.

[0071] 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: The method was the same as in Example 2; the recovery rate of the callus tissue from *Populus silvereris* was 67.90% after thawing and culturing for 30 days.

[0072] Example 5 This embodiment verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed based on the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: Same as in Example 2; 2. Pre-culture treatment: Same as in Example 2; 3. Freezing treatment: 100 mg of pre-cultured *Populus silvereris* callus was placed in a 2 ml cryovial. The cryoprotectant (group ⑤, 10% DMSO + 8% glucose + 10% polyethylene glycol) was added to an ice bath and kept for 20 min. Then, a gradient cooling treatment was applied: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After this treatment, the cryovial was directly immersed in liquid nitrogen (-196℃) for ultra-low temperature cryopreservation.

[0073] 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: The method was the same as in Example 2; the recovery rate of the Populus spp. callus tissue was 45.74% after thawing and culturing for 30 days.

[0074] Example 6 This embodiment verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed based on the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: Same as in Example 2; 2. Pre-culture treatment: Same as in Example 2; 3. Freezing treatment: 100 mg of pre-cultured Populus spp. callus tissue was placed in a 2 ml cryovial. Group ② (10% DMSO + 8% hydrolyzed milk protein) cryoprotectant was added to the ice bath and kept for 20 min. Then, a gradient cooling treatment was performed: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After the above treatment, the cryovial was directly immersed in liquid nitrogen (-196℃) for ultra-low temperature freezing.

[0075] 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: The method was the same as in Example 2; the recovery rate of the callus tissue from *Populus silvereris* was 37.88% after thawing and culturing for 30 days.

[0076] Example 7 It should be noted that, based on the MD simulation results, the predicted cryoprotectant effects are as follows: Group ⑦ has the best cryopreservation effect, Groups ① and ③ have good cryopreservation effects, Groups ④, ⑤, and ⑥ have the next best cryopreservation effects, and Group ② has the worst cryopreservation effect. Examples 2, 3, 4, 5, and 6 above have already verified the predicted cryopreservation effects of the 7 cryoprotectants across the four gradients. Furthermore, the specific cryopreservation effect of Group ⑥, which belongs to the third gradient and is predicted to have a better cryopreservation effect (higher bound water content, specifically 10.8%), was further selectively verified, while Group ④, which has a lower bound water content (10.2%), was not verified.

[0077] This embodiment verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed based on the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: Same as in Example 2; 2. Pre-culture treatment: Same as in Example 2; 3. Freezing treatment: 100 mg of pre-cultured Populus spp. callus was placed in a 2 ml cryovial. Group 6 (10% DMSO + 10% glycerol + 5% sucrose) cryoprotectant was added to the ice bath. The cryoprotectant was added and then the tissue was directly immersed in liquid nitrogen (-196℃) for cryopreservation at ultra-low temperature.

[0078] 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: The method was the same as in Example 2; the recovery rate of the callus tissue from *Populus silvereris* was 51.59% after 30 days of thawing and culture.

[0079] Example 8 This embodiment verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed based on the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: Same as in Example 2; 2. Pre-culture treatment: Same as in Example 2; 3. Freezing treatment: Same as in Example 2; 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: The method was the same as in Example 2; two weeks later, the cryotubes stored in the liquid nitrogen biological container were taken out and thawed in a water bath at room temperature (27°C). The recovery rate of the poplar callus tissue was 64.41% after 30 days of culture following thawing.

[0080] Example 9 This embodiment verifies the specific cryopreservation effect of the cryopreservation scheme for Populus tomentosa callus designed based on the above screening results, including the following steps: 1. Culture of callus material from *Populus silvereris*: Same as in Example 2; 2. Pre-culture treatment: Same as in Example 2; 3. Freezing treatment: Same as in Example 2; 4. Thawing and reculturing of cryopreserved *Populus silvereris* callus: The method was the same as in Example 2; two weeks later, the cryotubes stored in the liquid nitrogen biological container were taken out and thawed in a 30°C water bath. The recovery rate of the Populus alba callus tissue was 68.25% after 30 days of thawing and culture.

