Chirag J. Patel

Univariate differential expression analysis dominates single-cell transcriptomic studies of Alzheimer’s disease, yet marginal testing alone cannot resolve multivariate disease-associated structure or distinguish clinical from neuropathological signatures. Here we introduce TriSCOPE, a framework integrating multivariate predictive modeling, donor-aware differential expression, and pseudotime inference, applied to six Alzheimer’s disease snRNA-seq datasets comprising ~4.2 million nuclei from 627 donors. Across six major brain cell types and 41 subclusters, we identify cell-type-specific predictively expressed genes that are largely non-overlapping with differentially expressed genes, revealing disease-relevant signals invisible to univariate approaches. Conditional GWAS–eQTL colocalization anchors a subset of these predictively expressed genes to inherited Alzheimer’s disease risk loci, including 27 genetically supported gene–cell type pairs (among them ARL17B , JAZF1 ) with no significant differential expression. Integration across all three analytical branches nominates high-confidence convergent programs including astrocytic stress markers (such as GFAP , MT2A ), neuronal signaling genes (like RASGEF1B , LINGO1 , NRG1 ), and microglial and oligodendrocyte regulators (including SYTL3 , SPP1 , APOD ). Comparison of clinical and neuropathological models reveals cell-type-specific divergence between cognitive and pathological Alzheimer’s disease signatures. Key predictively expressed genes including ARL17B , LINGO1 , and APOE replicate across independent cohorts, with cross-cohort generalizability confirmed across all three analytical branches of TriSCOPE.

INTRODUCTION: Environmental exposures contribute substantially to Alzheimer's disease and related dementias (ADRD), yet comprehensive exposome‐wide investigations systematically evaluating multi‐domain environmental factors remain limited and lacking utility for precision medicine. METHODS: We conducted an exposome‐wide association study (ExWAS) among 499,992 UK Biobank participants, examining 193 exposures across five domains (social, physical, chemical, lifestyle, and ecosystems). Cox proportional hazard models with Bonferroni correction identified ADRD‐associated exposures. We used the XGBoost classifier to build a predictive model and further developed an exposomic risk score (ERS) for risk stratification. RESULTS: During a 15.3‐year median follow‐up, 8881 participants (1.77%) developed ADRD, including 4000 with Alzheimer's disease. ExWAS identified 76 exposures after Bonferroni correction. The prediction model achieved area under the curve (AUC) = 0.718, outperforming demographics+apolipoprotein E ( APOE ) status alone (AUC = 0.678). The ERS stratified ADRD risk (highest vs lowest quartile hazard ratio [HR] = 2.97, 95% confidence interval [CI]: 2.59–3.41). DISCUSSION: The ERS has the potential to contribute to clinically relevant prevention strategies for ADRD.

Importance: A 2025 systematic review concluded there was "strong evidence of a likely relationship" between prenatal acetaminophen exposure and neurodevelopmental disorders (NDDs), directly informing a federal health advisory, based on qualitative synthesis without quantitative pooling or bias correction. Objective: To determine, via meta-analysis, whether the association claimed between prenatal acetaminophen exposure and attention-deficit/hyperactivity disorder (ADHD), autism spectrum disorder (ASD), and other NDDs survives publication-bias correction and sensitivity methods applied to the same underlying studies. Data Sources: The 46 studies identified in a 2025 Navigation Guide systematic review, based on a PubMed search (through February 25, 2025) supplemented by Web of Science and Google Scholar searches that yielded no additional studies; no additional search was performed for this reanalysis. Study Selection: Of the 46 studies, 24 (52%) reported a ratio measure (hazard ratio, odds ratio, relative risk, or incidence rate ratio) and were retained; studies reporting only continuous outcomes were excluded. Data Extraction and Synthesis: Estimates were collapsed to 1 per study, outcome group, and design to correct a unit-of-analysis error in the original review. Random-effects meta-analysis was used. We assessed eight bias-correction methods, a credibility ceiling analysis, and 5 quality-filtered subsets, then repeated the analyses after harmonizing odds and hazard ratios onto a common risk ratio scale. Main Outcomes and Measures: Pooled ratio and 95% CI for ADHD, ASD, and other NDDs under each method, before and after harmonization. Results: Among 30 estimates from 24 studies (sample sizes, 307-2,480,797), the ASD association lost statistical significance (pooled ratio, 1.08; 95% CI, 0.97-1.19) from corrected weighting alone, before any further bias correction. The ADHD association remained positive under most methods but lost statistical significance under a 15% credibility ceiling. The other NDD association was not robust under bias correction, reversing sign under 1 method and failing a test for evidential value (p-curve, P = .71). Conclusions and Relevance: The claim of "strong evidence" for an acetaminophen-NDD association was not supported by meta-analysis of the same studies. A modest, method-dependent ADHD signal persisted with most methods, but the ASD and other-NDD associations did not survive bias-correction and sensitivity methods. Key Points: Question: Does the "strong evidence" verdict linking prenatal acetaminophen exposure to attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD) survive standard meta-analytic and bias-correction methods applied to the same underlying studies? Findings: In this meta-analysis of 30 study-level estimates from a 2025 systematic review, ¹ the ASD association was not significant after corrected study weighting alone, and the association with other neurodevelopmental disorders was not robust under scrutiny. The ADHD association survived most, but not all, of eight publication-bias-correction methods and was not statistically significant under a modest 15% credibility ceiling. Meaning: These findings suggest that the original review’s claim of strong evidence linking prenatal acetaminophen use to neurodevelopmental disorders is not supported by quantitative reanalysis of the same underlying studies.

