An aging prediction method, device and equipment based on age enhancement and a storage medium

Abstract

The application provides an aging prediction method and device based on age enhancement, equipment and a storage medium. A joint latent vector is constructed according to RNA sequencing data and DNA methylation data of health sample data through a plurality of omics variational autoencoders of a prediction network. Age-ordered enhancement contrast learning is performed according to a multilayer perceptron of the prediction network and the joint latent vector, and an age baseline arranged in an age order in a latent space is determined, which is beneficial to improving in-depth understanding and analysis accuracy of the aging process. The prediction network is trained according to the joint latent vector and a first loss function, and the trained prediction network is fine-tuned through RNA sequencing data and DNA methylation data of pan-cancer sample data to obtain a prediction model. The age prediction result of each age group is determined according to the age baseline of the prediction model and the pan-cancer sample data, so as to determine the aging condition of each individual and improve the judgment accuracy of the aging condition.

Description

Technical Field

[0001] This application relates to the field of deep learning, and in particular to an age-enhanced aging prediction method, device, equipment, and storage medium. Background Technology

[0002] Aging is a complex process characterized by declining bodily functions and the accumulation of cellular and molecular damage, influenced by a variety of factors. With the development of high-throughput sequencing technology, multi-omics data has enabled the estimation of biological age at the molecular level, thus driving the development of different aging clocks. "Age difference," as a key indicator in aging research, reflects the degree of acceleration of individual aging and is closely related to health and lifespan. A positive age difference indicates accelerated aging and an increased risk of disease and death, while a smaller or negative age difference indicates slower aging and is often associated with better health and longer lifespan. Currently, there are aging studies utilizing age factors and based on deep learning, but most methods rely on single-omics data, such as DNA methylation or transcriptomics, lacking multi-omics integration, which limits a comprehensive and in-depth understanding of the aging process. Summary of the Invention

[0003] This application provides an age-enhancing aging prediction method, apparatus, device, and storage medium to address at least one problem existing in related technologies. The technical solution is as follows:

[0004] In a first aspect, embodiments of this application provide a method for age-enhancing aging prediction, comprising:

[0005] We obtained healthy sample data and pan-cancer sample data from several individuals in different age groups. Both the healthy sample data and the pan-cancer sample data included paired RNA sequencing data and DNA methylation data.

[0006] By using a multi-omics variational autoencoder of the prediction network, a joint latent vector is constructed based on RNA sequencing data and DNA methylation data of healthy samples. Then, based on the multilayer perceptron of the prediction network and the joint latent vector, age-order-based enhanced contrastive learning is performed to determine the age baseline in the latent space arranged in age order. The age baseline is used for age prediction.

[0007] Based on the joint latent vector and the first loss function, a prediction network is trained, and the trained prediction network is fine-tuned using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain a prediction model.

[0008] Based on pan-cancer sample data and the age baseline of the prediction model, the age prediction results for each age group are determined, and based on the age prediction results, the aging status of each individual is determined.

[0009] In one implementation, the construction of a joint latent vector based on RNA sequencing data and DNA methylation data from healthy sample data using a multi-omics variational autoencoder of a prediction network includes:

[0010] The first omics latent vector of RNA sequencing data and the second omics latent vector of DNA methylation data in healthy sample data are generated by the multiomics variational autoencoder of the prediction network.

[0011] Using the POE framework, the posterior distribution of the joint latent vector of the first and second single-group latent vectors is calculated, and the posterior distribution of the joint latent vector is constrained by KL divergence to generate the joint latent vector of each individual in different age groups.

[0012] In one implementation, the step of performing age-order-based augmented contrastive learning based on the multilayer perceptrons of the prediction network and the joint latent vectors to determine the age baselines arranged in age order in the latent space includes:

[0013] By using a multilayer perceptron in the prediction network, the agent vector of each age group center is determined based on the joint latent vector.

[0014] Calculate the cosine similarity between each joint latent vector and the surrogate vector of the same age group to which the joint latent vector belongs, and use it as the first weight of the positive sample constraint.

[0015] Calculate the age difference between each joint latent vector and the center of the different age groups to which the joint latent vector belongs, and use the age difference to dynamically weight negative samples as the second weight;

[0016] Based on the first weight and the second weight, the age baseline of each joint latent vector in the latent space is determined in order of age.

[0017] In one implementation, determining the age baseline of each joint latent vector arranged in age order in the latent space according to the first weight and the second weight includes:

[0018] The orientation loss constraint ensures that the orientation of each joint latent vector is consistent with the orientation of the corresponding surrogate vector.

[0019] Based on the first weight, second weight, and directional consistency of each joint latent vector, the age baseline of each joint latent vector arranged in age order in the latent space is determined.

