Prediction method for survival state of duck against hepatitis A virus infection

Abstract

The invention discloses a method for predicting the survival state of duck against hepatitis A virus infection. The method comprises the following steps: (1) constructing a PCA (Principal Component Analysis) prediction model by taking the contents of IFN-alpha and IFN-beta of duck blood as variables; (2) measuring the contents of IFN-alpha and IFN-beta in the duck blood to be measured; and (3) substituting the measured contents of IFN-alpha and IFN-beta in the duck blood into the prediction model, and predicting the anti-hepatitis A virus infection survival state of the duck to be detected according to the family background of the individual in combination with the output result of the prediction model. According to the method, ROC curves and corresponding AUC values of different prediction models in the aspect of predicting the survival state are compared, and the result shows that the AUC value of the PCA prediction model is the highest, and the sensitivity and the specificity are most prominent. Based on the prediction model provided by the invention, the Beijing duck individuals or strains with strong disease resistance can be rapidly screened, and the prediction model has an application prospect in duck disease resistance breeding or duck virus hepatitis prevention.

Description

Technical Field

[0001] This invention relates to a method for predicting the survival status of poultry against viral infection, and more particularly to a method for predicting the survival status of ducks against hepatitis A virus infection, belonging to the field of prediction or assessment of duck survival status against viral infection. Background Technology

[0002] Duck meat is an important source of animal protein for the Chinese population. However, diseases have limited the development of the duck meat industry. Among these diseases, duck viral hepatitis is a significant acute and highly contagious disease. In China, duck hepatitis A virus type 3 (DHAV-3) is the main pathogen causing duck viral hepatitis.

[0003] Currently, there are no commercially available DHAV-3 vaccines, posing significant challenges to disease control. Screening for DHAV-3 resistant Beijing ducks primarily relies on post-infection observation and statistics. This method cannot predict the flock's disease resistance before infection, requires multiple generations of selection, is time-consuming, and struggles to dynamically predict individual disease resistance. Furthermore, traditional survival analysis methods often ignore the influence of individual heterogeneity and familial aggregation, limiting the accuracy of predictions. Rapid, low-cost, and high-precision disease resistance prediction methods are needed to support precision breeding and disease control. Type I interferon, as a key immunomodulatory molecule, plays a crucial role in resisting viral infections; its content and activity are correlated with an animal's resistance to viruses. However, current technologies lack clear and mature methods for effectively utilizing this characteristic of type I interferon for disease-resistant breeding, especially for DHAV-3 resistance in Beijing ducks. Summary of the Invention

[0004] The main objective of this invention is to provide a rapid and accurate method for predicting the survival status of ducks resistant to hepatitis A virus infection.

[0005] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0006] A method for predicting the survival status of ducks against hepatitis A virus infection for non-diagnostic or therapeutic purposes, comprising:

[0007] (1) A predictive model for the survival status of ducks against hepatitis A virus infection was constructed using the levels of IFN-α and IFN-β in duck blood as variables.

[0008] (2) Determine the levels of IFN-α and IFN-β in the blood of the ducks to be tested;

[0009] (3) Substitute the IFN-α and IFN-β levels in the blood of the duck to be tested as determined in step (2) into the prediction model constructed in step (1), and predict the survival status of the duck against hepatitis A virus infection based on the individual's family background and the output of the prediction model.

[0010] In a preferred embodiment of the present invention, the method for constructing the prediction model in step (1) includes: using the contents of IFN-α and IFN-β in duck blood as variables, constructing a prediction model of the survival status of ducks against hepatitis A virus infection based on principal component analysis (PCA) combined with machine learning algorithms; wherein, the machine learning algorithm can be a classification or regression model.

[0011] Further preferredly, the dimensionality reduction processing of duck blood IFN-α and IFN-β content is performed in the construction of the prediction model.

