External validation of machine learning models including newborn metabolomic markers for postnatal gestational age estimation in East and South-East Asian infants

BioMapAI: Artificial Intelligence Multi-Omics Modeling of Myalgic Encephalomyelitis / Chronic Fatigue Syndrome

Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a chronic illness with a multifactorial etiology and heterogeneous symptomatology, posing major challenges for diagnosis and treatment. Here, we present BioMapAI, a supervised deep neural network trained on a four-year, longitudinal, multi-omics dataset from 249 participants, which integrates gut metagenomics, plasma metabolomics, immune cell profiling, blood laboratory data, and detailed clinical symptoms. By simultaneously modeling these diverse data types to predict clinical severity, BioMapAI identifies disease- and symptom-specific biomarkers and robustly classifies ME/CFS in both held-out and independent external cohorts. Using an explainable AI approach, we construct the first connectivity map spanning the microbiome, immune system, and plasma metabolome in health and ME/CFS, adjusted for age, gender, and additional clinical factors. This map uncovers disrupted associations between microbial metabolism (e.g., short-chain fatty acids, branched-chain amino acids, tryptophan, benzoate), plasma lipids and bile acids, and heightened inflammatory responses in mucosal and inflammatory T cell subsets (MAIT, γδT) secreting IFNγ and GzA. Overall, BioMapAI provides unprecedented systems-level insights into ME/CFS, refining existing hypotheses and hypothesizing new pathways associated to the disease's heterogeneous symptoms.

Estimating Gestational Age and Prediction of Preterm Birth Using Metabolomics Biomarkers

Preterm birth (PTB) is a leading cause of morbidity and mortality in children aged under 5 years globally, especially in low-resource settings. It remains a challenge in many low-income and middle-income countries to accurately measure the true burden of PTB due to limited availability of accurate measures of gestational age (GA), first trimester ultrasound dating being the gold standard. Metabolomics biomarkers are a promising area of research that could provide tools for both early identification of high-risk pregnancies and for the estimation of GA and preterm status of newborns postnatally.

Accuracy of prenatal and postnatal biomarkers for estimating gestational age: a systematic review and meta-analysis

Background

Knowledge of gestational age (GA) is key in clinical management of individual obstetric patients, and critical to be able to calculate rates of preterm birth and small for GA at a population level. Currently, the gold standard for pregnancy dating is measurement of the fetal crown rump length at 11-14 weeks of gestation. However, this is not possible for women first presenting in later pregnancy, or in settings where routine ultrasound is not available. A reliable, cheap and easy to measure GA-dependent biomarker would provide an important breakthrough in estimating the age of pregnancy. Therefore, the aim of this study was to determine the accuracy of prenatal and postnatal biomarkers for estimating gestational age (GA).

Methods

Systematic review prospectively registered with PROSPERO (CRD42020167727) and reported in accordance with the PRISMA-DTA. Medline, Embase, CINAHL, LILACS, and other databases were searched from inception until September 2023 for cohort or cross-sectional studies that reported on the accuracy of prenatal and postnatal biomarkers for estimating GA. In addition, we searched Google Scholar and screened proceedings of relevant conferences and reference lists of identified studies and relevant reviews. There were no language or date restrictions. Pooled coefficients of correlation and root mean square error (RMSE, average deviation in weeks between the GA estimated by the biomarker and that estimated by the gold standard method) were calculated. The risk of bias in each included study was also assessed.

Findings

Thirty-nine studies fulfilled the inclusion criteria: 20 studies (2,050 women) assessed prenatal biomarkers (placental hormones, metabolomic profiles, proteomics, cell-free RNA transcripts, and exon-level gene expression), and 19 (1,738,652 newborns) assessed postnatal biomarkers (metabolomic profiles, DNA methylation profiles, and fetal haematological components). Among the prenatal biomarkers assessed, human chorionic gonadotrophin measured in maternal serum between 4 and 9 weeks of gestation showed the highest correlation with the reference standard GA, with a pooled coefficient of correlation of 0.88. Among the postnatal biomarkers assessed, metabolomic profiling from newborn blood spots provided the most accurate estimate of GA, with a pooled RMSE of 1.03 weeks across all GAs. It performed best for term infants with a slightly reduced accuracy for preterm or small for GA infants. The pooled RMSEs for metabolomic profiling and DNA methylation profile from cord blood samples were 1.57 and 1.60 weeks, respectively.

