Aging rate calculation method and storage medium

Abstract

This disclosure provides a method and storage medium for calculating aging rate, relating to the field of biological data analysis technology. The method includes: modeling the complex nonlinear relationship between the actual age, biomarkers, and mortality risk of a biological object using a mortality risk prediction model constructed based on a neural network; obtaining a mortality risk function for the target biological object through the determined mortality risk prediction model, and extracting an aging rate index with clear biological significance from the mortality risk function; and mapping the target aging rate to biological age through an age-aging relationship function, thereby accurately reflecting the degree of aging of the target biological object.

Description

Technical Field

[0001] This disclosure relates to the field of biological data analysis technology, and in particular to methods for calculating aging rates and storage media. Background Technology

[0002] Aging is a complex process in which the physiological functions of an organism gradually decline over time. As a core indicator characterizing an individual's physiological state and degree of aging, the aging rate, compared to actual age which reflects the natural progression of time, can more accurately reflect an individual's physical condition, differences in aging, and potential disease risks. Therefore, it has significant application value in fields such as geriatrics, disease risk prediction, and health management. How to accurately quantify an individual's aging rate from observable biomarkers is one of the key technical challenges in the field of aging research.

[0003] Existing methods for calculating aging rates and biological age can be mainly categorized as follows:

[0004] The first category is based on classical statistical models. For example, the traditional Gompertz model is based on the exponential increase in mortality risk with age, but its parameters are usually given in a fixed form, which is more suitable for describing the average aging trend of a group and is difficult to directly reflect individual differences. Although the Cox proportional hazards model can introduce covariates, its linear assumption makes it difficult to characterize the complex nonlinear relationship between biomarkers and mortality risk, and it cannot directly output an aging rate indicator with clear biological significance.

[0005] The second category is biomarker-weighted scoring methods. These methods typically determine the weights of each biomarker through statistical regression and then sum them to obtain a biological age score. However, the scores obtained by these methods lack characterization of the dynamics of the aging process, fail to reflect the increasing trend of individual mortality risk over time, and have limited ability to model the complex interactions between multidimensional biomarkers.

[0006] The third category is machine learning-based methods. While these methods can improve their ability to fit the complex relationship between biomarkers and physiological states by using models such as deep neural networks, their outputs usually lack clear biological meaning and interpretability, making it difficult to directly correlate them with the dynamics of mortality risk. As a result, the calculated aging rate or biological age is difficult to correlate with the individual's actual physiological aging process.

[0007] In summary, existing technologies lack a computational method that can both characterize the complex nonlinear relationship between biomarkers and mortality risk and output an aging rate indicator with clear biological significance. Summary of the Invention

[0008] In view of the shortcomings of the prior art described above, the purpose of this disclosure is to provide a method for calculating aging rate and a storage medium to solve the problems in the related art.

[0009] This disclosure provides a method for calculating aging rate, comprising: acquiring a mortality risk prediction model; the mortality risk prediction model describing the mathematical relationship between the actual age and biomarkers of a biological object and a mortality risk function; inputting the actual age and biomarkers of multiple sample biological objects into the mortality risk prediction model to determine the mortality risk function corresponding to each sample biological object, and calculating the sample aging rate based on the mortality risk function of each sample biological object; fitting an age-aging relationship function to characterize the mapping relationship between the actual age and aging rate of a biological object based on the actual age and sample aging rate of each sample biological object; inputting the actual age and biomarkers of a target biological object into the mortality risk prediction model to determine the mortality risk function corresponding to the target biological object, and determining the target aging rate of the target biological object based on the mortality risk function corresponding to the target biological object; and obtaining the target actual age by mapping the target aging rate of the target biological object to the age-aging relationship function, which is used as the biological age of the target biological object.

[0010] In an embodiment of the first aspect, the mortality risk function includes: an exponential function with a mortality time factor to which a shape coefficient is applied as the exponential term, and a baseline risk coefficient applied to the exponential function; the shape coefficient and the baseline risk coefficient are determined by training the mortality risk prediction model respectively; before obtaining the mortality risk prediction model, the method includes: constructing a neural network model based on the mortality risk function; the input of the neural network model is the actual age and biomarkers defined in the mortality risk function, the output of the neural network model is the baseline risk coefficient and the shape coefficient defined in the mortality risk function, and a loss function is constructed based on the baseline risk coefficient, the shape coefficient, and the mortality indicator variable; a training sample set containing the actual age, biomarkers, and mortality indicator values ​​as labels of multiple sample biological objects is input into the neural network model to obtain the output shape coefficient and baseline risk coefficient of each sample biological object; a loss is calculated based on the output shape coefficient, baseline risk coefficient, and label according to the loss function, and the model parameters of the neural network model are updated according to the calculated loss to obtain the mortality risk prediction model.

[0011] In an embodiment of the first aspect, the loss function includes a basic loss term, which is a negative log-likelihood function. The method further includes: for each sample biological object, determining a risk function value and a cumulative risk function value corresponding to the sample biological object based on the mortality risk function corresponding to the sample biological object; determining a mortality indication value corresponding to the sample biological object based on the survival information of the sample biological object; and determining a negative log-likelihood value corresponding to the sample biological object as the basic loss term based on the risk function value, the cumulative risk function value, and the mortality indication value.

[0012] In an embodiment of the first aspect, the loss function further includes: a constraint loss term for constraining the training of the mortality risk prediction model; the method further includes: weighted summation of the basic loss term and the constraint loss term to obtain a composite loss function; wherein the constraint loss term includes: a hybrid regularization term and / or a gradient penalty term, the hybrid regularization term is constructed based on the absolute value and squared terms of the model parameters in the neural network model, and the gradient penalty term is constructed based on the gradient of the basic loss term with respect to the shape coefficient and the baseline risk coefficient; the model parameters include the network weights and network biases of the neural network model.

[0013] In an embodiment of the first aspect, the step of calculating the loss based on the loss function according to the output shape coefficient, baseline risk coefficient and label, and updating the model parameters of the neural network model according to the calculated loss to obtain the mortality risk prediction model, further includes: iteratively training the neural network model with the goal of minimizing the composite loss function to update the shape coefficient and baseline risk coefficient to obtain the mortality risk prediction model.

[0014] In an embodiment of the first aspect, the sample aging rate is determined based on the risk growth rate of the mortality risk function corresponding to the sample biological object at a target time; the target aging rate is determined based on the risk growth rate of the mortality risk function corresponding to the target biological object at a target time.

[0015] In an embodiment of the first aspect, the shape coefficient determination mode includes a population-level mode and an individual-level mode, and the method further includes: in response to executing the individual-level mode, the neural network model includes a plurality of subnetworks for determining the shape coefficient, the subnetworks determining a shape coefficient corresponding to the sample biological object based on the actual age and biomarkers of each sample biological object; in response to executing the population-level mode, the neural network model determines a shape coefficient shared by all the sample biological objects based on the actual age and biomarkers of the sample biological objects.

