Benchmarking clinical risk prediction algorithms with ensemble machine learning for the noninvasive diagnosis of liver fibrosis in NAFLD.

BACKGROUND AND AIMS: Ensemble machine-learning methods, like the superlearner, combine multiple models into a single one to enhance predictive accuracy. Here we explore the potential of the superlearner as a benchmarking tool for clinical risk prediction, illustrating the approach to identifying significant liver fibrosis among patients with NAFLD. APPROACH AND RESULTS: We used 23 demographic/clinical variables to train superlearner(s) on data from the NASH-clinical research network observational study (n = 648) and validated models with data from the FLINT trial (n = 270) and National Health and Nutrition Examination Survey (NHANES) participants with NAFLD (n = 1244). Comparing the superlearner's performance to existing models (Fibrosis-4 [FIB-4], NAFLD fibrosis score, Forns, AST to Platelet Ratio Index [APRI], BARD, and Steatosis-Associated Fibrosis Estimator [SAFE]), it exhibited strong discriminative ability in the FLINT and NHANES validation sets, with AUCs of 0.79 (95% CI: 0.73-0.84) and 0.74 (95% CI: 0.68-0.79) respectively. CONCLUSIONS: Notably, the SAFE score performed similarly to the superlearner, both of which outperformed FIB-4, APRI, Forns, and BARD scores in the validation data sets. Surprisingly, the superlearner derived from 12 base models matched the performance of one with 90 base models. Overall, the superlearner, being the "best-in-class" machine-learning predictor, excelled in detecting fibrotic NASH, and this approach can be used to benchmark the performance of conventional clinical risk prediction models.

Heterogeneous Treatment Effects of Intensive Glycemic Control on Kidney Microvascular Outcomes and Mortality in ACCORD.

SIGNIFICANCE STATEMENT: Identifying and quantifying treatment effect variation across patients is the fundamental challenge of precision medicine. Here we quantify heterogeneous treatment effects of intensive glycemic control in the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, considering three outcomes of interest-a composite kidney outcome (driven by macroalbuminuria), all-cause mortality, and first assisted hypoglycemic event. We demonstrate that the effects of intensive glycemic control vary with risk of kidney failure, as predicted by the kidney failure risk equation (KFRE). Participants at highest risk of kidney failure gain the largest absolute kidney benefit of intensive glycemic control but also experience the largest absolute risk of death and hypoglycemic events. Our findings illustrate the value of identifying clinically meaningful treatment heterogeneity, particularly when treatments have different effects on multiple end points. OBJECTIVE: Clear criteria to individualize glycemic targets in patients with type II diabetes are lacking. In this post hoc analysis of the ACCORD, we evaluate whether the KFRE can identify patients for whom intensive glycemic control confers more benefit in preventing kidney microvascular outcomes. RESEARCH DESIGN AND METHODS: We divided the ACCORD trial population into quartiles on the basis of 5-year kidney failure risk using the KFRE. We estimated conditional treatment effects within each quartile and compared them with the average treatment effect in the trial. The treatment effects of interest were the 7-year restricted mean survival time (RMST) differences between intensive and standard glycemic control arms on ( 1 ) time-to-first development of severely elevated albuminuria or kidney failure and ( 2 ) all-cause mortality. RESULTS: We found evidence that the effect of intensive glycemic control on kidney microvascular outcomes and all-cause mortality varies with baseline risk of kidney failure. Patients with elevated baseline risk of kidney failure derived the most from intensive glycemic control in reducing kidney microvascular outcomes (7-year RMST difference of 114.8 [95% confidence interval 58.1 to 176.4] versus 48.4 [25.3 to 69.6] days in the entire trial population) However, this same patient group also experienced a shorter time to death (7-year RMST difference of -56.7 [-100.2 to -17.5] v. -23.6 [-42.2 to -6.6] days). CONCLUSIONS: We found evidence of heterogenous treatment effects of intensive glycemic control on kidney microvascular outcomes in ACCORD as a function of predicted baseline risk of kidney failure. Patients with higher kidney failure risk experienced the most pronounced reduction in kidney microvascular outcomes but also experienced the highest risk of all-cause mortality.

The effect of socioeconomic status on pediatric heart transplant outcomes at a single institution between 2013 and 2022.