[0020] In one implementation, training the prediction network based on the joint latent vector and the first loss function includes:

[0021] ELBO based on multilayer perceptron decodes and reconstructs joint latent vectors to obtain reconstructed RNA sequencing data and reconstructed DNA methylation data.

[0022] Based on the reconstructed RNA sequencing data, the reconstructed DNA methylation data, and the first loss function, the loss value is determined, and the prediction network is trained based on the loss value until the loss value is less than a preset threshold.

[0023] In one implementation, the step of fine-tuning the trained prediction network using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain the prediction model includes:

[0024] Using RNA sequencing data and DNA methylation data from pan-cancer sample data, the trained prediction network was trained based on the second loss function. During the training process, the parameters of the multi-omics variational autoencoder and the multilayer perceptron were adjusted using Adam optimizers with different learning rates until the training termination condition was met, thus obtaining the prediction model.

[0025] In one implementation, determining the aging status of each individual based on age prediction results includes:

[0026] Based on the age prediction results for each age group, the average predicted age for each age group is calculated and used as the anchor age for that age group.

[0027] Calculate the age difference between the predicted age of each individual and the anchor age of the actual age group to which the individual belongs, and calculate the standard deviation σ based on the age differences of all individuals;

[0028] When the age difference is greater than σ, the aging status of the corresponding individual is determined to be accelerated aging; when -σ ≤ age difference ≤ σ, the aging status of the corresponding individual is determined to be normal aging; when the age difference is less than -σ, the aging status of the corresponding individual is determined to be decelerated aging.

[0029] Secondly, embodiments of this application provide an age-enhancing aging prediction device, comprising:

[0030] The acquisition module is used to acquire healthy sample data and pan-cancer sample data of several individuals in different age groups. Both the healthy sample data and the pan-cancer sample data include paired RNA sequencing data and DNA methylation data.

[0031] The processing module is used to construct a joint latent vector based on RNA sequencing data and DNA methylation data of healthy sample data through the multi-omics variational autoencoder of the prediction network, and perform age-order-based enhanced contrastive learning based on the multilayer perceptron of the prediction network and the joint latent vector to determine the age baseline arranged in age order in the latent space. The age baseline is used for age prediction.

[0032] The training module is used to train the prediction network based on the joint latent vector and the first loss function, and to fine-tune the trained prediction network using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain the prediction model.

[0033] The determination module is used to determine the age prediction results for each age group based on pan-cancer sample data and the age baseline of the prediction model, and to determine the aging status of each individual based on the age prediction results.

[0034] Thirdly, embodiments of this application provide an electronic device, including: a processor and a memory, wherein the memory stores instructions that are loaded and executed by the processor to implement the methods in any of the above-described embodiments.

[0035] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed, implements the methods in any of the above-described embodiments.

[0036] The beneficial effects of the above technical solution include at least the following:

[0037] By acquiring healthy sample data and pan-cancer sample data from several individuals in different age groups, including paired RNA sequencing and DNA methylation data, a joint latent vector is constructed using a multi-omics variational autoencoder of the prediction network based on the RNA sequencing and DNA methylation data of the healthy sample data. Then, based on the multilayer perceptron of the prediction network and the joint latent vector, age-order-based augmented contrastive learning is performed to determine the age baseline in the latent space arranged in age order. Analysis is then performed based on the multi-omics data combined with the prediction network and age-order-based augmented contrastive learning. The age baseline used for age prediction is represented through the latent space, which helps improve the understanding and accuracy of the aging process. The prediction network is trained based on the joint latent vector and a first loss function, and then fine-tuned using the RNA sequencing and DNA methylation data of the pan-cancer sample data to obtain the prediction model. Based on the pan-cancer sample data and the age baseline of the prediction model, the age prediction results for each age group are determined, and the aging status of each individual is determined based on the age prediction results, which helps improve the accuracy of aging status assessment.

[0038] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, these aspects, embodiments, and features will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0039] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments disclosed in this application and should not be construed as limiting the scope of this application.

[0040] Figure 1 This is a schematic flowchart illustrating the steps of an age-enhancing aging prediction method according to an embodiment of this application;

[0041] Figure 2 This is a schematic diagram of the processing procedure of a prediction network according to an embodiment of this application;

[0042] Figure 3 This is a schematic diagram of a PAAG according to an embodiment of this application;

[0043] Figure 4 This is a structural block diagram of an age-enhanced aging prediction device according to an embodiment of this application;

[0044] Figure 5 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0045] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of this application. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0046] In existing deep learning-based aging research, models, besides those based solely on single-omics data, are typically regression models trained using actual age as the dependent variable. This can easily lead to prediction bias, ignore individual differences, and result in inaccurate assessments of aging in specific individuals. Furthermore, the training features of these models are prone to overfitting. Compared to high-dimensional omics data, the datasets used for training in these methods are often small, and due to the characteristics of deep learning, overfitting is likely, leading to poor generalization performance. In this application, a large amount of paired RNA sequencing data and DNA methylation data are used to avoid overfitting.

