Enhancing personalized gene expression prediction from DNA sequences using genomic foundation models.

Artificial intelligence (AI)/deep learning (DL) models that predict molecular phenotypes like gene expression directly from DNA sequences have recently emerged. While these models have proven effective at capturing the variation across genes, their ability to explain inter-individual differences has been limited. We hypothesize that the performance gap can be narrowed through the use of pre-trained embeddings from the Nucleotide Transformer, a large foundation model trained on 3,000+ genomes. We train a transformer model using the pre-trained embeddings and compare its predictive performance to Enformer, the current state-of-the-art model, using genotype and expression data from 290 individuals. Our model significantly outperforms Enformer in terms of correlation across individuals, and narrows the performance gap with an elastic net regression approach that uses just the genetic variants as predictors. Although simple regression models have their advantages in personalized prediction tasks, DL approaches based on foundation models pre-trained on diverse genomes have unique strengths in flexibility and interpretability. With further methodological and computational improvements with more training data, these models may eventually predict molecular phenotypes from DNA sequences with an accuracy surpassing that of regression-based approaches. Our work demonstrates the potential for large pre-trained AI/DL models to advance functional genomics.

A Penalized Regression Method for Genomic Prediction Reduces Mismatch between Training and Testing Sets.

Genomic selection (GS) is changing plant breeding by significantly reducing the resources needed for phenotyping. However, its accuracy can be compromised by mismatches between training and testing sets, which impact efficiency when the predictive model does not adequately reflect the genetic and environmental conditions of the target population. To address this challenge, this study introduces a straightforward method using binary-Lasso regression to estimate ß coefficients. In this approach, the response variable assigns 1 to testing set inputs and 0 to training set inputs. Subsequently, Lasso, Ridge, and Elastic Net regression models use the inverse of these ß coefficients (in absolute values) as weights during training (WLasso, WRidge, and WElastic Net). This weighting method gives less importance to features that discriminate more between training and testing sets. The effectiveness of this method is evaluated across six datasets, demonstrating consistent improvements in terms of the normalized root mean square error. Importantly, the model's implementation is facilitated using the glmnet library, which supports straightforward integration for weighting ß coefficients.

expHRD: an individualized, transcriptome-based prediction model for homologous recombination deficiency assessment in cancer.

BACKGROUND: Homologous recombination deficiency (HRD) stands as a clinical indicator for discerning responsive outcomes to platinum-based chemotherapy and poly ADP-ribose polymerase (PARP) inhibitors. One of the conventional approaches to HRD prognostication has generally centered on identifying deleterious mutations within the BRCA1/2 genes, along with quantifying the genomic scars, such as Genomic Instability Score (GIS) estimation with scarHRD. However, the scarHRD method has limitations in scenarios involving tumors bereft of corresponding germline data. Although several RNA-seq-based HRD prediction algorithms have been developed, they mainly support cohort-wise classification, thereby yielding HRD status without furnishing an analogous quantitative metric akin to scarHRD. This study introduces the expHRD method, which operates as a novel transcriptome-based framework tailored to n-of-1-style HRD scoring. RESULTS: The prediction model has been established using the elastic net regression method in the Cancer Genome Atlas (TCGA) pan-cancer training set. The bootstrap technique derived the HRD geneset for applying the expHRD calculation. The expHRD demonstrated a notable correlation with scarHRD and superior performance in predicting HRD-high samples. We also performed intra- and extra-cohort evaluations for clinical feasibility in the TCGA-OV and the Genomic Data Commons (GDC) ovarian cancer cohort, respectively. The innovative web service designed for ease of use is poised to extend the realms of HRD prediction across diverse malignancies, with ovarian cancer standing as an emblematic example. CONCLUSIONS: Our novel approach leverages the transcriptome data, enabling the prediction of HRD status with remarkable precision. This innovative method addresses the challenges associated with limited available data, opening new avenues for utilizing transcriptomics to inform clinical decisions.

Associations between exposure to pesticides mixture and semen quality among the non-occupationally exposed males: Four statistical models.