[0016] In an embodiment of the first aspect, establishing the mapping relationship between the sample aging rate and the actual age includes: averaging the sample aging rates of multiple sample biological objects with the same actual age in each group to obtain the average aging rate, and obtaining the actual age and average aging rate of each sample biological object; and fitting the age-aging relationship function to the actual age and average sample aging rate of each sample biological object.

[0017] In an embodiment of the first aspect, the biomarkers include global-level biomarkers and organ-specific biomarkers corresponding to the target organ.

[0018] A second aspect of this disclosure provides a computer-readable storage medium storing a computer program or instructions which are executed to perform the aging rate calculation method as described in any one of the first aspects.

[0019] As described above, this disclosure provides a method and storage medium for calculating aging rate. The method includes: acquiring a mortality risk prediction model; the mortality risk prediction model describes the mathematical relationship between the actual age and biomarkers of a biological object and a mortality risk function; inputting the actual age and biomarkers of multiple sample biological objects into the mortality risk prediction model to determine the mortality risk function corresponding to each sample biological object, and calculating the sample aging rate based on the mortality risk function of each sample biological object; fitting an age-aging relationship function to characterize the mapping relationship between the actual age and aging rate of the biological object based on the actual age and sample aging rate of each sample biological object; inputting the actual age and biomarkers of a target biological object into the mortality risk prediction model to determine the mortality risk function corresponding to the target biological object, and determining the target aging rate of the target biological object based on the mortality risk function corresponding to the target biological object; and mapping the target aging rate of the target biological object to the age-aging relationship function to obtain the target actual age, which is taken as the biological age of the target biological object. This disclosure obtains a biological age that accurately reflects the degree of aging of the target biological object by constructing a mortality risk prediction model, determining the mortality risk function corresponding to the biological object, and establishing an age-aging relationship function between the actual age and aging rate of the sample biological object. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the aging rate calculation method in one embodiment of this disclosure is shown.

[0021] Figure 2 A flowchart illustrating the aging rate calculation method in one embodiment of this disclosure is shown.

[0022] Figure 3A schematic diagram of a neural network model in one embodiment of this disclosure is shown.

[0023] Figure 4 This diagram illustrates the influence of the strength of the first norm term in the mixed regularization term in one embodiment of the present disclosure.

[0024] Figure 5 This diagram illustrates the influence of the gradient penalty term's strength in one embodiment of the present disclosure.

[0025] Figure 6 A schematic diagram illustrating the age-related aging relationship is shown in one embodiment of this disclosure.

[0026] Figure 7 A schematic diagram illustrating the age-related aging relationship is shown in one embodiment of this disclosure.

[0027] Figure 8 A flowchart illustrating the aging rate calculation method in yet another embodiment of this disclosure is shown.

[0028] Figure 9 A schematic diagram of the aging rate calculation system in one embodiment of this disclosure is shown.

[0029] Figure 10 A schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0030] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed herein. This disclosure can also be implemented or applied through other different specific embodiments, and several details in this disclosure can also be modified or changed according to different viewpoints and application modules without departing from the spirit of this disclosure. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this disclosure can be combined with each other.

[0031] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings, so that those skilled in the art to which this disclosure pertains can readily implement it. This disclosure may be embodied in many different forms and is not limited to the embodiments described herein.

[0032] In this disclosure, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples" indicates that a specific feature, structure, material, or characteristic represented in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. Furthermore, the specific features, structures, materials, or characteristics represented may be combined in any suitable manner in any one or a group of embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples represented in this disclosure, as well as the features of those different embodiments or examples.

[0033] Furthermore, the terms "first" and "second" are used for illustrative 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 disclosure, "a set" means two or more, unless otherwise explicitly specified.

[0034] For the purpose of clarity, devices unrelated to the description are omitted, and the same or similar components throughout the specification are given the same reference numerals.

[0035] Throughout this specification, when it is said that a device is "connected" to another device, this includes not only "direct connection" but also "indirect connection" by placing other components in between. Furthermore, when it is said that a device "comprises" a certain constituent element, unless otherwise stated otherwise, this does not exclude other constituent elements, but rather implies that other constituent elements may be included.

[0036] Although the terms first, second, etc., are used in some examples herein to refer to several elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another. For example, first interface and second interface, etc., are used. Furthermore, as used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms “comprising,” “including,” and “including” indicate the presence of the stated feature, step, operation, element, module, item, kind, and / or group, but do not exclude the presence, occurrence, or addition of one or more other features, steps, operations, elements, modules, items, kinds, and / or groups. The terms “or” and “and / or” as used herein are to be interpreted as inclusive, or mean any one or any combination thereof. Thus, “A, B, or C” or “A, B, and / or C” means “any one of the following: A; B; C; A and B; A and C; B and C; A, B, and C.” An exception to this definition will only occur if the combination of elements, functions, steps, or operations is inherently mutually exclusive in some way.

[0037] The technical terms used herein are for reference only to specific embodiments and are not intended to limit the scope of this disclosure. The singular form used herein includes the plural form unless the statement explicitly indicates otherwise. The word "comprising" as used in this specification means to specify a particular characteristic, region, integer, step, operation, element, and / or component, and does not exclude the presence or addition of other characteristics, regions, integers, steps, operations, elements, and / or components.

[0038] Although not explicitly defined, all terms, including technical and scientific terms used herein, shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. Terms defined in commonly used dictionaries shall be further interpreted as having a meaning consistent with the relevant technical literature and the message of the present disclosure, and shall not be over-interpreted as having an ideal or overly formulaic meaning unless otherwise defined.

[0039] Therefore, this disclosure provides a method for calculating the aging rate, used to determine the biological age of a target biological object based on its actual age and biomarkers. The method determines the mortality risk function corresponding to each sample biological object and the target biological object using a mortality risk prediction model, and further extracts the aging rate based on the mortality risk function. Then, it fits an age-aging relationship function based on the relationship between the actual age and the sample aging rate of multiple sample biological objects, and finally determines the corresponding age in the age-aging relationship function based on the target aging rate of the target biological object, which is then used as the biological age of the target biological object. Thus, while preserving the biological interpretability of the mortality risk function, it utilizes a neural network model to enhance the ability to characterize the complex nonlinear relationship between actual age, biomarkers, and mortality risk, and calculates the aging rate based on the aging rate.

[0040] like Figure 1 The diagram shown illustrates a flowchart of the aging rate calculation method in an embodiment of this disclosure.

[0041] exist Figure 1 The aging rate calculation method includes:

[0042] Step S110: Obtain the mortality risk prediction model.

[0043] The mortality risk prediction model describes the mathematical relationship between the actual age and biomarkers of a biological subject and the mortality risk function.