[0047] Reference Figure 1 The flowchart illustrates an embodiment of an age-enhanced aging prediction method according to this application. This age-enhanced aging prediction method may include at least steps S100-S400:

[0048] S100: Obtain health sample data and pan-cancer sample data from several individuals in different age groups.

[0049] The healthy sample data and the pan-cancer sample data both include paired RNA sequencing data and DNA methylation data.

[0050] S200: Using a multi-omics variational autoencoder of the prediction network, a joint latent vector is constructed based on RNA sequencing data and DNA methylation data of healthy sample data. Based on the multilayer perceptron of the prediction network and the joint latent vector, age-order-based enhanced contrastive learning is performed to determine the age baseline arranged in age order in the latent space. The age baseline is used for age prediction.

[0051] S300. Based on the joint latent vector and the first loss function, train the prediction network, and fine-tune the trained prediction network using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain the prediction model.

[0052] S400. Based on pan-cancer sample data and the age baseline of the prediction model, determine the age prediction results for each age group, and based on the age prediction results, determine the aging status of each individual.

[0053] The technical solution of this application embodiment acquires healthy sample data and pan-cancer sample data from several individuals in different age groups. Both the healthy sample data and the pan-cancer sample data include paired RNA sequencing data and DNA methylation data. A multi-omics variational autoencoder of the prediction network is used to construct a joint latent vector based on the RNA sequencing data and DNA methylation data of the healthy sample data. Then, based on the multilayer perceptron of the prediction network and the joint latent vector, age-order-based enhanced contrastive learning is performed to determine the age baseline arranged in age order in the latent space. Analysis is performed based on multi-omics data combined with the prediction network and age-order-based enhanced contrastive learning. The age baseline used for age prediction is represented through the latent space, which helps improve the understanding and accuracy of the aging process. The prediction network is trained based on the joint latent vector and a first loss function. The trained prediction network is then fine-tuned using the RNA sequencing data and DNA methylation data of the pan-cancer sample data to obtain a prediction model. Based on the pan-cancer sample data and the age baseline of the prediction model, the age prediction results for each age group are determined. Based on the age prediction results, the aging status of each individual is determined, which helps improve the accuracy of aging status assessment.

[0054] In one implementation, several healthy sample data and pan-cancer sample data can be obtained from the TCGA database and / or GTEx database. For example, the healthy sample data includes data from 7862 healthy tissue samples (31 types of tissues / organs, each with corresponding tissue / organ tags) from several individuals, and the pan-cancer sample data includes data from 9621 pan-cancer samples from several individuals. Each healthy tissue sample and each pan-cancer sample includes paired RNA sequencing data (RNA-seq data) and DNA methylation data. Optionally, after obtaining the data, appropriate preprocessing can be performed, and the preprocessed data can then be used for subsequent analysis. For example, RNA sequencing data: after filtering low-expression genes, 11,816 features were retained and normalized to log2(TPM+1). DNA methylation data: probes with β values ​​<0.3 in 90% of samples were excluded, and 6,617 CpG sites were retained. At the same time, based on a preset age separator K (e.g., K=10), as an age label for subsequent analysis, the acquired data were divided according to the individual's actual age, for example, into multiple age groups (age intervals) such as 10-20 years old, 20-30 years old, 30-40 years old, 40-50 years old, and 50-60 years old. Then, each age group contains data on healthy tissue samples and pan-cancer samples corresponding to the individual.

[0055] In this embodiment, the prediction network is AOE-Net (Age Order Enhanced Network), which includes a multi-omics variational autoencoder and an MLP (Multilayer Perceptron, also referred to as an age estimator in this application).

[0056] In one implementation, step S200 involves constructing a joint latent vector using a multi-omics variational autoencoder of a prediction network, based on RNA sequencing data and DNA methylation data from healthy sample data, including steps S210-S220:

[0057] S210. The first single-omics latent vector of RNA sequencing data and the second single-omics latent vector of DNA methylation data in the healthy sample data are generated by the multi-omics variational autoencoder of the prediction network.

[0058] like Figure 2 As shown, optionally, RNA sequencing data (including mRNA) and DNA methylation data from healthy samples of several individuals in different age groups are used as pre-training datasets. These datasets are encoded by a multi-omics variational autoencoder (referred to as encoder) of the prediction network to generate the first single-omics latent vector of RNA sequencing data (including mRNA) and the second single-omics latent vector of DNA methylation data from the healthy sample data, respectively.