Most epidemiological studies on the associations between pesticides exposure and semen quality have been based on a single pesticide, with inconsistent major results. In contrast, there was limited human evidence on the potential effect of pesticides mixture on semen quality. Our study aimed to investigate the relationship of pesticide profiles with semen quality parameters among 299 non-occupationally exposed males aged 25-50 without any clinical abnormalities. Serum concentrations of 21 pesticides were quantified by gas chromatography-tandem mass spectrometry (GC-MS/MS). Semen quality parameters were abstracted from medical records. Generalized linear regression models (GLMs) and three mixture approaches, including weighted quantile sum regression (WQS), elastic net regression (ENR) and Bayesian kernel machine regression (BKMR), were applied to explore the single and mixed effects of pesticide exposure on semen quality. In GLMs, as the serum levels of Bendiocarb, ß-BHC, Clomazone, Dicrotophos, Dimethenamid, Paclobutrazole, Pentachloroaniline and Pyrimethanil increased, the straight-line velocity (VSL), linearity (LIN) and straightness (STR) decreased. This negative association also occurred between the concentration of ß-BHC, Pentachloroaniline, Pyrimethanil and progressive motility, total motility. In the WQS models, pesticides mixture was negatively associated with total motility and several sperm motility parameters (ß: -3.07∼-1.02 per decile, FDR-P<0.05). After screening the important pesticides derived from the mixture by ENR model, the BKMR models showed that the decreased qualities for VSL, LIN, and STR were also observed when pesticide mixtures were at ≥ 70th percentiles. Clomazone, Dimethenamid, and Pyrimethanil (Posterior inclusion probability, PIP: 0.2850-0.8900) were identified as relatively important contributors. The study provides evidence that exposure to single or mixed pesticide was associated with impaired semen quality.

An elastic net regression model for predicting the risk of ICU admission and death for hospitalized patients with COVID-19.

This study aimed to develop and validate prediction models to estimate the risk of death and intensive care unit admission in COVID-19 inpatients. All RT-PCR-confirmed adult COVID-19 inpatients admitted to Fujian Provincial Hospital from October 2022 to April 2023 were considered. Elastic Net Regression was used to derive the risk prediction models. Potential risk factors were considered, which included demographic characteristics, clinical symptoms, comorbidities, laboratory results, treatment process, prognosis. A total of 1906 inpatients were included finally by inclusion/exclusion criteria and were divided into derivation and test cohorts in a ratio of 8:2, where 1526 (80%) samples were used to develop prediction models under a repeated cross-validation framework and the remaining 380 (20%) samples were used for performance evaluation. Overall performance, discrimination and calibration were evaluated in the validation set and test cohort and quantified by accuracy, scaled Brier score (SbrS), the area under the ROC curve (AUROC), and Spiegelhalter-Z statistics. The models performed well, with high levels of discrimination (AUROCICU [95%CI]: 0.858 [0.803,0.899]; AUROCdeath [95%CI]: 0.906 [0.850,0.948]); and good calibrations (Spiegelhalter-ZICU: - 0.821 (p-value: 0.412); Spiegelhalter-Zdeath: 0.173) in the test set. We developed and validated prediction models to help clinicians identify high risk patients for death and ICU admission after COVID-19 infection.

The interaction effect of metals exposure and dietary habit on cognitive function in Chinese older adult cohort.

BACKGROUND: As the important factors in cognitive function, dietary habits and metal exposures are interactive with each other. However, fewer studies have investigated the interaction effect of them on cognitive dysfunction in older adults. METHODS: 2,445 registered citizens aged 60-85 years from 51 community health centers in Luohu District, Shenzhen, were recruited in this study based on the Chinese older adult cohort. All subjects underwent physical examination and Mini-cognitive assessment scale. A semi quantitative food frequency questionnaire was used to obtain their food intake frequency, and 21 metal concentrations in their urine were measured. RESULTS: Elastic-net regression model, a machine learning technique, identified six variables that were significantly associated with cognitive dysfunction in older adults. These variables included education level, gender, urinary concentration of arsenic (As) and cadmium (Cd), and the frequency of monthly intake of egg and bean products. After adjusting for multiple factors, As and Cd concentrations were positively associated with increased risk of mild cognitive impairment (MCI) in the older people, with OR values of 1.19 (95% CI: 1.05-1.42) and 1.32 (95% CI: 1.01-1.74), respectively. In addition, older adults with high frequency of egg intake (≥30 times/month) and bean products intake (≥8 times/month) had a reduced risk of MCI than those with low protein egg intake (<30 times/month) and low bean products intake (<8 times/month), respectively. Furthermore, additive interaction were observed between the As exposure and egg products intake, as well as bean products. Cd exposure also showed additive interactions with egg and bean products intake. CONCLUSIONS: The consumption of eggs and bean products, as well as the levels of exposure to the heavy metals Cd and As, have been shown to have a substantial influence on cognitive impairment in the elderly population.