[0044] In some embodiments, the mortality risk prediction model is constructed based on Gompertz's law, which reveals the law governing the exponential growth trend of human life. Based on this law, the mortality risk prediction model can characterize the mortality risk level of a biological object and its growth trend over time through a mortality risk function. The mortality risk function includes: an exponential function with a death time factor with a shape coefficient applied as the exponential term, and a baseline risk coefficient applied to the exponential function; the shape coefficient and the baseline risk coefficient are determined through training the mortality risk prediction model respectively, and are used to describe the law governing the growth of an individual's mortality risk over time, and can be expressed as:

[0045] (1)

[0046] In equation (1), x represents the input feature, which includes at least the actual age of the biological object and biomarkers. Specifically, the biomarkers include proteomics features. The proteomics features can be Olink proteomics data from the UK Biobank. It is the baseline risk coefficient, used to characterize the risk level of a biological object at the initial moment. The shape coefficient is used to characterize the rate at which an individual's risk of death increases over time.

[0047] In some embodiments, if the target biological object is a target organism, a mortality risk prediction model for the biological object at the overall level is used, wherein the biomarker is an overall level biomarker, which may include one or any combination of 124 plasma proteins retained after screening Olink proteomics data from the UK Biobank.

[0048] In some embodiments, the training process of the mortality risk prediction model includes the following steps.

[0049] like Figure 2 The diagram shown illustrates a flowchart of the aging rate calculation method in an embodiment of this disclosure.

[0050] Step S111: Construct a neural network model based on the mortality risk function; the input of the neural network model is the actual age and biomarkers defined in the mortality risk function, the output of the neural network model is the baseline risk coefficient and shape coefficient defined in the mortality risk function, and a loss function is constructed based on the baseline risk coefficient, shape coefficient and mortality indicator variable.

[0051] The neural network model is used to predict parameters in the mortality risk function based on input features. The input features include at least the actual age and biomarkers of the sample biological object. By feeding the input features into the neural network model, the output is a baseline risk coefficient and a shape coefficient used to construct the mortality risk function. For example, the neural network model generates a two-dimensional parameter vector from the baseline risk coefficient and the shape coefficient. , Output the result.

[0052] In some embodiments, based on Figure 3 The diagram shows a neural network model, which includes an input layer 100, a shared feature extraction layer 200, and a parameter output layer 300.

[0053] Input layer 100 is used to receive the actual age and biomarkers of the sample biological object. Input layer 100 includes several input feature nodes. , , ... , , The number of input features is given, with each input feature node corresponding to either actual age or a biomarker. A shared feature extraction layer 200 is used to extract features and perform non-linear mapping on the actual age and biomarkers to obtain hidden features. The shared feature extraction layer 200 includes at least one hidden layer, and each hidden layer includes several hidden feature nodes. , , ... , , To hide the number of features, Figure 3 All hidden layers shown are fully connected layers. The parameter output layer 300 is used to output the baseline risk coefficient and shape coefficient corresponding to the sample biological object based on the hidden features.

[0054] Therefore, a neural network model can be used to model the complex nonlinear relationship between actual age, multiple biomarkers and baseline risk coefficients, and shape coefficients, and to iterate the model parameters of the neural network model, which include network weights and / or network biases.

[0055] In some embodiments, if the target biological object is a target organism, the neural network model can be configured as a model at the overall level of the biological object. In response to the neural network model being configured as a model at the overall level of the biological object, biomarkers in the input features may include a set of overall-level biomarkers, such as selected and retained plasma proteins. Thus, changes in mortality risk at the overall level of the biological object can be modeled.

[0056] In some embodiments, based on Figure 3 The diagram illustrates a neural network model, which also includes a residual path indicated by dashed lines. Only a portion of the residual path is shown in the diagram. The residual path is used to pass hidden features from previous layers to hidden features from subsequent layers. By using summing nodes, the passed hidden features are added to and fused with the corresponding hidden features from the subsequent layers. This mitigates the information decay problem that may occur as the number of network layers increases and improves the stability of neural network model training.

[0057] In some embodiments, network weights in the neural network model are used to characterize the connection strength between input features and hidden features, as well as between hidden features at different layers. Each biomarker participates in the construction of hidden features through its corresponding network weights, and further influences the determination of the baseline risk coefficient and shape coefficient. The network weights and biases serve as model parameters and are updated based on the loss function during subsequent training. Specifically, the neural network model uses the network weights and biases to perform weighted combinations and nonlinear transformations on the input features, mapping the actual age and effective information from the biomarkers layer by layer into hidden features related to the mortality risk function parameters.

[0058] In some embodiments, based on Figure 3 The diagram illustrates the neural network model. The parameter output layer 300 includes a dual-branch architecture 400, consisting of a baseline risk branch 500 and a shape coefficient branch 600. The baseline risk branch 500 outputs a baseline risk coefficient based on the hidden features, and the shape coefficient branch 600 outputs a shape coefficient based on the hidden features. Corresponding network weights are set in both the baseline risk branch 500 and the shape coefficient branch 600 to perform weighted mapping on the hidden features output from the shared feature extraction layer, thereby determining the baseline risk coefficient and the shape coefficient respectively. Therefore, based on this dual-branch architecture 400, the mortality risk level and risk growth rate can be learned separately based on the shared feature extraction.

[0059] Specifically, the baseline risk coefficient branch 500 includes a dimension reduction layer 510 and a feature weight allocation layer 520. The dimension reduction layer 510 includes a first fully connected layer and a second fully connected layer. For example, the first fully connected layer is connected to the shared feature extraction layer 200 and can be composed of eight hidden neurons. The second fully connected layer is connected to the first fully connected layer and the feature weight allocation layer 520 and can be composed of four hidden neurons. The hidden features output from the shared feature extraction layer are sequentially input into the first and second fully connected layers, enabling the neural network model to extract more compact hidden features from high-dimensional features through layer-by-layer weighted mapping and nonlinear transformation. The feature weight allocation layer 520 is used to further assign feature weights to the hidden features processed by the dimension reduction layer 510 to enhance the features related to the baseline risk coefficient. The baseline risk coefficient output layer 530 is used to output the baseline risk coefficient based on the weighted hidden features.

[0060] In some embodiments, the baseline risk coefficient output layer 530 can constrain the output result to a preset numerical range through an exponential mapping function or other non-negative mapping function to ensure the reasonableness of the baseline risk coefficient value. For example, the range of the baseline risk coefficient can be set to (0,1).

[0061] In some embodiments, the determination mode of the shape coefficient includes a group-level mode and an individual-level mode. Based on Figure 3 A schematic diagram of the neural network model is shown. The shared parameter output structure 610 in the shape coefficient branch 600 is used to execute the group-level mode; the individual parameter output structure 620 in the shape coefficient branch 600 is used to execute the individual-level mode. The group-level mode and the individual-level mode can be determined by external selection.