[0059] S220. Using the POE framework, calculate the posterior distribution of the joint latent vector of the first and second single-group latent vectors, and generate the joint latent vector of each individual in different age groups by constraining the posterior distribution of the joint latent vector through KL divergence.

[0060] Optionally, the first and second single-set latent vectors are jointly processed using the POE (Perl Object Environment) framework. The posterior distribution of the joint latent vector of the first and second single-set latent vectors is calculated, and the posterior distribution of the joint latent vector is constrained by KL divergence to approximate a standard Gaussian distribution (e.g., ...). Figure 2 The latent vectors of multiple omics are fused into the latent space (or simply latent space) to avoid distribution shift, thereby generating a joint latent vector for each individual in different age groups.

[0061] In one implementation, step S200 involves performing age-order-based augmented contrastive learning based on the multilayer perceptrons and joint latent vectors of the prediction network to determine age baselines arranged in age order in the latent space, including steps S230-S260:

[0062] S230. Using the multilayer perceptron of the prediction network, determine the agent vector of the center of each age group based on the joint latent vector.

[0063] Optionally, since the joint latent vector follows a Gaussian distribution, the surrogate vector can be sampled by using a multilayer perceptron (MLP) of the prediction network to determine the center of each age group and the surrogate vector of the age group center based on the joint latent vector.

[0064] S240. Calculate the cosine similarity between each joint latent vector and the surrogate vector of the same age group to which the joint latent vector belongs, and use it as the first weight of the positive sample constraint.

[0065] Optionally, the cosine similarity between each joint latent vector and the surrogate vector of the same age group to which the joint latent vector belongs is calculated as the first weight of the positive sample constraint, and the higher the cosine similarity, the greater the first weight.

[0066] S250. Calculate the age difference between each joint latent vector and the center of the different age groups to which the joint latent vector belongs, and use the age difference to dynamically weight negative samples as the second weight.

[0067] Optionally, the age difference between each joint latent vector and the centers of different age groups to which the joint latent vector belongs is calculated. Negative samples are dynamically weighted using this age difference as the second weight; the larger the age difference, the lower the second weight of the negative samples, thereby reducing cross-age group interference. Adjusting the magnitudes of the first and second weights corresponds to... Figure 2The loss of the middle triplet is narrowed and pushed back; the embodiments of this application, by adjusting the first weight and the second weight, take into account individual age differences, which is conducive to ensuring the accuracy of subsequent aging assessment.

[0068] S260. Based on the first weight and the second weight, determine the age baseline of each joint latent vector arranged in the latent space according to age order.

[0069] In this embodiment, in addition to adjusting the magnitudes of the first and second weights, a directional loss constraint is introduced. This constraint ensures that the direction of the latent vectors of samples from different age groups is consistent with the direction of their corresponding surrogate vectors (e.g., the direction of the joint latent vector for ages 30-40 is restricted to be between the surrogate vectors for ages 20-30 and surrogate vectors for ages 40-50). This guarantees that the age intervals in the latent space are arranged in an ordered manner according to size 1→2→3, such as age interval 1: 20-30 years old, age interval 2: 30-40 years old, and age interval 3: 40-50 years old (corresponding to...). Figure 2 The age ranges are 1, 2, and 3. Finally, based on the first weight, the second weight, and directional consistency, the spatial structure of the joint latent vectors in the latent space is optimized to obtain the age baseline of each joint latent vector arranged in age order in the latent space, that is, the comprehensive age baseline after considering the age order loss, which is used for subsequent age prediction.

[0070] In one implementation, step S300 involves training a prediction network based on the joint latent vector and the first loss function, including steps S310-S320:

[0071] S310, based on the multilayer perceptron ELBO, decodes and reconstructs the joint latent vector to obtain reconstructed RNA sequencing data and reconstructed DNA methylation data.

[0072] Optionally, the joint latent vector is decoded and reconstructed based on the ELBO (Evidence Lower Bound) of the decoder in the multilayer perceptron, thereby obtaining the reconstructed RNA sequencing data and the reconstructed DNA methylation data.

[0073] S320. Based on the reconstructed RNA sequencing data, the reconstructed DNA methylation data, and the first loss function, determine the loss value, and train the prediction network based on the loss value until the loss value is less than a preset threshold.

[0074] Optionally, the first loss function is:

[0075] L pretrain =L recon +β1KL+β2L order

[0076] Among them, L pretrainLet L be the loss value of the first loss function. recon The difference between reconstructed methylation data, reconstructed DNA methylation data, and RNA sequencing and DNA methylation data from healthy samples is calculated. KL is the value obtained by weighting the logarithmic ratio of the distribution and probability density, with β1 and β2 as weights, and L... order This is due to age order loss.