An X Chromosome Transcriptome Wide Association Study Implicates ARMCX6 in Alzheimer's Disease.

Background: The X chromosome is often omitted in disease association studies despite containing thousands of genes that may provide insight into well-known sex differences in the risk of Alzheimer's disease (AD). Objective: To model the expression of X chromosome genes and evaluate their impact on AD risk in a sex-stratified manner. Methods: Using elastic net, we evaluated multiple modeling strategies in a set of 175 whole blood samples and 126 brain cortex samples, with whole genome sequencing and RNA-seq data. SNPs (MAF > 0.05) within the cis-regulatory window were used to train tissue-specific models of each gene. We apply the best models in both tissues to sex-stratified summary statistics from a meta-analysis of Alzheimer's Disease Genetics Consortium (ADGC) studies to identify AD-related genes on the X chromosome. Results: Across different model parameters, sample sex, and tissue types, we modeled the expression of 217 genes (95 genes in blood and 135 genes in brain cortex). The average model R2 was 0.12 (range from 0.03 to 0.34). We also compared sex-stratified and sex-combined models on the X chromosome. We further investigated genes that escaped X chromosome inactivation (XCI) to determine if their genetic regulation patterns were distinct. We found ten genes associated with AD at p < 0.05, with only ARMCX6 in female brain cortex (p = 0.008) nearing the significance threshold after adjusting for multiple testing (α = 0.002). Conclusions: We optimized the expression prediction of X chromosome genes, applied these models to sex-stratified AD GWAS summary statistics, and identified one putative AD risk gene, ARMCX6.

Depression and fatigue six months post-COVID-19 disease are associated with overlapping symptom constellations: A prospective, multi-center, population-based cohort study.

BACKGROUND: Depression and fatigue are commonly observed sequelae following viral diseases such as COVID-19. Identifying symptom constellations that differentially classify post-COVID depression and fatigue may be helpful to individualize treatment strategies. Here, we investigated whether self-reported post-COVID depression and post-COVID fatigue are associated with the same or different symptom constellations. METHODS: To address this question, we used data from COVIDOM, a population-based cohort study conducted as part of the NAPKON-POP platform. Data were collected in three different German regions (Kiel, Berlin, Würzburg). We analyzed data from >2000 individuals at least six months past a PCR-confirmed COVID-19 disease, using elastic net regression and cluster analysis. The regression model was developed in the Kiel data set, and externally validated using data sets from Berlin and Würzburg. RESULTS: Our results revealed that post-COVID depression and fatigue are associated with overlapping symptom constellations consisting of difficulties with daily activities, perceived health-related quality of life, chronic exhaustion, unrestful sleep, and impaired concentration. Confirming the overlap in symptom constellations, a follow-up cluster analysis could categorize individuals as scoring high or low on depression and fatigue but could not differentiate between both dimensions. LIMITATIONS: The data presented are cross-sectional, consisting primarily of self-reported questionnaire or medical records rather than biometric data. CONCLUSIONS: In summary, our results suggest a strong link between post-COVID depression and fatigue, highlighting the need for integrative treatment approaches.

A genomic predictor for age at sexual maturity for mammalian species.