[0062] If the neural network model is set to the population-level mode, the shared parameter output structure 610 may include a shape coefficient output layer that fits data to the actual age and biomarkers of different sample biological objects and obtains a fixed value of the shape coefficient as a "prior estimate" through, for example, maximum likelihood estimation. Exemplarily, a fixed value of the shape coefficient applicable to all samples can be obtained through maximum likelihood estimation or other parameter estimation methods.

[0063] If the neural network model is set to the individual-level mode, the individual parameter output structure 620 is a sub-network for determining the shape coefficients corresponding to multiple sample biological objects. This sub-network determines the shape coefficient corresponding to each sample biological object based on its actual age and biomarkers. The shape coefficient sub-network may also include a hierarchical structure similar to the baseline risk coefficient branch 500. For example, the shape coefficient sub-network may include one or more fully connected layers to map shared hidden features, a feature weight allocation layer to enhance the feature representation related to the shape coefficient, and a shape coefficient output layer to output the individual shape coefficient corresponding to the current sample biological object.

[0064] Therefore, in the individual-level model, different shape coefficients can be output according to the actual age and biomarkers of different sample biological objects to enhance the ability of the neural network model to express the differences in the rate of increase of mortality risk among individuals; in the population-level model, the shared shape coefficient of the population can be determined based on the actual age and biomarkers of all sample biological objects to use a unified parameter to characterize the rate of increase of mortality risk of the population sample over time.

[0065] Therefore, by switching between the group-level mode and the individual-level mode, a fixed shape coefficient can be obtained through fitting, i.e., the shape coefficient shared by the group; or, a shape coefficient estimated by the trained neural network model can be obtained that reflects the difference in the rate of increase of mortality risk among individuals.

[0066] In some embodiments, the shared shape coefficients of the population can be determined first based on the shared parameter output structure 610, and then the shape coefficients corresponding to each sample biological object can be estimated based on the shared shape coefficients under the individual parameter output structure 620. The shared shape coefficients of the population can be used as prior information for shape coefficient training in the individual-level mode. For example, the shared shape coefficients of the population can be used as the initial values ​​for individual shape coefficient training and participated in the loss calculation to improve training efficiency. In another optional example, the shared shape coefficients of the population and the individual shape coefficients can also be obtained independently, and a weighted sum can be calculated to obtain the "fused" shape coefficients.

[0067] In some embodiments, to enhance the biological rationality of the parameters, an activation mapping module is set at the output of the subnetwork, which can utilize the Sigmoid function. and linear transformation The shape factor is converted to a preset range, where b' is the intermediate value output by the previous connection layer at the output end. This is the lower limit of the preset range. This represents the upper limit of the preset range. For example, when the shape factor needs to be constrained to the interval [0.01, 0.2], the linear transformation formula for the shape factor is: .

[0068] Step S112: Input the training sample set containing the actual age, biomarkers, and mortality indicator values ​​as labels of multiple sample biological objects into the neural network model to obtain the shape coefficient and baseline risk coefficient of each sample biological object as output.

[0069] Among them, the death indicator value, which serves as a label in the sample biological object, is used to characterize the event state of whether a death event has occurred in the sample biological object at the observation time; when the event occurs, the death indicator value can take a first preset value; when the event does not occur, the death indicator value can take a second preset value.

[0070] The cumulative risk function is derived from the time integral of the mortality risk function, which can be found in equation (1). The cumulative risk function characterizes the accumulation of risk from the initial moment to the current moment and can be expressed as:

[0071] (2)

[0072] Therefore, the survival function can be further obtained according to equation (2). The survival function is used to characterize the probability that the target event has not occurred before the current time, and can be expressed as:

[0073] (3)

[0074] In some embodiments, the loss function includes a base loss term, which may be a negative log-likelihood function. The negative log-likelihood function models the mortality indicator value for each sample observation and characterizes how well the current mortality risk function fits the survival outcome of the sample biological object.

[0075] In terms of mortality risk estimation, the observations of a single sample include the probability, represented by the survival function, that the sample has not experienced a mortality event before the observation time, and the probability, represented by the risk function, that the sample experiences a mortality event at the observation time. The probability density function, which is the product of the mortality risk function and the survival function, can be expressed as:

[0076] (4)

[0077] Therefore, when the target event occurs at the observation time, the likelihood function of a single sample can be represented by the probability density function; when the target event does not occur at the observation time, the likelihood function of a single sample can be represented by the survival function. Thus, the likelihood function of a single sample can be expressed as:

[0078] (5)

[0079] Wherein, for sample biological object i, the likelihood function of equation (5) is: The observation time is Its mortality indicator value, which represents survival information, is When a sample experiences a death event during the observation period, then When the target event does not occur in the sample within the observation period, .

[0080] Based on equation (5), taking the logarithm and negative of the likelihood function for a single sample yields the negative log-likelihood function for that single sample:

[0081] (6)

[0082] in, Let be the negative log-likelihood value of sample i. Therefore, in the basic loss term, one part comes from the cumulative risk function value H(t_i|x_i), corresponding to the cumulative risk term of the sample before the current observation time, and the other part comes from the risk term of the sample at the time the death event occurs.

[0083] Optionally, in some embodiments, the negative log-likelihood values ​​of a subset of samples can be summed or averaged to obtain the negative log-likelihood function of the entire sample, which serves as the base loss term. For example, the negative log-likelihood function NLL can be expressed as:

[0084] (7)

[0085] Therefore, by using equation (7), the mortality indicator value of the sample biological object can be optimized to establish an association with the shape coefficient and baseline risk coefficient output by the neural network model. In the subsequent iterative training process, the neural network model is trained to minimize the negative log-likelihood function to train the shape coefficient and baseline risk coefficient, so that the neural network model is updated in the direction of more accurately representing the mortality risk pattern.

[0086] In some embodiments, the loss function further includes a constraint loss term for constraining the training of the mortality risk prediction model. As an example, the constraint loss term includes a hybrid regularization term.

[0087] Specifically, the hybrid regularization term is constructed based on the absolute value and squared terms of the model parameters in the neural network model, and the hybrid regularization term can be expressed as:

[0088] (8)

[0089] in, This represents the set of model parameters in the neural network model. The model parameters include the network weights and network biases trained by the neural network model in the aforementioned embodiments. This represents the sum of squares of all parameters in the model parameters, i.e., the square of the second norm, used to constrain the magnitude of the model parameters from becoming too large, in order to prevent the neural network model from overfitting. The first norm represents the sum of the absolute values ​​of all parameters in the model parameters. By introducing this first norm term into the loss function, some model parameters can be reduced to near zero during training, thereby reducing the impact of irrelevant or weakly correlated features on the neural network model. It is a balance coefficient with a value range of [0,1]. It can balance the smooth stability of the second norm squared term with the sparse selection ability of the first norm term.