[0012] In a preferred embodiment of the present invention, the formula of the prediction model in step (1) is as follows: survival_status~pca_score+(1|family_id);

[0013] In a preferred embodiment of the present invention, the present invention can use methods such as ELISA to detect the content of IFN-α or IFN-β in duck blood; the duck blood includes duck plasma and duck serum;

[0014] The duck plasma can be prepared using any of the following methods (1)-(3):

[0015] (1) Plasma obtained by centrifuging duck blood with heparin sodium anticoagulation;

[0016] (2) Plasma obtained by centrifuging duck blood with EDTA anticoagulation;

[0017] (3) Plasma obtained by anticoagulating duck blood using other methods and centrifuging.

[0018] The duck serum can be prepared by any one of the following methods (1) or (2):

[0019] (1) Serum obtained by centrifuging duck blood after it has been left to stand;

[0020] (2) Serum obtained from duck blood using other methods.

[0021] In a preferred embodiment of the present invention, step (3) of predicting the survival status of the ducks against hepatitis A virus infection by combining the individual's family background with the output of the prediction model includes:

[0022] 1. Retrieve the family_id (such as S4A, Z3B, etc.) of the ducks to be tested from the database (which records the family information of the ducks to be tested and the dataset of IFN-α and IFN-β), match the original IFN-α and IFN-β data of all individuals in the family, and calculate the basic IFN level of the family.

[0023] 2. Detection and standardization of IFN indicators in ducks under test: The actual values ​​of IFN-α (denoted as Xα) and IFN-β (denoted as Xβ) of the ducks under test were detected and standardized to obtain the standardized values ​​of IFN-α or IFN-β: Standardized value = (measured value - mean of the whole dataset) / standard deviation of the whole dataset;

[0024] 3. Prediction model input and probability output: Input the standardized values ​​of IFN-α and IFN-β and the base level of IFN for the family (family_id as a random effect) into the prediction model;

[0025] 4. Prediction criteria and result determination based on family background:

[0026] Basic decision threshold: Based on the optimal threshold of the ROC curve of the entire dataset (derived from roc_curve analysis), assuming the optimal threshold P = 0.55, that is, P ≥ 0.55 predicts survival, and P < 0.55 predicts death.

[0027] Family background correction:

[0028] If the mortality rate of the duck family being tested (mortality = number of dead individuals in the family / total number of individuals in the family) is less than 50% (e.g., the mortality rate of the S4A family = 3 / 8 = 37.5% < 50%), then no correction is required;

[0029] If a family's mortality rate is ≥50% (e.g., 6 / 10 = 60%), the threshold for judgment will be lowered by 10% (i.e., P ≥ 0.495 predicts Alive).

[0030] If the average IFN-α or average IFN-β of the duck family to be tested is more than 20% lower than the mean of the whole dataset (e.g., the mean_ifn_alpha of the whole dataset is 65.23, and the mean_ifn_alpha of a certain family is 52.18, which is less than 20%), then the judgment threshold will be raised by 5% (i.e., P≥0.5775 is predicted as Alive).

[0031] Final output: Combining the base threshold and the family-adjusted threshold, output "Predicted as Alive (probability XX%)" or "Predicted as Dead (probability XX%)", and indicate the impact of family background on the result (e.g., "Due to the high family mortality rate, the judgment threshold is lowered, and the confidence of predicting as Alive is increased").

[0032] The duck used in this invention is preferably the Beijing duck.