[0077] It should be noted that the age order loss L order This is used to utilize the age order information of samples to guide multi-omics data to form a reasonable and ordered representation in the latent space, specifically:

[0078] L order =L contrast +λL direction

[0079] In the formula, L contrast For agent-based contrast loss L contrast L direction The loss is for sequential perception of direction, and λ is the balancing weight, for example, λ is 0.5.

[0080] Among them, the comparison loss based on the agent L contrast The goal is to cluster samples belonging to the same age group (age segment) in the latent space, while dynamically expanding samples from different age segments based on age differences:

[0081] 1. The model learns a surrogate vector p representing the center k of each age group (age segment). k For a given age label (e.g., a specific age) as a i Sample embedding Z i (Sample latent vector), and its corresponding surrogate vector i is a positive agent, and all other agent vectors are negative agents, where i and k are integers.

[0082] 2. Similarity Calculation: The similarity s(z) between the sample embedding and the surrogate vector. i ,p k ) through cosine similarity cos(z) i ,p k )calculate: Where τ is the temperature hyperparameter, which defaults to 0.9.

[0083] 3. Negative sample weighting: Negative samples are dynamically weighted according to their age difference with positive samples. The greater the age difference, the higher the weight.

[0084] Finally, the loss function L contrast definition:

[0085]

[0086] Where N is the number of samples.

[0087] Additionally, the sequential perception orientation loss L direction Used to ensure sample embedding z in the latent space i It can maintain an arrangement consistent with the actual age order, and the loss of sequence perception direction is L. direction Designed to align direction vectors between sample embeddings and between proxies:

[0088] 1. Calculate the direction vector: Calculate the embedding z of any two samples. i z j The direction vector d between i,j , and z i z j The corresponding proxy vectors The direction vector d′ between i,j d i,j =normalize(z i -zj,di,j′=normalizepai-paj;

[0089] 2. Calculate the directional similarity s dir (i,j): L direction Definition: Using a mask M i,j (when a i >a j The value is 1 if the condition is met and 0 otherwise, to ensure that only sample pairs that conform to the age order are considered.

[0090]

[0091] Therefore, based on the reconstructed RNA sequencing data, the reconstructed DNA methylation data, and the first loss function, the corresponding loss value can be calculated. The prediction network is then trained based on the loss value. During the training process, the network parameters are continuously adjusted until the loss value is less than the preset threshold, thus obtaining the trained prediction network.

[0092] In one implementation, step S300 involves fine-tuning the trained prediction network using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain a prediction model, including step S330:

[0093] S330. Using RNA sequencing data and DNA methylation data from pan-cancer sample data, the trained prediction network is trained based on the second loss function. During the training process, Adam optimizers with different learning rates are used to adjust the parameters of the multi-omics variational autoencoder and the multilayer perceptron until the training termination condition is met, thus obtaining the prediction model.

[0094] Optionally, to further improve the accuracy of the prediction network, it can be fine-tuned using pan-cancer sample data. Specifically, the prediction network is trained using RNA sequencing data and DNA methylation data from the pan-cancer sample data, based on a second loss function. Similarly, a new joint latent vector is generated based on POE to determine the latent distribution. However, care is taken to maintain the latent space structure of the healthy baseline (i.e., the age baseline) during training, using Adam optimizers with different learning rates (e.g., a multi-omics variational autoencoder with a learning rate of 10). -6 The learning rate of the multilayer perceptron is {10}^{-3}. The parameters of the multi-group variational autoencoder and the multilayer perceptron are adjusted respectively until the training termination condition is reached (e.g., 50 epochs, and the loss value of the second loss function has not changed for 10 epochs). The fine-tuned prediction network is obtained, that is, the prediction model. The prediction model can output the predicted biological age based on its age baseline.

[0095] Optionally, the second loss function consists of cross-entropy loss and regression loss. The cross-entropy loss uses a Gaussian target distribution to ensure that the distribution of the predicted biological age is consistent with the actual age; the regression loss reduces the absolute error between the predicted biological age and the actual age. The two are balanced by α weights, ultimately making the prediction more accurate. It should be noted that the stability of the predicted biological age (hereinafter referred to as the predicted age) can be evaluated through 5-fold cross-validation to guide whether the prediction model needs to be fine-tuned.

[0096] In this embodiment, the prediction network is fine-tuned to predict physiological age using actual age as a supervisory label. Through transfer learning, it can demonstrate superior accuracy and clinical relevance to existing aging clocks in various chronic diseases.