Age at sexual maturity is a key life history trait that can be used to predict population growth rates and develop life history models. In many wild animal species, the age at sexual maturity is not accurately quantified. This results in a reduced ability to accurately model demography of wild populations. Recent studies have indicated the potential for CpG density within gene promoters to be predictive of other life history traits, specifically maximum lifespan. Here, we have developed a machine learning model using gene promoter CpG density to predict the mean age at sexual maturity in mammalian species. In total, 91 genomes were used to identify 101 unique gene promoters predictive of age at sexual maturity across males and females. We found these gene promoters to be most predictive of age at sexual maturity in females (R 2 = 0.881) compared to males (R 2 = 0.758). The median absolute error rate was also found to be lower in females (0.427 years) compared to males (0.785 years). This model provides a novel method for species-level age at sexual maturity prediction without the need for long-term monitoring. This study also highlights a potential epigenetic mechanism for the onset of sexual maturity, indicating the possibility of using epigenetic biomarkers for this important life history trait.

Tissue-specific RNA methylation prediction from gene expression data using sparse regression models.

N6-methyladenosine (m6A) is a highly prevalent and conserved post-transcriptional modification observed in mRNA and long non-coding RNA (lncRNA). Identifying potential m6A sites within RNA sequences is crucial for unraveling the potential influence of the epitranscriptome on biological processes. In this study, we introduce Exp2RM, a novel approach that formulates single-site-based tissue-specific elastic net models for predicting tissue-specific methylation levels utilizing gene expression data. The resulting ensemble model demonstrates robust predictive performance for tissue-specific methylation levels, with an average R-squared value of 0.496 and a median R-squared value of 0.482 across all 22 human tissues. Since methylation distribution varies among tissues, we trained the model to incorporate similar patterns, significantly improves accuracy with the median R-squared value increasing to 0.728. Additonally, functional analysis reveals Exp2RM's ability to capture coefficient genes in relevant biological processes. This study emphasizes the importance of tissue-specific methylation distribution in enhancing prediction accuracy and provides insights into the functional implications of methylation sites.

Spatial Relationships in the Tumor Microenvironment Demonstrate Association with Pathologic Response to Neoadjuvant Chemoimmunotherapy in Muscle-invasive Bladder Cancer.

BACKGROUND: Platinum-based neoadjuvant chemotherapy (NAC) is standard for patients with muscle-invasive bladder cancer (MIBC). Pathologic response (complete: ypT0N0 and partial:

A hybrid of regularization method and generalized path analysis: modeling single-vehicle run-off-road crashes in a cross-sectional study.

BACKGROUND: Determining risk factors of single-vehicle run-off-road (SV-ROR) crashes, as a significant number of all the single-vehicle crashes and all the fatalities, may provide infrastructure for quicker and more effective safety measures to explore the influencing and moderating variables in SV-ROR. Therefore, this paper emphasizes utilizing a hybrid of regularization method and generalized path analysis for studying SV-ROR crashes to identify variables influencing their happening and severity. METHODS: This cross-sectional study investigated 724 highway SV-ROR crashes from 2015 to 2016. To drive the key variables influencing SV-ROR crashes Ridge, Least Absolute Shrinkage and Selection Operator (Lasso), and Elastic net regularization methods were implemented. The goodness of fit of utilized methods in a testing sample was assessed using the deviance and deviance ratio. A hybrid of Lasso regression (LR) and generalized path analysis (gPath) was used to detect the cause and mediators of SV-ROR crashes. RESULTS: Findings indicated that the final modified model fitted the data accurately with [Formula: see text]= 16.09, P < .001, [Formula: see text]/ degrees of freedom = 5.36 > 5, CFI = .94 > .9, TLI = .71 < .9, RMSEA = 1.00 > .08 (90% CI = (.06 to .15)). Also, the presence of passenger (odds ratio (OR) = 2.31, 95% CI = (1.73 to 3.06)), collision type (OR = 1.21, 95% CI = (1.07 to 1.37)), driver misconduct (OR = 1.54, 95% CI = (1.32 to 1.79)) and vehicle age (OR = 2.08, 95% CI = (1.77 to 2.46)) were significant cause of fatality outcome. The proposed causal model identified collision type and driver misconduct as mediators. CONCLUSIONS: The proposed HLR-gPath model can be considered a useful theoretical structure to describe how the presence of passenger, collision type, driver misconduct, and vehicle age can both predict and mediate fatality among SV-ROR crashes. While notable progress has been made in implementing road safety measures, it is essential to emphasize that operative preventative measures still remain the most effective approach for reducing the burden of crashes, considering the critical components identified in this study.