BACKGROUND: Disparities in pediatric heart transplant outcomes based on socioeconomic status (SES) have been previously observed. However, there is a need to reevaluate these associations in contemporary settings with advancements in transplant therapies and increased awareness of health disparities. This retrospective study aims to investigate the relationship between SES and outcomes for pediatric heart transplant patients. METHODS: Data were collected through a chart review of 176 pediatric patients who underwent first orthotopic heart transplantation (OHT) at a single center from 2013 to 2021. The Area Deprivation Index (ADI), a composite score based on U.S. census data, was used to quantify SES. Cox proportional hazards models and generalized linear models were employed to analyze the association between SES and graft failure, rejection rates, and hospitalization rates. RESULTS: The analysis revealed no statistically significant differences in graft failure rates, rejection rates, or hospitalization rates between low-SES and high-SES pediatric heart transplant patients for our single-center study. CONCLUSION: There may be patient education, policies, and social resources that can help mitigate SES-based healthcare disparities. Additional multi-center research is needed to identify post-transplant care that promotes patient equity.

Statistical methods for Mendelian models with multiple genes and cancers.

Risk evaluation to identify individuals who are at greater risk of cancer as a result of heritable pathogenic variants is a valuable component of individualized clinical management. Using principles of Mendelian genetics, Bayesian probability theory, and variant-specific knowledge, Mendelian models derive the probability of carrying a pathogenic variant and developing cancer in the future, based on family history. Existing Mendelian models are widely employed, but are generally limited to specific genes and syndromes. However, the upsurge of multigene panel germline testing has spurred the discovery of many new gene-cancer associations that are not presently accounted for in these models. We have developed PanelPRO, a flexible, efficient Mendelian risk prediction framework that can incorporate an arbitrary number of genes and cancers, overcoming the computational challenges that arise because of the increased model complexity. We implement an 11-gene, 11-cancer model, the largest Mendelian model created thus far, based on this framework. Using simulations and a clinical cohort with germline panel testing data, we evaluate model performance, validate the reverse-compatibility of our approach with existing Mendelian models, and illustrate its usage. Our implementation is freely available for research use in the PanelPRO R package.

A likelihood-based approach to assessing frequency of pathogenicity among variants of unknown significance in susceptibility genes.

Commercialized multigene panel testing brings unprecedented opportunities to understand germline genetic contributions to hereditary cancers. Most genetic testing companies classify the pathogenicity of variants as pathogenic, benign, or variants of unknown significance (VUSs). The unknown pathogenicity of VUSs poses serious challenges to clinical decision-making. This study aims to assess the frequency of VUSs that are likely pathogenic in disease-susceptibility genes. Using estimates of probands' probability of having a pathogenic mutation (ie, the carrier score) based on a family history probabilistic risk prediction model, we assume the carrier score distribution for probands with VUSs is a mixture of the carrier score distribution for probands with positive results and the carrier score distribution for probands with negative results. Under this mixture model, we propose a likelihood-based approach to assess the frequency of pathogenicity among probands with VUSs, while accounting for the existence of possible pathogenic mutations on genes not tested. We conducted simulations to assess the performance of the approach and show that under various settings, the approach performs well with very little bias in the estimated proportion of VUSs that are likely pathogenic. We also estimate the positive predictive value across the entire range of carrier scores. We apply our approach to the USC-Stanford Hereditary Cancer Panel Testing cohort, and estimate the proportion of probands that have VUSs in BRCA1/2 that are likely pathogenic to be 10.12% [95%CI: 0%, 43.04%]. This approach will enable clinicians to target high-risk patients who have VUSs, allowing for early prevention interventions.

Evaluating the utility of multi-gene, multi-disease population-based panel testing accounting for uncertainty in penetrance estimates.

Panel germline testing allows for the efficient detection of deleterious variants for multiple conditions, but the benefits and harms of identifying these variants are not always well understood. We present a multi-gene, multi-disease aggregate utility formula that allows the user to consider adding or removing each gene in a panel based on variant frequency, estimated penetrances, and subjective disutilities for testing positive but not developing the disease and testing negative but developing the disease. We provide credible intervals for utility that reflect uncertainty in penetrance estimates. Rare, highly penetrant deleterious variants tend to contribute positive net utilities for a wide variety of user-specified disutilities, even when accounting for parameter estimation uncertainty. However, the clinical utility of deleterious variants with moderate, uncertain penetrance depends more on assumed disutilities. The decision to include a gene on a panel depends on variant frequency, penetrance, and subjective utilities and should account for uncertainties around these factors.