[0033] This invention first analyzed the differences in IFN-α and IFN-β levels between ducks susceptible to and resistant to duck hepatitis virus, as well as between different families. The results showed that the levels of IFN-α and IFN-β in resistant ducks were significantly higher than those in ducks susceptible to duck hepatitis virus, and there was a significant difference between the two. There were also significant differences in IFN-α and IFN-β levels between different families. Based on this, this invention uses IFN-α and IFN-β content as variables and employs multiple algorithms to construct predictive models for the survival status of ducks resistant to hepatitis A virus infection. By comparing the AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion) values ​​of the constructed predictive models, the goodness of fit and complexity of each model were evaluated. The results show that the predictive model for duck survival status resistant to hepatitis A virus infection constructed based on principal component analysis (PCA) combined with machine learning algorithms (PCA prediction model) and the Combined prediction model both performed best in terms of AIC and BIC values, indicating that these two prediction models have significant advantages in data fitting and complexity balance. This invention further compares the ROC (Receiver Operating Characteristic) curves and their corresponding AUC (Area Under the Curve) values ​​of different prediction models in predicting survival status. From the AUC value, the PCA prediction model has the highest AUC value, indicating that this prediction model has the best performance in predicting survival status. In addition, the ROC curve of the PCA prediction model is more prominent in the high sensitivity and high specificity regions, indicating that this prediction model has better overall performance in predicting survival status. Based on the predictive model provided by this invention, individuals or strains of Beijing ducks with strong disease resistance can be quickly screened, providing data-driven decision support for disease-resistant breeding, promoting the innovative application of data analysis in the aquaculture industry, improving biosecurity levels, and having important application prospects in duck disease-resistant breeding or the prevention of duck viral hepatitis. Attached Figure Description

[0034] Figure 1 The ROC curve (AUC = 0.86) of the GLMM-Model 1 prediction model on the training set is shown.

[0035] Figure 2 The ROC curve of the GLMM-Model 3 model on the training set (AUC is 0.85).

[0036] Figure 3 The ROC curve of the GLMM-Model 4 model on the training set (AUC is 0.86).

[0037] Figure 4 The ROC curve of the PCA model on the training set (AUC is 0.87).

[0038] Figure 5 The ROC curve (AUC = 0.86) of the Combined IFN Model on the training set.

[0039] Figure 6 This is the ROC curve of the PCA model on the validation set.

[0040] Figure 7 This represents the PCA score and survival prediction results of the PCA model on the validation set. Detailed Implementation

[0041] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer with the description. However, it should be understood that the embodiments described are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but such modifications or substitutions all fall within the protection scope of the present invention.

[0042] Experiment Example 1: Analysis of the differences in IFN-α and IFN-β levels between susceptible and resistant ducks and between different families.

[0043] 1. Biomaterials and Experimental Methods

[0044] 1.1 Experimental animals: 185 natural individuals of Beijing ducks (Beijing Duck Conservation Farm, Changping Experimental Base, Beijing Institute of Animal Husbandry and Veterinary Medicine, Chinese Academy of Agricultural Sciences).

[0045] 1.2 Detection of resistance indicators

[0046] 1.2.1 Blood Sample Collection

[0047] At 2 days of age, wing vein blood was collected from 54 Z3 strain Beijing ducks and 131 S4 strain Beijing ducks using heparin sodium anticoagulant blood collection tubes and stored at -20℃ for later use. DHAV-3 infection experiments were conducted on 3-day-old ducks, and deaths were recorded.

[0048] 1.2.2 Plasma Preparation

[0049] Take the venous blood obtained in step 1.2.1 and centrifuge it at 3500 r / min for 15 min to obtain plasma.

[0050] 1.2.3 Detection of IFN-α and IFN-β contents

[0051] DHAV-3 infection resulted in the death of 74 ducklings. The levels of IFN-α and IFN-β in the plasma of the 74 dead ducks and the plasma of the 111 surviving ducks were measured and statistically analyzed using ELISA.

[0052] 2 Experimental Results

[0053] The statistical results are shown in Table 1.

[0054] Table 1. Analysis of differences in plasma parameters among 185 Beijing ducks

[0055] index Resistant ducks (n=111) Susceptible ducks (n=74) p-value IFN-α (pg / mL) 69.80±9.522 64.54±11.12 0.0007 IFN-β (pg / mL) 71.42±9.724 65.49±10.53 0.0001

[0056] Statistical results showed that the levels of IFN-α and IFN-β differed significantly between dead and surviving ducks.

