The epigenetic clock evaluates human biological age based on DNA methylation patterns. It takes the form of a regression model where the methylation ratio at CpG sites serves as the predictor and age as the response variable. Due to the large number of CpG sites and their correlation, Elastic Net is commonly used to train the models. However, existing standard epigenetic clocks, trained on multiracial data, may exhibit biases due to genetic and environmental differences among specific racial groups. Developing epigenetic clocks suitable for a specific single-race population requires collecting and analyzing hundreds or thousands of samples, which costs a lot of time and money. Therefore, an efficient method to construct accurate epigenetic clocks with smaller sample sizes is needed. We propose Transfer Elastic Net, a transfer learning approach that trains a model in the target population using the information of parameters estimated by the Elastic Net in a source population. Using this method, we constructed Horvath’s, Hannum’s, and Levine’s types of epigenetic clocks from blood samples of 143 Japanese subjects. The DNA methylation data were transformed through principal component analysis to obtain more reliable clocks. The developed clocks demonstrated the smallest prediction errors compared to both the original clocks and those trained with the Elastic Net on the same Japanese data. Transfer Elastic Net can also be applied to develop epigenetic clocks for other specific populations, and is expected to be applied in various fields.
Epigenetic clocks are regression models that predict biological age based on the DNA methylation patterns. They are developed using DNA methylation array data; however, systematic differences in the predicted values between the Infinium Methylation Screening Array (MSA) and Infinium MethylationEPIC v2.0 BeadChip (EPICv2) platforms have not been evaluated. We quantified the systematic differences between the MSA and EPICv2 arrays for six major epigenetic clocks (Horvath, Hannum, PhenoAge, GrimAge, GrimAge v2, and DunedinPACE) using $166$ identical human blood samples measured on both platforms. The mean differences (MSA $-$ EPICv2) ranged from $-7.99$ years for PhenoAge to $21.8$ years for the Hannum clock, despite strong correlations between platforms ($r = 0.854$ -- $0.991$). Based on these training data, we developed three linear regression models to correct the MSA-based epigenetic clocks: Model 1 (offset location correction), Model 2 (slope and location correction), and Model 3 (slope and location correction with covariates). Validation on $48$ independent samples measured at a different institution showed that all models reduced cross-platform bias. Although Model 3 tended to achieve the lowest bias; $0.204$ years for PhenoAge and $0.383$ years for GrimAge v2, Model 1, which was the simplest, provided a sufficient correction: $-0.541$ years of bias for PhenoAge and $0.516$ for GrimAge v2. These methods are expected to enable reliable comparisons of epigenetic age across platforms and facilitate large-scale epigenetic studies.
1 January 2025 in Computational and Structural Biotechnology Journal
DNA methylation is associated with various diseases and aging; thus, longitudinal and repeated assessments of methylation patterns are crucial for revealing the mechanisms of disease onset and identifying factors associated with aging. The presence of batch effects influences the analysis of DNA methylation array data. As existing methods for correcting batch effects are designed to correct all samples simultaneously, when data are incrementally measured and included, the correction of newly added data affects previous data. Therefore, we propose an incremental framework for batch-effect correction based on ComBat, a location/scale adjustment approach using a Bayesian hierarchical model, and empirical Bayes estimation. Using numerical experiments and application to actual data, we demonstrate that the proposed method can correct newly included data without re-correcting the old data. The proposed method is expected to be useful for studies involving repeated measurements of DNA methylation, such as clinical trials of anti-aging interventions.