Resting-State Functional Connectivity Explained Psychotic-like Experiences in the General Population and Partially Generalized to Patients and Relatives.

Background: Psychotic-like experiences (PLEs) are considered the subclinical portion of the psychosis continuum. Research suggests that there are resting-state functional connectivity (rsFC) substrates of PLEs, yet it is unclear if the same substrates underlie more severe psychosis. Here, to our knowledge, we report the first study to build a cross-validated rsFC model of PLEs in a large community sample and directly test its ability to explain psychosis in an independent sample of patients with psychosis and their relatives. Methods: Resting-state FC of 855 healthy young adults from the WU-Minn Human Connectome Project (HCP) was used to predict PLEs with elastic net. An rsFC composite score based on the resulting model was correlated with psychotic traits and symptoms in 118 patients with psychosis, 71 nonpsychotic first-degree relatives, and 45 healthy control subjects from the psychosis HCP. Results: In the HCP, the cross-validated model explained 3.3% of variance in PLEs. Predictive connections spread primarily across the default, frontoparietal, cingulo-opercular, and dorsal attention networks. The model partially generalized to a younger, but not older, subsample in the psychosis HCP, explaining two measures of positive/disorganized psychotic traits (the Structured Interview for Schizotypy: ß = 0.25, pone-tailed = .027; the Schizotypy Personality Questionnaire positive factor: ß = 0.14, pone-tailed = .041). However, it did not differentiate patients from relatives and control subjects or explain psychotic symptoms in patients. Conclusions: Some rsFC substrates of PLEs are shared across the psychosis continuum. However, explanatory power was modest, and generalization was partial. It is equally important to understand shared versus distinct rsFC variances across the psychosis continuum.

Associations between multiple metals during early pregnancy and gestational diabetes mellitus under four statistical models.

Gestational diabetes mellitus (GDM) is one of the most common complications of pregnancy. Metal exposure is an emerging factor affecting the risk of GDM. However, the effects of metal mixture on GDM and key metals within the mixture remain unclear. This study was aimed at investigating the association between metal mixture during early pregnancy and the risk of GDM using four statistical methods and further at identifying the key metals within the mixture associated with GDM. A nested case-control study including 128 GDM cases and 318 controls was conducted in Beijing, China. Urine samples were collected before 13 gestational weeks and the concentrations of 13 metals were measured. Single-metal analysis (unconditional logistic regression) and mixture analyses (Bayesian kernel machine regression (BKMR), quantile g-computation, and elastic-net regression (ENET) models) were applied to estimate the associations between exposure to multiple metals and GDM. Single-metal analysis showed that Ni was associated with lower risk of GDM, while positive associations of Sr and Sb with GDM were observed. Compared with the lowest quartile of Ni, the ORs of GDM in the highest quartiles were 0.49 (95% CI 0.24, 0.98). In mixture analyses, Ni and Mg showed negative associations with GDM, while Co and Sb were positively associated with GDM in BKMR and quantile g-computation models. No significant joint effect of metal mixture on GDM was observed. However, interestingly, Ni was identified as a key metal within the mixture associated with decreased risk of GDM by all three mixture methods. Our study emphasized that metal exposure during early pregnancy was associated with GDM, and Ni might have important association with decreased GDM risk.

Two sides of the same coin: distinct neuroanatomical patterns predict crystallized and fluid intelligence in adults.

Background: Crystallized intelligence (Gc) and fluid intelligence (Gf) are regarded as distinct intelligence components that statistically correlate with each other. However, the distinct neuroanatomical signatures of Gc and Gf in adults remain contentious. Methods: Machine learning cross-validated elastic net regression models were performed on the Human Connectome Project Young Adult dataset (N = 1089) to characterize the neuroanatomical patterns of structural magnetic resonance imaging variables that are associated with Gc and Gf. The observed relationships were further examined by linear mixed-effects models. Finally, intraclass correlations were computed to examine the similarity of the neuroanatomical correlates between Gc and Gf. Results: The results revealed distinct multi-region neuroanatomical patterns predicted Gc and Gf, respectively, which were robust in a held-out test set (R2 = 2.40, 1.97%, respectively). The relationship of these regions with Gc and Gf was further supported by the univariate linear mixed effects models. Besides that, Gc and Gf displayed poor neuroanatomical similarity. Conclusion: These findings provided evidence that distinct machine learning-derived neuroanatomical patterns could predict Gc and Gf in healthy adults, highlighting differential neuroanatomical signatures of different aspects of intelligence.