[0057] ANOVA analysis showed that the differences in IFN-β among different families were highly significant (F = 6.434, p < 2e-16), indicating that IFN-β levels varied significantly among families. The Kruskal-Wallis test further validated this result, showing that the differences in IFN-β among different families were also highly significant (χ²). 2 =108.31, df=36, p=3.633e-09). ANOVA analysis showed that the differences in IFN-α among different families were also highly significant (F=9.97, p<2e-16), indicating that there are also significant differences in IFN-α levels among families.

[0058] The Kruskal-Wallis test further validated this result, showing that the differences in IFN-α among different families were also highly significant (χ²). 2 =130.52, df=36, p=1.211e-12).

[0059] Table 2. Family Correlation Analysis of IFN-α and IFN-β Levels

[0060]

[0061] The results showed that IFN-α and IFN-β levels differed significantly among families.

[0062] Experiment Example 2: Survival Status Prediction Experiment Based on Generalized Linear Mixture Model

[0063] The document primarily employs the following analytical methods: Generalized Linear Mixed Model (GLMM): used to analyze the relationship between survival status and multiple variables (such as log_α, log_β, IFN-α, IFN-β, etc.). Specifically, in the PCA model, the IFN-α and IFN-β content is dimensionality-reduced, with the PCA score used as a fixed effect; in the interaction term analysis, interaction terms (such as IFN-α*IFN-β) are introduced to assess the interaction between variables; and in the combined index analysis, IFN-α and IFN-β are combined into a single index (combined_IFN) for analysis. The results are shown in Table 3.

[0064] Table 3 Comparison of different models

[0065]

[0066]

[0067] The GLMM-Model1 model converged well, with significant p-values ​​for both IFN-α and IFN-β (<2e-16). For every 1 unit increase in IFN-α level, the individual survival probability decreased significantly (β=-0.027, P<0.001), while for every 1 unit increase in IFN-β level, the individual survival probability decreased significantly (β=-0.066, P<0.001). The variance among families was 1.424, indicating that family background has a significant impact on survival status.

[0068] The GLMM-Model2 model converged well, but the p-values ​​of log_α and log_β were high, indicating that these two variables had no significant impact on survival status. For every 1 unit increase in log_α, there was no significant change in the individual survival probability (β = -2.291, P = 0.344), while for every 1 unit increase in log_β, there was no significant change in the individual survival probability (β = -3.880, P = 0.097). The variance of variation among families was 1.553, indicating that family background had a significant impact on survival status.

[0069] The GLMM-Model3 model converged well. For every 1 unit increase in IFN-α level, the individual survival probability decreased significantly (β = -0.082, P = 0.001). The variance of variation among families was 1.400, indicating that family background has a significant impact on survival status.

[0070] The GLMM-Model4 model converged well. For every 1 unit increase in IFN-β level, the individual survival probability decreased significantly (β = -0.087, P = 0.001). The variance of variation among families was 1.362, indicating that family background has a significant impact on survival status.

[0071] The GLMM-Model5 model failed to converge completely. The interaction between IFN-α and IFN-β had no significant effect on individual survival probability (β = 0.000, P = 0.783). For every 1 unit increase in IFN-α level, there was no significant change in individual survival probability (β = -0.047, P = 0.548), while for every 1 unit increase in IFN-β level, there was no significant change in individual survival probability (β = -0.086, P = 0.272). The variance among families was 1.507, indicating that family background has a significant impact on survival status.

[0072] The PCA model converged well. For every 1 unit increase in the composite score (pca_score), the individual survival probability increased significantly (β = 0.716, P = 0.001). The variance of variation among families was 1.529, indicating that family background has a significant impact on survival status.

[0073] The CombinedIFN Model converged well. For every 1 unit increase in the combined index (combined_ifn), the individual survival probability decreased significantly (β = -0.097, P = 0.001). The variance of variation among families was 1.529, indicating that family background has a significant impact on survival status.