Background: Epigenetic clocks based on DNA methylation (DNAm) are widely used indicators of biological aging; however, most established models have been developed using EPIC arrays and non-Japanese populations. The Methylation Screening Array (MSA), a cost-efficient platform with reduced CpG content, has not been evaluated for its capacity to support biological age estimation and biomarker prediction in Japanese cohorts. Methods: DNAm profiles and clinical laboratory measurements were obtained from 166 Japanese participants for model development; an independent cohort of 48 individuals processed at a separate institute was used for validation. A linear regression model was trained using the Elastic Net method to predict phenotypic age from MSA-derived methylation data, and a two-stage modeling (residual learning) framework integrating EPIC-based clock predictions with MSA-specific residual predictions was evaluated. Additional models were constructed to examine the predictability of 59 clinical biomarkers and their log-transformed variants, including sex-stratified analyses. Results: The MSA-based model accurately predicted phenotypic age in the validation dataset; prediction performance improved when the EPIC-based estimates were incorporated through the residual learning framework. Several clinical biomarkers, particularly those related to leukocyte composition and sex hormone regulation, were also predicted from the MSA data, although some markers were strongly affected by sex. Some of the nine constituent phenotypic age biomarkers were not individually predicted. Conclusions: MSA methylation profiles contain sufficient biological information for reliable prediction of epigenetic aging markers in Japanese individuals. These findings demonstrate the feasibility of applying cost-efficient MSA-based DNAm profiling for biological age prediction and provide a methodological foundation for expanding epigenetic biomarker applications in Japan.
Motivation The epigenetic clock evaluates human biological age based on DNA methylation patterns. It takes the form of a regression model where the methylation ratio at CpG sites serves as the predictor, and chronological or adjusted age as the response variable. Due to the large number of CpG sites considered as candidate explanatory variables and their potential correlation, Elastic Net is commonly used to train the regression models. However, existing standard epigenetic clocks, trained on multiracial data, may exhibit biases due to genetic and environmental differences among specific racial groups. The development of epigenetic clocks suitable for a single-race population typically necessitates the collection of hundreds to thousands of samples to measure DNA methylation and other biomarkers, which costs a lot of time and money. Consequently, a method for developing accurate epigenetic clocks with relatively small sample sizes is needed. Results We propose Transfer Elastic Net, a transfer learning approach that uses the parameter information from a linear regression model trained with the Elastic Net to estimate another model. Using this method, we constructed Horvath’s, Hannum’s, and Levine’s types of epigenetic clocks using DNA methylation data from blood samples of 143 Japanese subjects. The data were transformed through principal component analysis to obtain more reliable clocks. The developed clocks demonstrated the smallest prediction errors compared to both the original clocks and those trained with the Elastic Net on the same Japanese data. Furthermore, the bias relative to the original clocks was reduced. Thus, we successfully developed epigenetic clocks that are well-suited for the Japanese population. Transfer Elastic Net can also be applied to develop epigenetic clocks for other specific populations, and is expected to be applied in various fields. Availability https://github.com/t-yui/TransferENet-EpigeneticClock
Background: Social cognitive impairments are a fundamental aspect of schizophrenia, exerting a substantial influence on patients’ functional outcomes. However, to date, there have been no meta-analyses of comprehensive pharmacological interventions covering all domains of social cognition. The aim of the present study was to address this knowledge gap by conducting a network meta-analysis, a comprehensive approach that systematically compares the efficacy of pharmacological interventions across all domains of social cognition. Methods: A literature search for randomized controlled trials (RCTs) was conducted using PubMed, Embase, the Cochrane Central Register of Controlled Trials, PsycINFO, ClinicalTrials.gov, and the International Clinical Trials Registry Platform. The Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines were followed. Results: A total of 8,752 records were screened, and 60 RCTs involving 4,270 subjects were included in the systematic review. Thirty-six pharmacological interventions were extracted, but no compounds had a significant ameliorative effect on social cognition in comparison with placebo. In each domain of social cognition, the following compounds were identified as the most probable candidates for treatment selection: selective glycine uptake inhibitor (standardized mean difference [SMD], 0.46; 95% credible interval [CI], −0.52 to 1.44) and stimulant (SMD, 0.44; 95% CI, −0.57 to 1.45) for emotion perception in comparison with placebo. In the context of emotion processing, γ-aminobutyric acid (A) α2/α3 partial agonist (SMD, 0.33; 95% CI, −0.53 to 1.19) emerged as the top compound. Conclusions: To date, no pharmacological interventions have demonstrated efficacy for social cognitive impairments in schizophrenia.