Factors associated with heartworm preventative use in the golden retriever lifetime study.

Introduction: Heartworm disease is preventable with use of heartworm preventatives, but the reported prevalence of heartworm preventative use in the United States is low, some estimates falling around 50% of dogs. However, there are very few estimates of prevalence and its associated factors. Methods: We aimed to estimate prevalence and evaluate factors, including vaccination status, demographics, lifestyle, physical conditions, medications and supplements, and environment and living conditions, for their association with heartworm preventative use in a large dataset from the Golden Retriever Lifetime Study (N = 2,998). Due to the large number of predictors evaluated, we built a bootstrapped elastic net logistic regression model, which is robust to overfitting and multicollinearity. Variables were evaluated by calculating covariate stability (>80%) and statistical significance (p<0.02). Results: In our sample, the prevalence of heartworm use was 39.5%. In our elastic net model, receiving vaccinations (rabies, Bordetella, or any other vaccine), being located in the Southern U.S., being altered, having an infectious disease or ear/ nose/throat system disease diagnosis, being on heartworm preventatives in the past, currently being on tick preventative, having sun exposure in an area with concrete flooring, living in a house with more rooms with carpeted floors, and spending time on hardwood flooring inside were associated with greater odds of heartworm preventative use. Supplementation use and being in the top quartile of height were associated with lower odds of heartworm preventative use. Discussion: The explanatory factors we identified can be used to improve client communication. In addition, target populations for educational interventions and outreach can be identified. Future studies can validate the findings in a more diverse population of dogs.

An X Chromosome Transcriptome Wide Association Study Implicates ARMCX6 in Alzheimer's Disease.

Background: The X chromosome is often omitted in disease association studies despite containing thousands of genes which may provide insight into well-known sex differences in the risk of Alzheimer's Disease. Objective: To model the expression of X chromosome genes and evaluate their impact on Alzheimer's Disease risk in a sex-stratified manner. Methods: Using elastic net, we evaluated multiple modeling strategies in a set of 175 whole blood samples and 126 brain cortex samples, with whole genome sequencing and RNA-seq data. SNPs (MAF>0.05) within the cis-regulatory window were used to train tissue-specific models of each gene. We apply the best models in both tissues to sex-stratified summary statistics from a meta-analysis of Alzheimer's disease Genetics Consortium (ADGC) studies to identify AD-related genes on the X chromosome. Results: Across different model parameters, sample sex, and tissue types, we modeled the expression of 217 genes (95 genes in blood and 135 genes in brain cortex). The average model R2 was 0.12 (range from 0.03 to 0.34). We also compared sex-stratified and sex-combined models on the X chromosome. We further investigated genes that escaped X chromosome inactivation (XCI) to determine if their genetic regulation patterns were distinct. We found ten genes associated with AD at p 0.05, with only ARMCX6 in female brain cortex (p = 0.008) nearing the significance threshold after adjusting for multiple testing (α = 0.002). Conclusions: We optimized the expression prediction of X chromosome genes, applied these models to sex-stratified AD GWAS summary statistics, and identified one putative AD risk gene, ARMCX6.

Face recognition ability can be predicted by microstructural properties of white matter: a study of diffusion tensor imaging (DTI).