[0090] In some embodiments, the constraint loss term includes a gradient penalty term. Exemplarily, the gradient penalty term can be constructed based on the norm of the gradient of the model output with respect to the base loss term, so that the gradient remains within a reasonable scale. The gradient penalty term can be expressed as:

[0091] (9)

[0092] Where L represents the basic loss term, and when the negative log-likelihood function is used as the basic loss term, L is NLL in equation (7). This represents the model output value, which includes the shape coefficient and baseline risk coefficient obtained in this training round. E represents the expected value, which can be approximated by calculating the mean. The sensitivity of the gradient penalty term to the output indirectly improves the model parameter update process.

[0093] In some embodiments, the basic loss term and the constraint loss term are weighted and summed to obtain a composite loss function.

[0094] (10)

[0095] in, This is the total loss function obtained from this neural network training. The weights of the mixed regularization term are relative to the loss terms other than the mixed regularization term, and can change the strength of the mixed regularization term's influence. The gradient penalty term weight is the relative weight of the loss term other than the gradient penalty term, and it can change the strength of the gradient penalty term.

[0096] In some embodiments, the first-order norm term and the second-order norm squared term in the mixed regularization term can be weighted and combined according to a preset ratio; the combination of the basic loss term, the mixed regularization term, and the gradient penalty term in the composite loss function can also be set according to training requirements.

[0097] In some embodiments, based on Figure 4The diagram illustrates the influence of the first-order norm term in the mixed regularization terms. The horizontal axis, "Weight Value (w)," represents the value of the network weight w in the model parameters; the vertical axis, "Total Loss," represents the total loss function value. By weighted summing of the basic loss term and the constraint loss term, the curves showing the change of the total loss function under different regularization term weights can be obtained.

[0098] Specifically, the curve corresponding to "No Regularization" in the figure represents the relationship between the total loss function and network weights without mixed regularization. The other curves represent the relationship between the total loss function and network weights under different mixed regularization weights. The magnitude of the network weights is positively correlated with the value of the total loss function. Without the introduction of mixed regularization, the total loss function exerts the least constraint on the magnitude of the network weights. As the parameters increase, the influence of the first-order norm term in the mixed regularization term on the total loss function gradually strengthens. When the magnitude of the network weights is large, the first-order norm term's effect on the total loss function becomes more significant, thus prompting the network weights to converge towards smaller magnitudes. Therefore, this can suppress the increase in model complexity caused by excessively large network weights and reduce the risk of overfitting the neural network model to the training samples.

[0099] In some embodiments, based on Figure 5 The diagram shows the influence of the gradient penalty term. The horizontal axis "GradientNorm" represents the gradient norm of the basic loss term with respect to the model output, i.e., in equation (10). The vertical axis “GradientPenalty” represents the value of the gradient penalty term. The three curves in the figure represent the values ​​of the gradient penalty term under different gradient penalty weights, i.e., in equation (10). Below is the correspondence between the gradient norm and the gradient penalty term. A straight line, Target Norm = 1, indicates that the target gradient norm is 1. When the gradient norm is close to the target gradient norm of 1, the gradient penalty term has a small value; when the gradient norm deviates from the target gradient norm of 1, the gradient penalty term gradually increases, and the greater the deviation, the more significant the increase in the penalty value. Therefore, by adjusting the gradient penalty weight, the constraint strength of the gradient corresponding to the model output can be controlled, keeping the gradient within a reasonable range. This improves the stability of the neural network model training process and reduces the adverse effects of excessively large or small gradients on model parameter updates.

[0100] Step S113: Calculate the loss based on the output shape coefficient, baseline risk coefficient and label according to the loss function, and update the model parameters of the neural network model according to the calculated loss to obtain the mortality risk prediction model.

[0101] In some embodiments, the neural network model is iteratively trained with the goal of minimizing the composite loss function in the above embodiments to update the network weights and biases, thereby obtaining the mortality risk prediction model. Specifically, in each iteration, training samples are input into the neural network model, and the baseline risk coefficient and the shape coefficient are calculated through forward propagation. Subsequently, the basic loss term is calculated according to equation (7), the regularization constraint term is calculated according to equation (8), the gradient penalty term is calculated according to equation (9), and the loss is obtained by summing based on the total loss function. The gradient with respect to each model parameter in the neural network model is calculated based on the loss, and the model parameters of the neural network model are updated. The above iterative process is repeated to train the training set of the sample biological objects for multiple rounds until the preset training stopping condition is met. The training stopping condition includes at least one of the following: the total loss function value converges to a preset threshold, the preset maximum number of iterations is reached, and the performance index of the neural network model (e.g., the C-index of the neural network model) no longer increases. The neural network model with the final determined shape coefficient and baseline risk coefficient calculation capability obtained after training is determined as the mortality risk prediction model.

[0102] Therefore, compared to existing technologies, the baseline risk parameters of traditional Gompertz models have only a fixed form, making it difficult to adapt to the complex nonlinear relationships in real-world data. This disclosure combines deep neural networks with Gompertz models to address this issue. Designed as a nonlinear function dynamically estimated by a neural network, this model can flexibly fit the nonlinear association between covariates such as age and biomarkers and risk. Specifically, in simulations on nonlinear datasets, the model's concordance index is 0.664, higher than the 0.513 concordance index of the traditional Gompertz model and the 0.487 concordance index of the Cox model. The neural network, through a combination of multi-layer linear mappings and nonlinear activation functions (such as the sigmoid function and exponential activation functions), can model the multi-order feature combination relationships and nonlinear interactions between actual age, biomarkers, and the mortality risk function, thus overcoming the limitation of traditional models' limited ability to characterize the relationship between input features and the mortality risk function.

[0103] Therefore, the shape coefficient and baseline risk coefficient obtained by this model more accurately reflect the survival information of the sample biological objects, have a stronger ability to characterize nonlinear relationships, and thus help improve the accuracy of mortality risk prediction.

[0104] Step S120: Input the actual age and biomarkers of multiple sample biological objects into the mortality risk prediction model, determine the mortality risk function corresponding to each sample biological object, and calculate the sample aging rate based on the mortality risk function of each sample biological object.

[0105] The sample biological objects are historical samples used to fit the relationship between aging rate and age. They can be the training set, validation set, or other sample biological objects in the sample library of the neural network model. For each sample biological object, after its actual age and biomarkers are input into the trained mortality risk prediction model, the baseline risk coefficient and shape coefficient corresponding to that sample individual can be determined, and the mortality risk function for that sample individual can be constructed accordingly.

[0106] In some embodiments, the rate of increase in mortality risk of the mortality risk function at the initial time of t=0 for each sample is defined as the individual rate of aging (ROA). The individual rate of aging, based on the mortality risk function of equation (1), can be expressed as:

[0107] (11)

[0108] Therefore, based on the mortality risk prediction model, the baseline risk coefficient and shape coefficient corresponding to each sample biological object are obtained, and these are converted into a sample aging rate index with clear biological significance. Thus, the resulting sample aging rate index is also a composite index that combines the baseline risk coefficient, representing the individual's baseline vulnerability, with the shape coefficient, representing the individual / group's risk growth rate.