Low infectivity among asymptomatic patients with a positive severe acute respiratory coronavirus virus 2 (SARS-CoV-2) admission test at a tertiary care center, 2020-2022.

We used a strand-specific RT-qPCR to evaluate viral replication as a surrogate for infectiousness among 242 asymptomatic inpatients with a positive severe acute respiratory coronavirus virus 2 (SARS-CoV-2) admission test. Only 21 patients (9%) had detectable SARS-CoV-2 minus-strand RNA. Because most patients were found to be noninfectious, our findings support the suspension of asymptomatic admission testing.

Nirmatrelvir-Ritonavir and Symptoms in Adults With Postacute Sequelae of SARS-CoV-2 Infection: The STOP-PASC Randomized Clinical Trial.

Importance: There is an urgent need to identify treatments for postacute sequelae of SARS-CoV-2 infection (PASC). Objective: To assess the efficacy of a 15-day course of nirmatrelvir-ritonavir in reducing the severity of select PASC symptoms. Design, Setting, and Participants: This was a 15-week blinded, placebo-controlled, randomized clinical trial conducted from November 2022 to September 2023 at Stanford University (California). The participants were adults with moderate to severe PASC symptoms of 3 months or longer duration. Interventions: Participants were randomized 2:1 to treatment with oral nirmatrelvir-ritonavir (NMV/r, 300 mg and 100 mg) or with placebo-ritonavir (PBO/r) twice daily for 15 days. Main Outcomes and Measures: Primary outcome was a pooled severity of 6 PASC symptoms (fatigue, brain fog, shortness of breath, body aches, gastrointestinal symptoms, and cardiovascular symptoms) based on a Likert scale score at 10 weeks. Secondary outcomes included symptom severity at different time points, symptom burden and relief, patient global measures, Patient-Reported Outcomes Measurement Information System (PROMIS) measures, orthostatic vital signs, and sit-to-stand test change from baseline. Results: Of the 155 participants (median [IQR] age, 43 [34-54] years; 92 [59%] females), 102 were randomized to the NMV/r group and 53 to the PBO/r group. Nearly all participants (n = 153) had received the primary series for COVID-19 vaccination. Mean (SD) time between index SARS-CoV-2 infection and randomization was 17.5 (9.1) months. There was no statistically significant difference in the model-derived severity outcome pooled across the 6 core symptoms at 10 weeks between the NMV/r and PBO/r groups. No statistically significant between-group differences were found at 10 weeks in the Patient Global Impression of Severity or Patient Global Impression of Change scores, summative symptom scores, and change from baseline to 10 weeks in PROMIS fatigue, dyspnea, cognitive function, and physical function measures. Adverse event rates were similar in NMV/r and PBO/r groups and mostly of low grade. Conclusions and Relevance: The results of this randomized clinical trial showed that a 15-day course of NMV/r in a population of patients with PASC was generally safe but did not demonstrate a significant benefit for improving select PASC symptoms in a mostly vaccinated cohort with protracted symptom duration. Further studies are needed to determine the role of antivirals in the treatment of PASC. Trial Registration: ClinicalTrials.gov Identifier: NCT05576662.

Benchmarking clinical risk prediction algorithms with ensemble machine learning: An illustration of the superlearner algorithm for the non-invasive diagnosis of liver fibrosis in non-alcoholic fatty liver disease.