[0074] The above results indicate that both IFN-α and IFN-β have a significant impact on survival status, and the variance among families is relatively high in all models, suggesting that family background has a significant impact on survival status. The coefficients of the fixed effects in GLMM-Model1, 2, 3, 4, and CombinedIFNModel are negative, which is inconsistent with the actual results, while GLMM-Model5 and PCA models are consistent with the actual results.

[0075] Experiment Example 3: Comparison of AIC and BIC performance of various prediction models

[0076] By comparing the AIC (Akaike Information Criterion) and BIC (Bayes Information Criterion) values ​​of multiple models, the goodness of fit and complexity of each model were evaluated. Both AIC and BIC tend to select models that fit the data well without being overly complex, and lower AIC and BIC values ​​generally indicate better models. The results are shown in Table 4.

[0077] Table 4. Performance Comparison of Multiple Models

[0078]

[0079] The PCA and Combined models have the lowest AIC values ​​(224.1144 and 224.1177, respectively), indicating they perform best in fitting the data. The GLMM-Model 4 model has the second highest AIC value (224.3947), also showing good fitting. The GLMM-Model 3 model has the highest AIC value (227.4613), indicating its relatively poor fitting. The PCA and Combined models have the lowest BIC values ​​(233.7754 and 233.7787, respectively), indicating they perform best in balancing goodness of fit and complexity. The GLMM-Model 4 model has the second highest BIC value (234.0557), also showing good balance. The GLMM-Model 1 model has the highest BIC value (238.6851), indicating its high complexity and relatively poor fitting.

[0080] The results show that the PCA model and the Combined model perform best in both AIC and BIC values, indicating their significant advantages in data fitting and complexity balance.

[0081] Experiment Example 4: Performance Comparison of Multiple Prediction Models

[0082] The ROC (Receiver Operating Characteristic) curves of different models in predicting survival status and their corresponding AUC (Area Under the Curve) values. Each file includes the specificity and sensitivity axes, as well as the model's AUC value. Figure 1 The image shows the ROC curve (AUC = 0.86) of the GLMM-Model 1 model on the training set. Figure 2 The image shows the ROC curve (AUC = 0.85) of the GLMM-Model 3 model on the training set. Figure 3 The image shows the ROC curve (AUC = 0.86) of the GLMM-Model 4 model on the training set. Figure 4 The image shows the ROC curve of the PCA model on the training set (AUC is 0.87). Figure 5 The ROC curve (AUC = 0.86) of the Combined IFN Model on the training set.

[0083] In terms of AUC values, the PCA model has the highest AUC, indicating that it has the best performance in predicting survival status. The GLMM-Model 1, Combined IFN Model, and GLMM-Model 4 models all have the same AUC value of 0.86, indicating that their performance is comparable. The GLMM-Model 3 model has an AUC value of 0.85, slightly lower than the other models.

[0084] The ROC curves of the GLMM-Model 1, Combined IFN Model, and GLMM-Model 4 models have similar shapes, indicating that their sensitivity and specificity performance is relatively consistent across different thresholds. Although the AUC value of the GLMM-Model 3 model is slightly lower, its ROC curve still shows good sensitivity and specificity at certain thresholds. The ROC curve of the PCA model is more prominent in the high sensitivity and high specificity regions, indicating that it has better overall performance in predicting survival status.

[0085] Experiment Example 5: PCA Model Training and Results

[0086] Stratified sampling by family was used to partition the data, ensuring that the representative distribution of each family in the training and validation sets remained consistent. Specifically, family ID was used as the stratification variable, and samples were randomly partitioned at a ratio of 80%:20%. This ensured that the model could learn common features within families while also validating its generalization ability on data from unknown families. Sample details: Total samples: 185 individuals; Training set samples: 163 individuals (88% of the total); Validation set samples: 22 individuals (12% of the total). The training set covered 37 different families, and the validation set contained representative samples from the families in the training set. Model performance validation results: The PCA mixed-effects model constructed using the training set showed good predictive performance on the validation set: the validation accuracy reached 72.7%; the validation AUC value was 0.762, indicating that the model has good discriminative ability. Figure 6 The difference between the training set AUC (0.87) and the validation set AUC (0.76) was 0.11, indicating that although the model has slight overfitting, it still maintains good generalization performance.