[0097] In one implementation, step S400 determines the age prediction results for each age group based on the pan-cancer sample data and the age baseline of the prediction model. Specifically, the RNA sequencing data and DNA methylation data of the pan-cancer sample data are input into the prediction model to obtain the age prediction results for each age group.

[0098] In one implementation, step S400, based on the age prediction results, determines the aging status of each individual, including steps S410-S430:

[0099] S410. Based on the age prediction results of each age group, calculate the average predicted age of each age group as the anchor age for that age group.

[0100] Optionally, taking the 20-30 age group as an example, this age group has N individuals, each with a corresponding predicted age. In this case, the average predicted age of the 20-30 age group can be obtained by summing the predicted ages of the N individuals and dividing by N, and used as the anchor age for the 20-30 age group. Therefore, based on this principle, the anchor age for each age group can be calculated.

[0101] S420. Calculate the age difference between the predicted age of each individual and the anchor age of the actual age group to which the individual belongs, and calculate the standard deviation σ based on the age difference of all individuals.

[0102] Optionally, an individual might be predicted to be between 20 and 30 years old, but their actual age group is 30 to 40 years old. Therefore, the age difference between the individual's predicted age and the anchor age of their actual age group (30-40 years old) is calculated. Based on this principle, the age difference for all individuals can be calculated, and then the corresponding standard deviation σ can be calculated based on the age difference for all individuals.

[0103] S430. When the age difference is greater than σ, the aging status of the corresponding individual is determined to be accelerated aging; when -σ ≤ age difference ≤ σ, the aging status of the corresponding individual is determined to be normal aging; when the age difference is less than -σ, the aging status of the corresponding individual is determined to be decelerated aging.

[0104] like Figure 3As shown in the embodiments of this application, the standard deviation σ is used as the threshold. When the age difference > σ, the aging status of the corresponding individual is determined to be accelerated aging; when -σ ≤ age difference ≤ σ, the aging status of the corresponding individual is determined to be normal aging; and when the age difference < -σ, the aging status of the corresponding individual is determined to be decelerated aging. Ultimately, the aging status of all individuals can be accurately determined. This application embodiment proposes to determine the aging status of individuals based on the aforementioned multi-omics quantitative method PAAG (Population-Aware Age Gap), improving the traditional age gap indicator by assessing the aging status through the age difference between the predicted age and the corresponding anchor age, thus reflecting the aging trend of the population more meticulously. In this embodiment, a stratified analysis of overall survival and immune response in a pan-cancer population composed of pan-cancer individuals was further performed using PAAG. The results showed that: using 5-fold cross-validation to predict and assess physiological age, and then dividing the pan-cancer samples into three groups according to PAAG—a decelerating aging group, a normal aging group, and an accelerated aging group—significant differences in overall survival were observed among these groups (Kaplan-Meier multivariate Logrank test, p = 1.93 × 10⁻¹⁵), with the accelerated aging group showing the worst prognosis, consistent with previous findings. Furthermore, based on the model's predictive ability for different cancer types, the hazard ratio (HR) of PAAG revealed different risk patterns. For example, low-grade glioma (LGG) showed a strong association between accelerated aging and poor prognostic outcomes (p = 9.25 × 10⁻⁴). Especially in comparison with existing multi-omics aging clocks (such as DeepMAge, BiT-Age) and traditional AG indicators in ablation experiments, AOE-Net+PAAG outperformed these methods, achieving a higher concordance index (C-index) and HR. If AOE-Net is not used, updating existing aging clock estimates with PAAG alone also improves prognostic accuracy compared to traditional AG, highlighting the value of PAAG as a reliable biomarker for aging.

[0105] Furthermore, the association between PAAG and tumor immunity is a key focus, as senescent cells are known to influence immune infiltration in the tumor microenvironment; studies using related techniques have shown that senescent cells in the tumor microenvironment can enhance the immune response against tumor cells. In our results, individuals in the previously known immune subtype C5 (i.e., “immunely quiescent”) had the lowest PAAG values, while other subtypes showed higher PAAG values. Additionally, individuals within immune subtype C5 had the smallest actual age compared to other subtypes. Meanwhile, the immune checkpoint feature 5 (ICS5) score was significantly lower in the aging deceleration group (p = 4.44 × 10⁻²⁸). Consistent with previous studies, the normal aging and accelerated aging groups had higher immune response scores, likely due to the higher number of senescent cells in the microenvironment, while the aging deceleration group showed lower scores even in immunely quiescent environments, such as immune subtype C5.