[0081] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation, characterized in that... Includes the following steps: A machine learning-based model for predicting the stress resistance phenotype of Populus tomentosa was constructed. Based on the model, Populus tomentosa with a stress resistance phenotype was selected and used as Populus tomentosa callus material. The pre-culture medium components of Populus tomentosa callus were determined based on molecular dynamics simulations, and the determined pre-culture medium was used to pre-culture the Populus tomentosa callus. Cryoprotectants were screened based on molecular dynamics simulations, and the screened cryoprotectants were used to freeze-treat pre-cultured Populus tomentosa callus material. The thawing method was determined based on molecular dynamics simulations, and the determined thawing method was used to thaw and re-culture the frozen Populus spp. callus material.

2. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 1, characterized in that, Before constructing the machine learning-based model for predicting the stress resistance phenotype of Populus tomentosa, the following steps are also included: The first batch of sterile Populus spp. seedlings was obtained. The 3rd to 8th healthy leaves below the terminal bud of the branches were cut off. After the leaf edges were removed, the seedlings were inoculated into differentiation medium. The culture temperature was 23℃±2℃ and the humidity was 60%. The seedlings were cultured in the dark. After 15 days of culture, the first subculture of Populus spp. callus was obtained. Cut 2-5cm stem segments of first-generation Populus spp. callus, retaining 2-3 lateral buds, and inoculate them into the subculture medium. The culture temperature is 23℃±2℃, the humidity is 75%, the light intensity is 6000Lux, and the photoperiod is 16h. After 20 days of culture, the next generation of sterile seedlings is obtained, and the leaves grown from them are used to culture the second generation of Populus spp. callus. The above operation was repeated to obtain the 10th generation of Populus spp. callus. The gene sequence information of all Populus spp. sterile seedlings obtained from the subculture was obtained by high-throughput sequencing.

3. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 2, characterized in that, The constructed machine learning-based Populus tomentosa stress resistance phenotype prediction model uses the gene sequence information of sterile Populus tomentosa seedlings as input files. By using a machine learning-based model to predict the stress resistance phenotype of Populus tomentosa, the influence of cell variations during subculture on the phenotype of Populus tomentosa seedlings was obtained, namely the significance mapping score of the subcultured sterile Populus tomentosa seedlings. The sterile Populus tomentosa seedlings with the highest significance mapping score were selected as Populus tomentosa seedlings with stress resistance phenotype, and their leaves were cut to cultivate Populus tomentosa callus tissue.

4. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 1, characterized in that, The constructed machine learning-based Populus tomentosa stress resistance phenotype prediction model includes: an input layer, a normal noise layer, a hidden layer, and an output layer; The input layer is a C×S matrix, where C represents the number of entities and S represents the number of markers or genomic data types, used to input the gene sequence information of aseptic seedlings of Populus tomentosa. The normal noise layer is used to inject subtle normal noise with zero mean and fixed standard deviation into the gene sequence information of sterile Populus tomentosa seedlings; The hidden layer includes a two-stream convolutional layer, a batch normalization layer, and a feature compression layer; The dual-stream convolutional layer comprises an upper stream and a lower stream. The upper stream is a first one-dimensional convolutional layer with a kernel size of 16 and 4 filters. This first one-dimensional convolutional layer is used for shallow feature extraction of the gene sequence information of sterile Populus tomentosa seedlings, reducing data dimensionality, extracting local variations in base pairs in the gene sequence information, and capturing local associations of SNP sites in the Populus tomentosa gene. The lower stream comprises a second one-dimensional convolutional layer and a third one-dimensional convolutional layer used in combination. The second one-dimensional convolutional layer has a kernel size of 16 and 4 filters, and the third one-dimensional convolutional layer has a kernel size of 16 and 4 filters, respectively. The kernel size of the first-dimensional convolutional layer is 16, and the number of filters is 20. The second one-dimensional convolutional layer is used to perform shallow feature extraction on the gene sequence information of the sterile seedlings of Populus tomentosa, reduce the data dimensionality, and extract the local sequence patterns in the gene sequence information of the sterile seedlings of Populus tomentosa, specifically the short sequence fragments that recur in the genomic DNA and have specific stress resistance functions or structural features. The third one-dimensional convolutional layer is used to perform deep feature extraction on the gene sequence information of the sterile seedlings of Populus tomentosa, capture the long-range dependencies in the gene sequence of Populus tomentosa, and identify the complex effects of SNPs in the Populus tomentosa genome. The batch normalization layer is used to accelerate training and reduce internal covariate bias. The feature compression layer includes a flattening layer and a fully connected layer. The flattening layer is used to convert multidimensional features into one-dimensional vectors, and the fully connected layer is used to integrate the extracted features and output phenotypic prediction values. The output layer is C×1 + significance mapping, which means outputting the phenotypic predicted value + significance mapping score. The phenotypic predicted value is used to assess whether there is a gene related to the stress-resistant phenotypic mutation. A positive phenotypic predicted value indicates that the model predicts that the sample has a gene related to the stress-resistant phenotypic mutation, and a negative phenotypic predicted value indicates that the model predicts that the sample does not have a gene related to the stress-resistant phenotypic mutation. The significance mapping score is used to assess the contribution of genomic markers to the phenotype of interest, representing the marker's association with a single phenotype. For each sample, the significance value ranges from 0 to 1. The higher the significance value, the higher the association between the sample's genome and a single phenotype.

5. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 1, characterized in that, The determination of the pre-culture medium composition for Populus tomentosa callus material based on molecular dynamics simulation includes the following steps: Prepare the basic pre-culture medium with the following formula: WPM 2.41 g / L + Sucrose 30 g / L + 6-BA 0.4 mg / L + NAA 0.2 mg / L; Five groups of hypertonic solutions were added to the basal pre-culture medium: Five pre-mediums were prepared by adding 0.75 mol / L mannitol, 5% sucrose, 0.75 mol / L L-sorbitol, 8% sucrose, and 0.5 mg / L abscisic acid. The pH of the pre-mediums was adjusted to 5.80 ± 0.

20. The pre-mediums were: (a) 0.75 mol / L mannitol added to the pre-medium; (b) 5% sucrose added to the pre-medium; (c) 0.75 mol / L L-sorbitol added to the pre-medium; (d) 8% sucrose added to the pre-medium; and (e) 0.5 mg / L abscisic acid added to the pre-medium. Dehydrogenases were added to five groups of pre-culture media, and room-temperature molecular dynamics simulations were performed to simulate the pre-culture process. The stability of the dehydrogenases during this process was compared, and the components of the pre-culture media used for the Populus tomentosa callus were determined based on the stability of the dehydrogenases.

6. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 5, characterized in that, When the determined pre-culture medium is a pre-culture medium containing 8% sucrose added to a basic pre-culture medium, the pre-culture treatment of *Populus silvereris* callus material using the determined pre-culture medium includes: The callus tissue of Populus spp. was cultured on the pre-culture medium for 3-5 days, and then directly used as a cryopreservation material after 3-5 days of culture. Before cryopreservation, 100 mg of Populus tomentosa callus tissue was cut and the viability of the callus tissue was determined using a plant root viability assay kit. The cell viability of the Populus tomentosa callus tissue after pre-culture treatment was 398.2514 μg TTC / (g·h), and the cell viability of the callus tissue before pre-culture treatment was 400.0262 μg TTC / (g·h).

7. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 1, characterized in that, The screening of cryoprotectants based on molecular dynamics simulations includes the following steps: Seven cryoprotectant formulations that may be suitable for freezing silver poplar were selected: Group ① (10% DMSO + 0.5 mol / L sorbitol), Group ② (10% DMSO + 8% hydrolyzed milk protein), Group ③ (10% DMSO + 10% sucrose), Group ④ (10% DMSO + 10% glycerol), Group ⑤ (10% DMSO + 8% glucose + 10% polyethylene glycol), Group ⑥ (10% DMSO + 10% glycerol + 5% sucrose), and Group ⑦ (PVS2). Room temperature molecular dynamics simulations were performed on 7 groups of cryoprotectant formulations to analyze the formation of cryoprotectant-water hydrogen bond networks after mixing the cryoprotectant with water. Cryoprotectants are screened based on the formation of hydrogen bond networks between the cryoprotectant and water molecules, as well as the number of water molecules bound by the cryoprotectant.

8. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 7, characterized in that, When the selected cryoprotectant is PVS2, the step of using the selected cryoprotectant to freeze the pre-cultured Populus spp. callus material refers to: 100 mg of pre-cultured Populus spp. callus was placed in a 2 ml cryovial, and PVS2 cryoprotectant was added to the ice bath. After holding for 20 min, a gradient cooling treatment was performed: -20℃ for 20 min, -40℃ for 40 min, and -80℃ for 60 min. After the above treatment, the cryovial was directly immersed in liquid nitrogen at -196℃ for ultra-low temperature cryopreservation.

9. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 1, characterized in that, The determination of the thawing method based on molecular dynamics simulation includes the following steps: Molecular dynamics was used to simulate the thawing process of different thawing methods. There were three different thawing methods: room temperature at 27°C, water bath at 30°C, and water bath at 40°C. Cryoprotectant and dehydrogenase were added to all three thawing methods; Comparing the stability of dehydrogenases under two thawing methods: direct immersion and gradient cooling. The direct input system was simulated by decreasing the temperature from 300K to 77.15K at a rate of 2K / ns. After the temperature dropped to 77.15K, the simulation continued for another 90ns to allow the direct input system to adapt to the temperature change from 300K to 77.15K. The total simulation time was set to 200ns. The gradient cooling process involves decreasing the temperature from 300 K to 273.15 K at a rate of 2 K / ns, holding the temperature at 273.15 K for 10 ns to allow the system to adapt to the temperature change. Next, the temperature is decreased from 273.15 K to 253.15 K at a rate of 2 K / ns, held for 20 ns to allow the system to adapt to the temperature change. Finally, the temperature is decreased from 253.15 K to 233.15 K at a rate of 2 K / ns, held for 40 ns to allow the system to adapt to the temperature change. The temperature change should be from 253.15K to 233.15K; then, continue from 233.15K to 193.15K at a rate of 2K / ns. After the temperature reaches 193.15K, hold for 80ns to allow the gradient cooling system to adapt to the temperature change from 233.15K to 193.15K; then, continue from 193.15K to 77.15K at a rate of 2K / ns. After the temperature reaches 77.15K, continue simulating for 140ns to allow the gradient cooling system to adapt to the temperature change from 193.15K to 77.15K. This completes the entire simulation process, with the total simulation time set to 400ns. The thawing method of Populus tomentosa callus was screened based on the dehydrogenase stability of two cryopreservation methods.

10. The method for cryopreservation of *Populus silvereris* callus based on machine learning and molecular dynamics simulation according to claim 1, characterized in that, When the thawing method is determined to be gradient cooling thawing, the step of thawing and reculturing the frozen Populus spp. callus material using the determined thawing method refers to: Two weeks later, the frozen tubes stored in the liquid nitrogen biological container were removed and thawed in a 40°C water bath. After the tissues were thawed, the cryoprotectant was filtered out. The poplar callus tissues in each cryotube were washed 3-4 times with liquid induction medium. The liquid medium on the surface of the poplar callus tissues was blotted dry with filter paper, and the poplar callus tissues were placed in solid proliferation medium.

Images