Background and Aims: Ensemble machine learning (ML) methods can combine many individual models into a single 'super' model using an optimal weighted combination. Here we demonstrate how an underutilized ensemble model, the superlearner, can be used as a benchmark for model performance in clinical risk prediction. We illustrate this by implementing a superlearner to predict liver fibrosis in patients with non-alcoholic fatty liver disease (NAFLD). Methods: We trained a superlearner based on 23 demographic and clinical variables, with the goal of predicting stage 2 or higher liver fibrosis. The superlearner was trained on data from the Non-alcoholic steatohepatitis - clinical research network observational study (NASH-CRN, n=648), and validated using data from participants in a randomized trial for NASH ('FLINT' trial, n=270) and data from examinees with NAFLD who participated in the National Health and Nutrition Examination Survey (NHANES, n=1244). We compared the performance of the superlearner with existing models, including FIB-4, NFS, Forns, APRI, BARD and SAFE. Results: In the FLINT and NHANES validation sets, the superlearner (derived from 12 base models) discriminates patients with significant fibrosis from those without well, with AUCs of 0.79 (95% CI: 0.73-0.84) and 0.74 (95% CI: 0.68-0.79). Among the existing scores considered, the SAFE score performed similarly to the superlearner, and the superlearner and SAFE scores outperformed FIB-4, APRI, Forns, and BARD scores in the validation datasets. A superlearner model derived from 12 base models performed as well as one derived from 90 base models. Conclusions: The superlearner, thought of as the "best-in-class" ML prediction, performed better than most existing models commonly used in practice in detecting fibrotic NASH. The superlearner can be used to benchmark the performance of conventional clinical risk prediction models.

Heterogeneous treatment effects of intensive glycemic control on kidney microvascular outcomes in ACCORD.

Objective: Clear criteria to individualize glycemic targets are lacking. In this post-hoc analysis of the Action to Control Cardiovascular Risk in Diabetes trial (ACCORD), we evaluate whether the kidney failure risk equation (KFRE) can identify patients who disproportionately benefit from intensive glycemic control on kidney microvascular outcomes. Research design and methods: We divided the ACCORD trial population in quartiles based on 5-year kidney failure risk using the KFRE. We estimated conditional treatment effects within each quartile and compared them to the average treatment effect in the trial. The treatment effects of interest were the 7-year restricted-mean-survival-time (RMST) differences between intensive and standard glycemic control arms on (1) time-to-first development of severely elevated albuminuria or kidney failure and (2) all-cause mortality. Results: We found evidence that the effect of intensive glycemic control on kidney microvascular outcomes and all-cause mortality varies with baseline risk of kidney failure. Patients with elevated baseline risk of kidney failure benefitted the most from intensive glycemic control on kidney microvascular outcomes (7-year RMST difference of 115 v. 48 days in the entire trial population) However, this same patient group also experienced shorter times to death (7-year RMST difference of -57 v. -24 days). Conclusions: We found evidence of heterogenous treatment effects of intensive glycemic control on kidney microvascular outcomes in ACCORD as a function of predicted baseline risk of kidney failure. Patients with higher kidney failure risk experienced the most pronounced benefits of treatment on kidney microvascular outcomes but also experienced the highest risk of all-cause mortality.

Barriers to Family Building Among Physicians and Medical Students.

Importance: Physicians and medical students who desire to build families face significant barriers due to the structure and culture of medicine. Objective: To understand the barriers and facilitators to family building for all people in medicine-not only individuals who can become pregnant-through an open-ended, qualitative analysis of survey responses. Design, Setting, and Participants: This qualitative study used a survey conducted in April and May 2021 with a broad sample of physicians and medical students. Participants were recruited through social media, targeting physician and medical student communities. Physicians (residents, fellows, and physicians in independent practice) and medical students of all gender identities and sexual orientations were included. Informed by a postpositivist approach, coding reliability thematic analysis was performed on 3 open-ended survey questions on family-building experiences (what they would do differently, what advice they have for others, and anything else they wished to share). Main Outcomes and Measures: Identified themes were mapped to the social-ecological model, a model used in public health to examine how a spectrum of factors is associated with health outcomes. Results: A total of 2025 people (1860 [92%] women; 299 [15%] Asian, 151 [8%] Black, and 1303 [64%] White; 1730 [85%] heterosexual; and 1200 [59%] physicians who had completed training) responded to at least 1 of 3 open-ended questions. Themes mapped to social-ecological model levels included: (1) cultural, eg, medical training being at odds with family building; (2) organizational, eg, lack of institutional support for the range of family-building routes; (3) interpersonal, eg, impact of social support on family building; and (4) individual, eg, socioeconomic status and other individual factors that facilitate or inhibit family building. Recommendations to improve family-building experiences include implementing family-building curricula at medical schools, providing adequate parental leave for all physicians and medical students who become parents, and providing insurance coverage for all family-building routes. Conclusions and Relevance: In this qualitative study of physicians and medical students, self-reported barriers to family building were identified at each level of the social-ecological model. Addressing these barriers is critical to creating a more equitable family-building environment for physicians and medical students.