[0106] In this embodiment, through the training and fine-tuning process of the aforementioned prediction network, the obtained prediction model achieves accurate prediction of biological age. It exhibits significant age-related structural features in the latent space—the latent space representations matching the actual age groups show a clear gradual process, and the pseudo-time trajectories derived from these latent space representations are closely related to the actual age, thus supporting the contrastive learning of the prediction model. It can successfully arrange the vectors corresponding to organizational samples according to biological age in the latent space. Simultaneously, the age-based latent space in healthy individuals provides a solid foundation of population age structure for downstream transfer learning of pan-cancer samples, thereby accurately detecting deviations from normal aging patterns caused by disease (i.e., the gap between the population's biological age and physiological age).

[0107] In summary, this application proposes a prediction network (AOE-Net) based on multi-omics data, which can accurately capture the correlation between different omics data. The age baseline (health baseline) obtained by enhanced contrastive learning based on age order can be applied to a wide range of downstream tasks. At the same time, a new multi-omics measurement method, PAAG, is developed based on the prediction network to overcome the limitations of existing methods in underestimating individual biological aging changes, accurately assess physiological age and accelerated aging, more meticulously reflect the aging trend of the population, and accurately mine aging risks in complex diseases.

[0108] Reference Figure 4 The diagram illustrates a structural block diagram of an age-enhanced aging prediction device according to an embodiment of this application. The device may include:

[0109] The acquisition module is used to acquire healthy sample data and pan-cancer sample data of several individuals in different age groups. Both the healthy sample data and the pan-cancer sample data include paired RNA sequencing data and DNA methylation data.

[0110] The processing module is used to construct a joint latent vector based on RNA sequencing data and DNA methylation data of healthy sample data through the multi-omics variational autoencoder of the prediction network, and perform age-order-based enhanced contrastive learning based on the multilayer perceptron of the prediction network and the joint latent vector to determine the age baseline arranged in age order in the latent space. The age baseline is used for age prediction.

[0111] The training module is used to train the prediction network based on the joint latent vector and the first loss function, and to fine-tune the trained prediction network using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain the prediction model.

[0112] The determination module is used to determine the age prediction results for each age group based on pan-cancer sample data and the age baseline of the prediction model, and to determine the aging status of each individual based on the age prediction results.

[0113] The functions of each module in the device of this application embodiment can be found in the corresponding description in the above method, and will not be repeated here.

[0114] Reference Figure 5 The diagram illustrates a structural block diagram of an electronic device according to an embodiment of this application. The electronic device includes a memory 310 and a processor 320. The memory 310 stores instructions that can be executed on the processor 320. The processor 320 loads and executes these instructions to implement the age-enhanced aging prediction method described in the above embodiment. The number of memories 310 and processors 320 can be one or more.

[0115] In one embodiment, the electronic device further includes a communication interface 330 for communicating with external devices and exchanging data. If the memory 310, processor 320, and communication interface 330 are implemented independently, they can be interconnected via a bus to communicate with each other. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0116] Optionally, in a specific implementation, if the memory 310, processor 320 and communication interface 330 are integrated on a single chip, the memory 310, processor 320 and communication interface 330 can communicate with each other through an internal interface.

[0117] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the age-enhancing aging prediction method provided in the above embodiments.

[0118] This application also provides a chip, which includes a processor for calling and executing instructions stored in a memory, causing a communication device on which the chip is installed to perform the method provided in this application.

[0119] This application also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, output interface, processor, and memory are connected through an internal connection path. The processor is used to execute code in the memory. When the code is executed, the processor is used to execute the method provided in the application embodiment.

[0120] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting the Advanced Reduced Instruction Set Computing (RISC) machine (ARM) architecture.

[0121] Further, optionally, the aforementioned memory may include read-only memory and random access memory, and may also include non-volatile random access memory. The memory may be volatile or non-volatile, or may include both. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. Many forms of RAM are available by way of example, but not limitation. Examples include static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0122] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0123] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0124] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0125] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0126] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0127] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0128] Furthermore, the functional units in the various embodiments of this application can be integrated into a single processing module, or each unit can exist physically separately, or two or more units can be integrated into a single module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. This storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0129] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An age-enhancing aging prediction method, characterized in that, include: We obtained healthy sample data and pan-cancer sample data from several individuals in different age groups. Both the healthy sample data and the pan-cancer sample data included paired RNA sequencing data and DNA methylation data. By using a multi-omics variational autoencoder of the prediction network, a joint latent vector is constructed based on RNA sequencing data and DNA methylation data of healthy samples. Then, based on the multilayer perceptron of the prediction network and the joint latent vector, age-order-based enhanced contrastive learning is performed to determine the age baseline in the latent space arranged in age order. The age baseline is used for age prediction. Based on the joint latent vector and the first loss function, a prediction network is trained, and the trained prediction network is fine-tuned using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain a prediction model. Based on pan-cancer sample data and the age baseline of the prediction model, the age prediction results for each age group are determined, and based on the age prediction results, the aging status of each individual is determined.