[0109] Step S130: Fit an age-aging relationship function to characterize the mapping relationship between the actual age and aging rate of each sample biological object based on the actual age and aging rate of each sample biological object.

[0110] In some embodiments, the average aging rate is obtained by averaging the aging rates of multiple sample biological objects of the same actual age in each group, thus obtaining the actual age and average aging rate of each sample biological object; the age-aging relationship function is obtained by fitting the actual age and average aging rate of each sample biological object. For example, the average aging rate of the aging rates calculated for all 30-year-old sample biological objects in the sample is taken as the average aging rate of the 30-year-old sample biological objects.

[0111] based on Figure 6The diagram illustrating the age-related aging relationship shows that the horizontal axis "Age" represents the actual age of the sample organisms, and the vertical axis "log(ROA)" represents the logarithmic value of the average aging rate of the sample organisms. The age-related aging relationship function for the entire sample organisms can be established using a non-linear approach. Specifically, the non-linear function g(age) can be obtained by fitting the logarithmic value of the average aging rate corresponding to different ages using a spline function. Using the logarithmic average aging rate for fitting helps reduce the numerical differences in the average aging rate across different ages and improves the stability of the fitting.

[0112] In some embodiments, based on Figure 7 The diagram illustrating the relationship between age and aging, with the coordinate system and... Figure 6 Similarly, if the actual age and log(ROA) satisfy an approximately linear relationship, the age-related aging function can also be established linearly, and can be expressed as:

[0113] (12)

[0114] Among them, a linear function relating age and aging is obtained by fitting the punctuation points of the sample biological objects in the coordinate system, and the slope of the linear function is... The intercept is The slope and intercept are approximate values ​​obtained from the fitting, and age is the actual age of the sample biological object.

[0115] Step S140: Input the actual age and biomarkers of the target biological object into the mortality risk prediction model, determine the mortality risk function corresponding to the target biological object, and determine the target aging rate of the target biological object based on the mortality risk function corresponding to the target biological object.

[0116] The target biological object refers to the individual whose aging rate is to be calculated. Similar to the processing of the sample biological object, after inputting the actual age and biomarkers of the target biological object into the mortality risk prediction model, the baseline risk coefficient and shape coefficient corresponding to the target individual can be obtained, and the corresponding mortality risk function and target aging rate can be further constructed.

[0117] In some embodiments, the target aging rate is the rate of risk growth of the mortality risk function corresponding to the target biological object at a target time. For example, when the target time is the starting time t=0, the mortality risk function corresponding to the target biological object can be differentiated with respect to time to obtain the risk growth rate at the starting time, and this risk growth rate is determined as the target aging rate ROA. Therefore, the target aging rate is a composite indicator determined by both the baseline risk coefficient and the shape coefficient, which can simultaneously reflect the baseline risk level of the target biological object at the starting time and the growth trend of mortality risk over time.

[0118] In some embodiments, the target aging rate can be logarithmically transformed to obtain the target biological object. This allows for matching with the age-related aging function based on log(ROA) established in step S130.

[0119] In some embodiments, if the target biological object exhibits a higher aging rate compared to a sample biological population of the same age, it indicates that the target individual is in a state of accelerated aging at the physiological level; if the target aging rate is low, it indicates that the target biological object exhibits a relatively slower aging rate compared to a sample biological population of the same age.

[0120] Step S150: Based on the target aging rate of the target biological object, determine the age corresponding to the target aging rate in the age-aging relationship function, and use it as the biological age of the target biological object.

[0121] Therefore, the dynamic aging rate of a target biological object can be mapped onto an age scale to obtain a biological age with intuitive interpretation. Compared with using only actual age or only static biomarker scores, this approach simultaneously considers the mortality risk function, the influence of biomarkers, and the aging rate, thus more accurately reflecting the individual's true physiological state.

[0122] In some embodiments, based on Figure 6 The diagram illustrating the age-related aging relationship uses a non-linear approach to establish the age-related aging relationship function, which can be further used to derive the inverse function f(log(ROA)) = g⁻¹(age). When the target biological object is determined... Then, by substituting the inverse function, the age value corresponding to the aging rate can be looked up, and this age value can be used as the biological age of the target biological object.

[0123] In some embodiments, based on Figure 7 The diagram illustrating the age-aging relationship uses a linear approach to establish the age-aging relationship function. After obtaining the aging rate of the target biological object using the prediction results of the mortality risk model, the biological age calculation formula is obtained based on the inverse function determined by equation (12). That is, the "actual age" obtained by mapping the aging rate of the target biological object through the inverse function is taken as the biological age, which can be expressed as:

[0124] (13)

[0125] like Figure 8 The diagram shown illustrates a flowchart of an aging rate calculation method according to another embodiment of this disclosure. Compared to previous embodiments, the target biological object in this embodiment is a target biological organ.

[0126] Step S210: Obtain an organ-specific mortality risk prediction model; the organ-specific mortality risk prediction model describes the mathematical relationship between the actual age of the target biological organ and the corresponding organ-specific biomarkers and the mortality risk function.

[0127] The organ-specific mortality risk prediction model can be considered an extension of the mortality risk prediction model in the aforementioned embodiments at the organ level, and the actual age of the target biological organ refers to the actual age of the biological individual to which the target biological organ belongs.

[0128] In some embodiments, organ-specific biomarkers are used to characterize the proteomics features corresponding to different organs. For example, candidate organ-specific proteins corresponding to multiple organs can be obtained by first screening candidate proteins based on preset organ-specific protein screening rules; then, feature screening is performed on the candidate organ-specific proteins for each organ to obtain a set of organ-specific biomarkers corresponding to each organ. For example, the organ-specific proteins may include 1131 organ-specific proteins covering 16 organs.

[0129] In some embodiments, the preset organ-specific protein screening rules include: determining the organ specificity of proteins by combining large-scale proteome maps with existing RNA and protein maps. Specifically, the specificity of proteins in target organs can be measured using the Global Label Score (GLS), and proteins with a GLS score greater than or equal to a preset threshold are identified as candidate organ-specific proteins. Furthermore, alternative screening rules can be used, i.e., when the expression level of a certain protein in a certain organ is higher than a preset fold or greater in any other organ, that protein is also identified as a candidate organ-specific protein.

[0130] Step S220: Input the actual age of multiple sample biological organs and the organ-specific biomarkers corresponding to the multiple sample biological organs into the organ-specific mortality risk prediction model, determine the mortality risk function corresponding to each sample biological organ, and calculate the sample aging rate based on the mortality risk function of each sample biological organ.

[0131] Step S230: Fit an organ age aging relationship function to characterize the mapping relationship between the actual age and aging rate of the target biological organ based on the actual age and aging rate of each sample biological organ.