Interpretation

We identified no antenatal biomarkers that accurately predict GA over a wide window of pregnancy. Postnatally, metabolomic profiling from newborn blood spot provides an accurate estimate of GA, however, as this is known only after birth it is not useful to guide antenatal care. Further prenatal studies are needed to identify biomarkers that can be used in isolation, as part of a biomarker panel, or in combination with other clinical methods to narrow prediction intervals of GA estimation.

Funding

The research was funded by the Bill and Melinda Gates Foundation (INV-000368). ATP is supported by the Oxford Partnership Comprehensive Biomedical Research Centre with funding from the NIHR Biomedical Research Centre funding scheme. The views expressed are those of the authors and not necessarily those of the UK National Health Service, the NIHR, the Department of Health, or the Department of Biotechnology. The funders of this study had no role in study design, data collection, analysis or interpretation of the data, in writing the paper or the decision to submit for publication.

Validation of gestational age determination from ultrasound or a metabolic gestational age algorithm using exact date of conception in a cohort of newborns conceived using assisted reproduction technologies

BACKGROUND

Accurate estimates of gestational age in pregnancy are important for the provision of optimal care. Although current guidelines generally recommend estimating gestational age via first-trimester ultrasound measurement of crown-rump length, error associated with this method can range from 3 to 8 days of gestation. In pregnancies resulting from assisted reproductive technology, estimated due date can be calculated on the basis of the age of the embryo and the date of embryo transfer, arguably providing the most accurate estimates possible. We have developed and extensively validated statistical models to estimate gestational age postnatally using metabolomic markers from blood samples in combination with clinical and demographic data. These models have shown high accuracy compared with first-trimester ultrasound, the recommended method for estimating gestational age in spontaneous pregnancies. We hypothesized that gestational age derived from date and stage of embryo at transfer in newborns conceived using assisted reproduction therapy would provide the most accurate reference standard possible to evaluate and compare the accuracy of both first-trimester ultrasound and metabolomic model-based gestational dating.

OBJECTIVE

This study aimed to validate both first-trimester ultrasound dating and postnatal metabolomic gestational age estimation models against gestational age derived from date and stage of embryo at transfer in a cohort of newborns conceived via assisted reproductive technology, both overall and in important subgroups of interest (preterm birth, small for gestational age, and multiple birth).

STUDY DESIGN

This was a retrospective cohort study of infants born in Ontario, Canada between 2015 and 2017 and captured in the provincial birth registry. Spontaneous conceptions were randomly partitioned into a model derivation sample (80%) and a test sample (20%) for model validation. A cohort of assisted conceptions resulting from fresh embryo transfers was derived to evaluate the accuracy of both ultrasound and model-based gestational dating. Postnatal gestational age estimation models were developed with multivariable linear regression using elastic-net regularization. Gestational age estimates from dating ultrasound and from postnatal metabolomic models were compared with date of embryo transfer reference gestational age in the independent test cohorts. Accuracy was quantified by calculating mean absolute error and the square root of mean squared error.

RESULTS

Our model derivation cohort included 202,300 spontaneous conceptions, and the testing cohorts included 50,735 spontaneous conceptions and 1924 assisted conceptions. In the assisted conception cohort, first-trimester dating ultrasound was accurate to within approximately ±1.5 days compared with date of embryo transfer reference overall (mean absolute error, 0.21 [95% confidence interval, 0.20-0.23]). When compared with gestational age derived from date of embryo transfer, the metabolomic estimation models were accurate to within approximately ±5 days overall (0.79 [0.76-0.81] weeks). When ultrasound was used as the reference in validating the metabolomic model, the mean absolute error was slightly higher overall (0.81 [0.78-0.84] weeks). In general, the accuracy of gestational age estimates derived from ultrasound or metabolomic models was highest in term infants and lower in preterm and small-for-gestational-age newborns.

CONCLUSION

Our findings support the accuracy of ultrasound as a gestational age dating tool. They also support the potential utility of metabolic gestational age dating algorithms in settings where ultrasound or other accurate methods of estimating gestational age are not available because of lack of infrastructure or specialized training (eg, low-income countries). However, the accuracy of metabolomic model-based dating was generally lower than that of ultrasound.