Multi-syndrome, multi-gene risk modeling for individuals with a family history of cancer with the novel R package PanelPRO.

Identifying individuals who are at high risk of cancer due to inherited germline mutations is critical for effective implementation of personalized prevention strategies. Most existing models focus on a few specific syndromes; however, recent evidence from multi-gene panel testing shows that many syndromes are overlapping, motivating the development of models that incorporate family history on several cancers and predict mutations for a comprehensive panel of genes.We present PanelPRO, a new, open-source R package providing a fast, flexible back-end for multi-gene, multi-cancer risk modeling with pedigree data. It includes a customizable database with default parameter values estimated from published studies and allows users to select any combinations of genes and cancers for their models, including well-established single syndrome BayesMendel models (BRCAPRO and MMRPRO). This leads to more accurate risk predictions and ultimately has a high impact on prevention strategies for cancer and clinical decision making. The package is available for download for research purposes at https://projects.iq.harvard.edu/bayesmendel/panelpro.

Genetic mutations that increase cancer risk can be passed down from parents to their children, which can affect families across many generations. In these families, multiple members may be affected by different types of cancer, and these cancers often develop at an early age. Unaffected family members are often referred to genetic counselling, where they can explore their own risk of cancer. Clinicians and genetic counselors can provide recommendations to minimize cancer risk and inform personal choices on how to manage that risk, such as opting for preventative surgeries or participating in regular screening. In genetic counselling sessions, highly trained clinicians and specialists use software that takes an individual's family history of cancer and uses it to estimate their individual risk of carrying certain genetic mutations. These estimates can in turn help to predict their future risk of cancer. Many existing software packages are limited to estimating risks based on mutations in well-known cancer-related genes, such as BRCA1 and BRCA2 in breast and ovarian cancer. However, emerging evidence suggests that many of the genes associated with cancer risk work as part of a complex and overlapping network. Since current risk-profiling software packages are only designed to consider such genes in isolation, they cannot generate the most robust, accurate or comprehensive cancer risk profiles. To address this challenge, Lee, Liang et al. have developed a new risk-profiling software that can integrate a large number of gene mutations and a wide range of potential cancer types to provide more accurate estimates of individual cancer risk. This software, called PanelPRO, uses evidence identified from extensive literature reviews to model the complex interplay between genes and cancer risk. The software not only calculates risks based on known genes, but also allows other developers to integrate new cancer-related genes that may be identified in the future. Importantly, the software is compatible with genetic counselling applications, since it returns answers within seconds when reasonable family and gene database sizes are used. PanelPRO is a new, modern, flexible and efficient software package that provides an important advance towards modelling the vast genetic and biological complexity that contributes to inherited cancer risk. This software is designed to provide a more accurate and comprehensive estimate of cancer risk for individuals with family histories of cancer. As an open-source software, it is freely available for research purposes, and can be licensed by software companies and healthcare organizations to integrate electronic patient records and rapidly identify at-risk individuals across larger patient groups. Ultimately, this software has the potential to improve cancer prevention strategies and optimize the personalized decision-making processes around cancer risk.

Matrix Linear Models for High-Throughput Chemical Genetic Screens.

We develop a flexible and computationally efficient approach for analyzing high-throughput chemical genetic screens. In such screens, a library of genetic mutants is phenotyped in a large number of stresses. Typically, interactions between genes and stresses are detected by grouping the mutants and stresses into categories, and performing modified t-tests for each combination. This approach does not have a natural extension if mutants or stresses have quantitative or nonoverlapping annotations (e.g., if conditions have doses or a mutant falls into more than one category simultaneously). We develop a matrix linear model (MLM) framework that allows us to model relationships between mutants and conditions in a simple, yet flexible, multivariate framework. It encodes both categorical and continuous relationships to enhance detection of associations. We develop a fast estimation algorithm that takes advantage of the structure of MLMs. We evaluate our method's performance in simulations and in an Escherichia coli chemical genetic screen, comparing it with an existing univariate approach based on modified t-tests. We show that MLMs perform slightly better than the univariate approach when mutants and conditions are classified in nonoverlapping categories, and substantially better when conditions can be ordered in dosage categories. Therefore, it is an attractive alternative to current methods, and provides a computationally scalable framework for larger and complex chemical genetic screens. A Julia language implementation of MLMs and the code used for this paper are available at https://github.com/janewliang/GeneticScreen.jl and https://bitbucket.org/jwliang/mlm_gs_supplement, respectively.