[0132] Step S240: Input the actual age of the target biological organ and organ-specific biomarkers into the organ-specific mortality risk prediction model, determine the mortality risk function corresponding to the target biological organ, and determine the target aging rate of the target biological organ based on the mortality risk function corresponding to the target biological organ.

[0133] Step S250: Determine the corresponding age in the organ age-aging relationship function according to the target aging rate of the target biological organ, and use it as the biological age of the target biological organ.

[0134] In this embodiment, steps S210-S250 correspond sequentially to steps S110-S150, and are used to calculate the aging rate of the target biological organ. The biomarker is an organ-specific biomarker corresponding to the target biological organ, and the biological age of the target biological object is the biological age of the target biological organ. Therefore, while determining the overall biological age of the target biological object, the biological age at the target organ level can be further obtained to characterize the aging of the target biological organ.

[0135] like Figure 9 The diagram shows a schematic of the aging rate calculation system in one embodiment of this disclosure. It should be noted that the principle and technical implementation of the aging rate calculation system can refer to the aging rate calculation method in previous embodiments, therefore, it will not be repeated in this embodiment.

[0136] The aging rate calculation system 700 includes:

[0137] The acquisition module 701 is used to acquire a mortality risk prediction model; the mortality risk prediction model describes the mathematical relationship between the actual age and biomarkers of a biological object and the mortality risk function.

[0138] The aging rate determination module 702 is used to input the actual age and biomarkers of multiple sample biological objects into the mortality risk prediction model, determine the mortality risk function corresponding to each sample biological object, and calculate the sample aging rate based on the mortality risk function of each sample biological object.

[0139] The age aging relationship function fitting module 703 is used to fit an age aging relationship function that characterizes the mapping relationship between the actual age and aging rate of the biological object based on the actual age and aging rate of each sample biological object.

[0140] The aging rate determination module 702 is further configured to input the actual age and biomarkers of the target biological object into the mortality risk prediction model, determine the mortality risk function corresponding to the target biological object, and determine the target aging rate of the target biological object based on the mortality risk function corresponding to the target biological object.

[0141] The biological age determination module 704 is used to obtain the target actual age based on the target aging rate of the target biological object through the mapping of the age aging relationship function, and use it as the biological age of the target biological object.

[0142] It should be noted that, in Figure 9 The various functional modules in the embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any combination thereof. When implemented in software, they can be implemented, in whole or in part, in the form of a computer program or instruction product. A computer program or instruction product includes one or more computer programs or instructions. When a computer program or instruction is loaded and executed on a computer, it produces, in whole or in part, the flow or function according to this disclosure. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0143] and, Figure 9 The apparatus disclosed in the embodiments can be implemented through other modular division methods. The apparatus embodiments shown above are merely illustrative. For example, the module division is only a logical functional division, and in actual implementation, there may be other division methods. For example, a group of modules or modules may be combined or dynamically integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces, and the indirect coupling or communication connection between devices or modules may be electrical or other forms.

[0144] in addition, Figure 9 The multiple functional modules and sub-modules in the embodiments can be integrated into one processing unit, or several modules can exist physically separately, or two or more modules can be dynamically integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module. If the integrated unit 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.

[0145] It should be specifically noted that the flowchart representations of the embodiments described above in this disclosure can be understood as representing a module, segment, or portion of code comprising one or more executable instructions configured to implement a specific logical function or process. Furthermore, the scope of the preferred embodiments of this disclosure includes other implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved.

[0146] For example, Figure 1 , Figure 2 , Figure 8 The order of several steps in the method embodiment may vary in specific scenarios and is not limited to the above representation.

[0147] like Figure 10 The diagram shown illustrates the structure of an electronic device according to an embodiment of this disclosure.

[0148] The electronic device 800 may be exemplified as a processing terminal, such as a server, desktop computer, laptop computer, tablet computer, or other terminal.

[0149] The electronic device 800 includes a bus 801, a processor 802, and a memory 803. The processor 802 and the memory 803 can communicate via the bus 801. The memory 803 can store computer programs or instructions. The processor 802 implements the method flow or function described in the previous embodiments by running the computer program or instructions stored in the memory 803, for example... Figure 1 , Figure 2 , Figure 8 .

[0150] Bus 801 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, although only one thick line is used in the diagram, this does not indicate that there is only one bus or one type of bus.

[0151] In some embodiments, processor 802 may be implemented as a central processing unit (CPU), microprocessor unit (MCU), system on chip (System on Chip), or field-programmable array (FPGA). Memory 803 may include volatile memory for temporary data storage during program execution, such as random access memory (RAM).

[0152] The memory 803 may also include non-volatile memory for data storage, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state disk (SSD).

[0153] In some embodiments, the electronic device 800 may further include a communicator 804. The communicator 804 is used for communication with external devices. In specific examples, the communicator 804 may include one or more wired and / or wireless communication circuit modules. For example, the communicator 804 may include one or more of, such as a wired network card, a USB module, a serial interface module, etc. The wireless communication protocols followed by the wireless communication module include, for example, Nearfield Communication (NFC) technology, Infrared (IR) technology, Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Time-Division Code Division Multiple Access (TD-SCDMA), Long Term Evolution (LTE), Bluetooth (BT), Global Navigation Satellite System (GNSS), etc.

[0154] This disclosure also provides a computer-readable storage medium storing a computer program or instructions, which, when run, implement the method flow or function of any of the previous embodiments.

[0155] That is, the method steps in the above embodiments are implemented as software or computer code that can be stored in a recording medium (such as CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or implemented as computer code that is originally stored in a remote recording medium or a non-transitory machine-readable medium and will be stored in a local recording medium after being downloaded via a network, so that the method represented herein can be stored in such software processing on a recording medium using a general-purpose computer, a special processor or programmable or special hardware (such as ASIC or FPGA).

[0156] This disclosure may also provide a computer program product, comprising one or more computer programs or instructions, which, when run, perform all or part of the processes or functions described in this disclosure. The computer program product includes one or more computer programs or instructions.

[0157] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0158] In summary, this disclosure provides a method and storage medium for calculating aging rate. The method includes: acquiring a mortality risk prediction model; the mortality risk prediction model describing the mathematical relationship between the actual age and biomarkers of a biological object and a mortality risk function; inputting the actual age and biomarkers of multiple sample biological objects into the mortality risk prediction model to determine the mortality risk function corresponding to each sample biological object, and calculating the sample aging rate based on the mortality risk function of each sample biological object; fitting an age-aging relationship function to characterize the mapping relationship between the actual age and aging rate of each sample biological object based on the actual age and sample aging rate of each sample biological object; inputting the actual age and biomarkers of a target biological object into the mortality risk prediction model to determine the mortality risk function corresponding to the target biological object, and determining the target aging rate of the target biological object based on the mortality risk function corresponding to the target biological object; and obtaining the target actual age by mapping the target aging rate of the target biological object to the age-aging relationship function, which is used as the biological age of the target biological object. This disclosure, through the mortality risk function determined by the mortality risk prediction model, can accurately determine the aging rate to accurately reflect the degree of aging of a target individual.