2. The age-enhancing aging prediction method according to claim 1, characterized in that: The construction of joint latent vectors using a multi-omics variational autoencoder through a prediction network, based on RNA sequencing data and DNA methylation data from healthy sample data, includes: The first omics latent vector of RNA sequencing data and the second omics latent vector of DNA methylation data in healthy sample data are generated by the multiomics variational autoencoder of the prediction network. Using the POE framework, the posterior distribution of the joint latent vector of the first and second single-group latent vectors is calculated, and the posterior distribution of the joint latent vector is constrained by KL divergence to generate the joint latent vector of each individual in different age groups.

3. The age-enhancing aging prediction method according to claim 2, characterized in that: The step of performing age-order-based augmented contrastive learning based on the multilayer perceptron and joint latent vectors of the prediction network to determine the age baseline arranged in age order in the latent space includes: By using a multilayer perceptron in the prediction network, the agent vector of each age group center is determined based on the joint latent vector. Calculate the cosine similarity between each joint latent vector and the surrogate vector of the same age group to which the joint latent vector belongs, and use it as the first weight of the positive sample constraint. Calculate the age difference between each joint latent vector and the center of the different age groups to which the joint latent vector belongs, and use the age difference to dynamically weight negative samples as the second weight; Based on the first and second weights, the age baselines of each joint latent vector in the latent space are determined in order of age.

4. The age-enhancing aging prediction method according to claim 3, characterized in that: The step of determining the age baseline for each joint latent vector arranged in age order in the latent space based on the first weight and the second weight includes: The orientation loss constraint ensures that the orientation of each joint latent vector is consistent with the orientation of the corresponding surrogate vector. Based on the first weight, second weight, and directional consistency of each joint latent vector, the age baseline of each joint latent vector arranged in age order in the latent space is determined.

5. The age-enhancing aging prediction method according to any one of claims 1-4, characterized in that: The step of training the prediction network based on the joint latent vector and the first loss function includes: ELBO based on multilayer perceptron decodes and reconstructs joint latent vectors to obtain reconstructed RNA sequencing data and reconstructed DNA methylation data. Based on the reconstructed RNA sequencing data, the reconstructed DNA methylation data, and the first loss function, the loss value is determined, and the prediction network is trained based on the loss value until the loss value is less than a preset threshold.

6. The age-enhancing aging prediction method according to claim 5, characterized in that: The prediction model obtained by fine-tuning the trained prediction network using RNA sequencing data and DNA methylation data from pan-cancer sample data includes: Using RNA sequencing data and DNA methylation data from pan-cancer sample data, the trained prediction network was trained based on the second loss function. During the training process, the parameters of the multi-omics variational autoencoder and the multilayer perceptron were adjusted using Adam optimizers with different learning rates until the training termination condition was met, thus obtaining the prediction model.

7. The age-enhancing aging prediction method according to any one of claims 1-4, characterized in that: The determination of each individual's aging status based on age prediction results includes: Based on the age prediction results for each age group, the average predicted age for each age group is calculated and used as the anchor age for that age group. Calculate the age difference between the predicted age of each individual and the anchor age of the actual age group to which the individual belongs, and calculate the standard deviation σ based on the age differences of all individuals; When the age difference is greater than σ, the aging status of the corresponding individual is determined to be accelerated aging; when -σ ≤ age difference ≤ σ, the aging status of the corresponding individual is determined to be normal aging; when the age difference is less than -σ, the aging status of the corresponding individual is determined to be decelerated aging.

8. An aging prediction device based on age-enhancing features, characterized in that, include: The acquisition module is used to acquire healthy sample data and pan-cancer sample data of several individuals in different age groups. Both the healthy sample data and the pan-cancer sample data include paired RNA sequencing data and DNA methylation data. The processing module is used to construct a joint latent vector based on RNA sequencing data and DNA methylation data of healthy sample data through the multi-omics variational autoencoder of the prediction network, and to perform age-order-based enhanced contrastive learning based on the multilayer perceptron of the prediction network and the joint latent vector to determine the age baseline arranged in age order in the latent space. The age baseline is used for age prediction. The training module is used to train the prediction network based on the joint latent vector and the first loss function, and to fine-tune the trained prediction network using RNA sequencing data and DNA methylation data from pan-cancer sample data to obtain the prediction model. The determination module is used to determine the age prediction results for each age group based on pan-cancer sample data and the age baseline of the prediction model, and to determine the aging status of each individual based on the age prediction results.

9. An electronic device, characterized in that, include: A processor and a memory, wherein instructions are stored in the memory and loaded and executed by the processor to implement the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program therein, which, when executed, implements the method as described in any one of claims 1-7.

Images