There is a great individual difference in people's face recognition ability (FRA). This study aimed to reveal the neural mechanism underlying such individual differences. Elastic-net regression models were constructed to predict FRA based on the white matter (WM) microstructural properties. We found that FRA can be accurately predicted by the WM microstructural properties. For the right inferior longitudinal fasciculus (ILF) and bilateral arcuate fasciculus (AF), FRA was correlated negatively to fractional anisotropy (FA), but positively to radial diffusivity (RD). In contrast, for the corpus callosum forceps minor (CFM), FRA was correlated positively to FA, but negatively to RD. Such various patterns of the WM microstructural properties suggested a positive correlation between FRA and fiber diameter for the right ILF and bilateral AF, but a negative correlation between FRA and diameter of the CFM. These findings reflected that FRA was correlated positively to connectivities of the right ILF and bilateral AF, but negatively to those of the CFM. These findings not only confirmed the significant role of the right ILF in face recognition, but also revealed the involvement of the bilateral AF and CFM in face recognition, particularly implying the important role of hemisphere lateralization modulated by transcallosal connectivity in face recognition.

Associations between exposure to a mixture of perfluoroalkyl and polyfluoroalkyl substances and age at menarche in adolescent girls utilizing three statistical models.

Exposure to perfluoroalkyl and polyfluoroalkyl substances (PFAS) is suggested to interfere with endocrine function and may affect female pubertal development. However, the epidemiological evidence on age at menarche associated with PFAS exposure is still inconsistent. Our objective was to investigate association of serum PFAS concentrations with age at menarche among 12-19 years old girls. We used data on 432 girls from National Health and Nutrition Examination Survey (NHANES) 2007-2012 cycles. NHANES reported serum concentrations of perfluorooctanoic acid (PFOA), perfluorooctanesulfonic acid (PFOS), perfluorohexane sulfonate (PFHxS), perfluorononanoic acid (PFNA) and perfluorodecanoic acid (PFDA) as quantified by liquid chromatography tandem mass spectrometry (LC-MS/MS). Age at menarche was self-reported by girls or their guardians. Multivariable linear regression models were applied to estimate the association of individual PFAS exposure with age at menarche. The combined effects of PFAS mixture exposures on age at menarche were assessed using three statistical methods, including Bayesian kernel machine regression (BKMR), weighted quantile sum regression (WQS), and elastic net regression (ENR). In the single-chemical model, girls in the middle tertile of serum PFOA concentration had a lower age at menarche [regression coefficient (ß) = -0.73 years, 95% confidence interval (CI): -1.44, -0.01; P = 0.047], compared with those in the lower tertile. Girls in the higher tertile of PFNA exposure were associated with older age at menarche (ß = 0.36 years, 95% CI: 0.03, 0.80; P = 0.027), compared with those in the lower tertile. In the multiple-chemical models using BKMR and ENR approaches, higher PFNA exposure was significantly associated with older age at menarche among girls, after adjusting for other PFAS. We found suggestive evidence that higher PFAS mixture exposures may be related to an increase in age at menarche using the BKMR model. In conclusion, exposure to PFNA was associated with the later timing of menarche among girls.

A machine learning-derived neuroanatomical pattern predicts delayed reward discounting in the Human Connectome Project Young Adult sample.

Delayed reward discounting (DRD) is defined as the extent to which person favors smaller rewards that are immediately available over larger rewards available in the future. Higher levels of DRD have been identified in individuals with a wide range of clinical disorders. Although there have been studies adopting larger samples and using only gray matter volume to characterize the neuroanatomical correlates of DRD, it is still unclear whether previously identified relationships are generalizable (out-of-sample) and how cortical thickness and cortical surface area contribute to DRD. In this study, using the Human Connectome Project Young Adult dataset (N = 1038), a machine learning cross-validated elastic net regression approach was used to characterize the neuroanatomical pattern of structural magnetic resonance imaging variables associated with DRD. The results revealed a multi-region neuroanatomical pattern predicted DRD and this was robust in a held-out test set (morphometry-only R2 = 3.34%, morphometry + demographics R2  = 6.96%). The neuroanatomical pattern included regions implicated in the default mode network, executive control network, and salience network. The relationship of these regions with DRD was further supported by univariate linear mixed effects modeling results, in which many of the regions identified as part of this pattern showed significant univariate associations with DRD. Taken together, these findings provide evidence that a machine learning-derived neuroanatomical pattern encompassing various theoretically relevant brain networks produces robustly predicts DRD in a large sample of healthy young adults.