Longitudinal study of leukocyte DNA methylation and biomarkers for cancer risk in older adults.

BACKGROUND: Changes in DNA methylation over the course of life may provide an indicator of risk for cancer. We explored longitudinal changes in CpG methylation from blood leukocytes, and likelihood of future cancer diagnosis. METHODS: Peripheral blood samples were obtained at baseline and at follow-up visit from 20 participants in the Health, Aging and Body Composition prospective cohort study. Genome-wide CpG methylation was assayed using the Illumina Infinium Human MethylationEPIC (HM850K) microarray. RESULTS: Global patterns in DNA methylation from CpG-based analyses showed extensive changes in cell composition over time in participants who developed cancer. By visit year 6, the proportion of CD8+ T-cells decreased (p-value = 0.02), while granulocytes cell levels increased (p-value = 0.04) among participants diagnosed with cancer compared to those who remained cancer-free (cancer-free vs. cancer-present: 0.03 ± 0.02 vs. 0.003 ± 0.005 for CD8+ T-cells; 0.52 ± 0.14 vs. 0.66 ± 0.09 for granulocytes). Epigenome-wide analysis identified three CpGs with suggestive p-values ≤10- 5 for differential methylation between cancer-free and cancer-present groups, including a CpG located in MTA3, a gene linked with metastasis. At a lenient statistical threshold (p-value ≤3 × 10- 5), the top 10 cancer-associated CpGs included a site near RPTOR that is involved in the mTOR pathway, and the candidate tumor suppressor genes REC8, KCNQ1, and ZSWIM5. However, only the CpG in RPTOR (cg08129331) was replicated in an independent data set. Analysis of within-individual change from baseline to Year 6 found significant correlations between the rates of change in methylation in RPTOR, REC8 and ZSWIM5, and time to cancer diagnosis. CONCLUSION: The results show that changes in cellular composition explains much of the cross-sectional and longitudinal variation in CpG methylation. Additionally, differential methylation and longitudinal dynamics at specific CpGs could provide powerful indicators of cancer development and/or progression. In particular, we highlight CpG methylation in the RPTOR gene as a potential biomarker of cancer that awaits further validation.

Profiling DNA methylation differences between inbred mouse strains on the Illumina Human Infinium MethylationEPIC microarray.

The Illumina Infinium MethylationEPIC provides an efficient platform for profiling DNA methylation in humans at over 850,000 CpGs. Model organisms such as mice do not currently benefit from an equivalent array. Here we used this array to measure DNA methylation in mice. We defined probes targeting conserved regions and performed differential methylation analysis and compared between the array-based assay and affinity-based DNA sequencing of methyl-CpGs (MBD-seq) and reduced representation bisulfite sequencing. Mouse samples consisted of 11 liver DNA from two strains, C57BL/6J (B6) and DBA/2J (D2), that varied widely in age. Linear regression was applied to detect differential methylation. In total, 13,665 probes (1.6% of total probes) aligned to conserved CpGs. Beta-values (ß-value) for these probes showed a distribution similar to that in humans. Overall, there was high concordance in methylation signal between the EPIC array and MBD-seq (Pearson correlation r = 0.70, p-value < 0.0001). However, the EPIC probes had higher quantitative sensitivity at CpGs that are hypo- (ß-value < 0.3) or hypermethylated (ß-value > 0.7). In terms of differential methylation, no EPIC probe detected a significant difference between age groups at a Benjamini-Hochberg threshold of 10%, and the MBD-seq performed better at detecting age-dependent change in methylation. However, the top most significant probe for age (cg13269407; uncorrected p-value = 1.8 x 10-5) is part of the clock CpGs used to estimate the human epigenetic age. For strain, 219 EPIC probes detected significant differential methylation (FDR cutoff 10%) with ~80% CpGs associated with higher methylation in D2. This higher methylation profile in D2 compared to B6 was also replicated by the MBD-seq data. To summarize, we found only a small subset of EPIC probes that target conserved sites. However, for this small subset the array provides a reliable assay of DNA methylation and can be effectively used to measure differential methylation in mice.