[0087] Logistic regression was used to fit the curves, showing the association between PCA scores and mortality risk. PCA analysis showed that IFN-α and IFN-β had equal positive weights (both 0.707) on the first principal component (PC1), indicating that they made comparable contributions to the prediction of survival and acted in the same direction. Figure 7 ).

Claims

1. A method for predicting the survival status of ducks against hepatitis A virus infection for non-diagnostic or therapeutic purposes, characterized in that, include: Step (1) Construct a predictive model for the survival status of ducks against hepatitis A virus infection using the levels of IFN-α and IFN-β in duck blood as variables; Step (2) Determine the levels of IFN-α and IFN-β in the blood of the duck to be tested; Step (3) Substitute the IFN-α and IFN-β levels in the blood of the duck to be tested, as determined in step (2), into the prediction model constructed in step (1). Based on the individual's family background and the output of the prediction model, predict the survival status of the duck to be tested against hepatitis A virus infection.

2. The prediction method according to claim 1, characterized in that, The method for constructing the prediction model described in step (1) includes: using the levels of IFN-α and IFN-β in duck blood as variables, and constructing a prediction model for the survival status of ducks against hepatitis A virus infection based on principal component analysis combined with machine learning algorithms.

3. The prediction method according to claim 2, characterized in that, The machine learning algorithm mentioned is a classification or regression model.

4. The prediction method according to claim 2, characterized in that, In constructing the prediction model, the dimensionality of IFN-α and IFN-β content in duck blood was reduced.

5. The prediction method according to claim 1, characterized in that, The formula for the prediction model described in step (1) is as follows: survival_status~pca_score+(1|family_id).

6. The prediction method according to claim 1, characterized in that, In step (2), the ELISA method was used to detect the content of IFN-α or IFN-β in duck blood.

7. The prediction method according to claim 1, characterized in that, The duck blood mentioned includes duck plasma or duck serum.

8. The prediction method according to claim 7, characterized in that, The duck plasma is prepared by any of the following methods (1)-(3): (1) Plasma obtained by centrifuging duck blood with heparin sodium anticoagulation; (2) Plasma obtained by centrifuging duck blood with EDTA anticoagulation; (3) Plasma obtained by anticoagulating duck blood using other methods and centrifuging it; The duck serum is prepared by either (1) or (2) below: (1) Serum obtained by centrifuging duck blood after it has been left to stand; (2) Serum obtained from duck blood using other methods.

9. The prediction method according to claim 1, characterized in that, Step (3) involves combining an individual's family background with the output of the prediction model to predict the survival status of the ducks against hepatitis A virus infection, including: (1) Retrieve the family_id of the duck to be tested from the dataset that records the family information of the duck to be tested and the IFN-α and IFN-β, match the original data of IFN-α and IFN-β of all individuals in the family, and calculate the basic IFN level of the family. (2) Detection and standardization of IFN index of ducks under test: The actual values ​​of IFN-α and IFN-β of ducks under test were detected and standardized to obtain standardized values: Standardized value = (measured value - mean of the whole dataset) / standard deviation of the whole dataset; (3) Prediction model input and probability output: Input the standardized values ​​of IFN-α and IFN-β and the basic level of IFN of the family into the prediction model; (4) Prediction is made based on the optimal threshold of the ROC curve of the entire dataset; if the optimal threshold P≥0.55, the prediction is survival, and if P<0.55, the prediction is death.

10. The prediction method according to any one of claims 1-9, characterized in that, The duck mentioned is the Peking duck.

Images