[0159] The above embodiments are merely illustrative of the principles and effects of this disclosure and are not intended to limit this disclosure. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this disclosure. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this disclosure should still be covered by the protection scope of this disclosure.

Claims

1. A method for calculating aging rate, characterized in that, include: Obtain a mortality risk prediction model; The mortality risk prediction model is a model based on a neural network used to describe the mathematical relationship between the actual age and biomarkers of a biological object and the mortality risk function. The mortality risk function includes an exponential function with a death time factor with a shape coefficient applied as the exponential term and a baseline risk coefficient applied to the exponential function. The input of the neural network includes the actual age and the biomarkers, and the output includes the baseline risk coefficient and the shape coefficient. The actual age and biomarkers of multiple sample biological objects are input into the mortality risk prediction model to determine the mortality risk function corresponding to each sample biological object, and the sample aging rate of each sample biological object is determined based on the risk growth rate of the mortality risk function of each sample biological object at the target time. Based on the actual age and aging rate of each sample biological object, an age-aging relationship function is fitted to characterize the mapping relationship between the actual age and aging rate of the biological object. The actual age and biomarkers of the target biological object are input into the mortality risk prediction model to determine the mortality risk function corresponding to the target biological object, and the target aging rate of the target biological object is determined based on the risk growth rate of the mortality risk function corresponding to the target biological object at the target time. The target actual age is obtained by mapping the target aging rate of the target biological object to the age-aging relationship function, and is taken as the biological age of the target biological object.

2. The aging rate calculation method according to claim 1, characterized in that, The shape coefficient and the baseline risk coefficient are determined by training the mortality risk prediction model, respectively. Before obtaining the mortality risk prediction model, the following steps are included: A neural network model is constructed based on the mortality risk function; the input of the neural network model is the actual age and biomarkers defined in the mortality risk function, and the output of the neural network model is the baseline risk coefficient and shape coefficient defined in the mortality risk function. A loss function is constructed based on the baseline risk coefficient, shape coefficient and mortality indicator variable. The training sample set, which includes the actual age, biomarkers, and mortality indication values ​​as labels of multiple sample biological objects, is input into the neural network model to obtain the shape coefficient and baseline risk coefficient of each sample biological object as output. The loss is calculated based on the output shape coefficient, baseline risk coefficient, and label using the loss function, and the model parameters of the neural network model are updated based on the calculated loss to obtain the mortality risk prediction model.

3. The aging rate calculation method according to claim 2, characterized in that, The loss function includes a basic loss term, which is a negative log-likelihood function. The method further includes: For each sample biological object, the risk function value and cumulative risk function value corresponding to the sample biological object are determined based on the mortality risk function corresponding to the sample biological object. Determine the mortality indicator value corresponding to the sample biological object based on the survival information of the sample biological object; Based on the risk function value, the cumulative risk function value, and the mortality indicator value, the negative log-likelihood value corresponding to the sample biological object is determined as the basic loss term.

4. The aging rate calculation method according to claim 2, characterized in that, The loss function further includes a constraint loss term used to constrain the training of the mortality risk prediction model, and the method further includes: The composite loss function is obtained by weighted summation of the basic loss term and the constraint loss term. The constraint loss term includes a hybrid regularization term and / or a gradient penalty term. The hybrid regularization term is constructed based on the absolute value and squared terms of the model parameters in the neural network model. The gradient penalty term is constructed based on the gradient of the shape coefficient and the baseline risk coefficient of the basic loss term. The model parameters include the network weights and network biases of the neural network model.

5. The aging rate calculation method according to claim 4, characterized in that, The method of calculating the loss based on the output shape coefficient, baseline risk coefficient, and label using the loss function, and updating the model parameters of the neural network model based on the calculated loss to obtain the mortality risk prediction model, further includes: The neural network model is iteratively trained with the goal of minimizing the composite loss function to update the shape coefficient and the baseline risk coefficient, thereby obtaining the mortality risk prediction model.

6. The aging rate calculation method according to claim 1, characterized in that, The aging rate of the sample is determined based on the risk growth rate of the mortality risk function corresponding to the biological object of the sample at the target time. The target aging rate is determined based on the risk growth rate of the mortality risk function corresponding to the target biological object at the target time.

7. The aging rate calculation method according to claim 2, characterized in that, The method for determining the shape coefficient includes a group-level model and an individual-level model, and the method further includes: In response to executing the individual-level pattern, the neural network model includes multiple subnetworks for determining the shape coefficients, each subnetwork determining a shape coefficient corresponding to the sample biological object based on the actual age and biomarkers of each sample biological object. In response to executing the population-level pattern, the neural network model determines a shape coefficient shared by all the sample biological objects based on the actual age and biomarkers of the sample biological objects.

8. The aging rate calculation method according to claim 1, characterized in that, The process of establishing a mapping relationship between the aging rate of the sample and the actual age includes: The average aging rate of multiple biological subjects with the same actual age is calculated to obtain the average aging rate corresponding to that actual age. The age-aging relationship function is obtained by fitting the data based on the actual age and its corresponding average aging rate.

9. The aging rate calculation method according to claim 1, characterized in that, The target biological object is a target organism or a target biological organ, and also includes: If the target organism is a target organism, the biomarkers include global-level biomarkers of the target organism, and the biological age is the biological age of the target organism; or, If the target biological object is a target biological organ, the biomarker includes organ-specific biomarkers corresponding to the target biological organ, and the biological age of the target biological object is the biological age of the target biological organ.

10. An aging rate calculation system, characterized in that, include: The acquisition module is used to acquire the mortality risk prediction model; the mortality risk prediction model is a model based on neural networks that describes the mathematical relationship between the actual age and biomarkers of a biological object and the mortality risk function. The aging rate determination module is used to input the actual age and biomarkers of multiple sample biological objects into the mortality risk prediction model, determine the mortality risk function corresponding to each sample biological object, and determine the sample aging rate of each sample biological object based on the mortality risk function of each sample biological object. And for inputting the actual age and biomarkers of the target biological object into the mortality risk prediction model, determining the mortality risk function corresponding to the target biological object, and determining the target aging rate of the target biological object based on the mortality risk function corresponding to the target biological object; The age-aging relationship function fitting module is used to fit an age-aging relationship function that characterizes the mapping relationship between actual age and aging rate based on the actual age and aging rate of each sample biological object. The biological age determination module is used to map the corresponding age value in the age-aging relationship function according to the target aging rate of the target biological object, and use it as the biological age of the target biological object.

11. An electronic device, characterized in that, include: Memory is used to store computer programs or instructions; A processor for executing a computer program or instructions stored in the memory to implement the aging rate calculation method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, include: A computer program or instructions for performing the aging rate calculation method as described in any one of claims 1 to 9.

Images