Identifying small, interpretable gene sets that robustly capture disease-associated variation in singlecell transcriptomic data remains a central challenge for biological interpretation and experimental followup. In practice, commonly used differential expression and sparsity-based approaches often produce large, unstable gene lists that fail to generalize across patients due to strong donor-specific confounding. We study sparse gene selection for reconstructing donor-robust, disease-aligned cellular trajectories in real single-cell RNA-seq datasets. We introduce Sparse Linear Manifold Control (SLMC), a practical workflow that defines a disease-aligned score after removing donor-associated variation and selects minimal gene programs whose expression reconstructs this score. We focus on diagnosing the structure of the resulting reconstruction objective and evaluating selection strategies under realistic health data conditions. Across five human single-cell datasets spanning oncology and neurodegeneration, we find that the reconstruction objective exhibits strong diminishing returns, explaining why simple greedy selection methods perform well in practice. Under strict donor-heldout evaluation, greedy methods consistently outperform LASSO at small gene budgets and achieve accurate reconstruction with as few as 25 genes. Together, these results highlight how careful objective design and empirical evaluation enable robust and interpretable gene selection for disease-aligned representation learning in single-cell health data. Data and Code Availability: All code used for data preprocessing, model training, evaluation, and feature importance analyses is available at: https://github.com/AdiVM/SLMC_single-cell . The singlecell data used in this study are publicly available human transcriptomic datasets generated by prior studies and accessible through the Gene Expression Omnibus (GEO). Analyses were performed using renal cell carcinoma single-cell RNA-seq data (GEO accession: GSE314072 ) and Alzheimer’s disease single-nucleus RNA-seq data from human cortex (GEO accession: GSE138852 ), with additional publicly available datasets used for cross-context diagnostic evaluation (GEO accessions: GSM8652069 , GSE308624 , and GSE227734 ). All datasets contain de-identified human samples and are available under standard publicuse terms via GEO. Institutional Review Board (IRB): This study analyzes de-identified, publicly available human transcriptomic data obtained from previously published studies. No new data were collected, and no identifiable private information was accessed. In accordance with institutional policy, this work was determined to constitute non-human subjects research and did not require additional IRB approval.

Genetic risk, as well as the physical environment and psychosocial factors, affect susceptibility to common, complex diseases. Although portions of patients' social and exposure histories are captured during routine healthcare, these data are rarely collected in a standardized, quantitative format that supports cross-disease risk stratification. We analyzed data from the Personalized Environment and Genes Study (PEGS) for 2,344 participants with genome sequencing who completed three exposome surveys. We derived polyexposure (PXS) and polysocial (PSS) scores from 489 variables and compared disease-classification performance of these scores, polygenic scores (PGS), and classification based on age/sex across nine self‑reported, ever-diagnosed diseases. For cross-sectional classification of disease risk, PXS had greater improvement than PGS or PSS compared to models with common clinical factors and demographic variables such as sex and age. The combined PXS+PSS+PGS model had the best performance overall (mean ΔAUC 0.10; maximum ΔAUC 0.27 for asthma). Importantly, the score types identified different individuals in high-score groups, and there was minimal overlap among these groups (PGS-PXS 0.94%; PGS-PSS 1.0%; triple overlap 0.18%, averaged across diseases). Further, PGS-PXS and PGS-PSS correlations were weak (mean r=0.04 and 0.03). This independence may enable identification of different individuals at risk of disease compared to existing clinical assessments and could help tailor interventions and care. As a next step in determining clinical feasibility, we propose a 66‑question minimum exposure question set, with 61 questions based on PXS and five on PSS, as a pragmatic framework for clinical collection of standardized data for multiple diseases.

Diet is a leading modifiable contributor to the global burden of disease, yet its effects are often communicated in metrics that are difficult to interpret; expressing dietary impacts as years of life gained or lost can make epidemiological evidence more transparent and actionable for policy and practice. Here, using 30 years of NHANES data linked to the National Death Index (n=44,666), we modeled associations between multiple dietary exposures and all-cause mortality using a penalized accelerated failure time framework, achieving strong discrimination for predicting age at death (C-index=0.78, 95% CI 0.77-0.79). Across all exposures, shifting from an unfavorable dietary profile to an optimized one is associated with a maximum predicted difference in lifespan of 12.2 years. The largest predicted gains are related to higher intakes of nuts and seeds, legumes, vegetables and fruits, and lower intakes of processed and red meats, dairy and poultry. We also introduce a raindrop plot visualization that integrates linear and non-linear intake-lifespan relationships in units of life expectancy. Our framework translates complex diet-mortality associations into interpretable, visual estimates of changes in life expectancy, providing a practical tool for clearer risk communication in research, guidelines and public health decision-making.

Artificial intelligence (AI) in environmental health science is revolutionizing data analysis and problem‐solving approaches. These technologies facilitate the prediction of environmental exposures and disease outcomes and enable the identification of causal relationships for subsequent hypothesis testing. AI techniques improve pollution research through the analysis of satellite imagery and the modeling of pollutant dispersion, while AI advances chemical safety evaluations in toxicology by examining extensive datasets. AI is instrumental in addressing pressing environmental challenges, including remediation of polluted sites and ensuring equitable healthcare applications to mitigate biases. The expanding availability of large‐scale environmental, geospatial, and health outcome databases offers unprecedented opportunities for innovative applications. Their predictive capabilities are essential in disaster management, enabling real‐time analysis and optimizing resource deployment amid climate‐related crises. AI‐driven approaches play a critical role in carbon capture and waste management efforts aimed at reducing environmental impact. Furthermore, AI can elucidate complex relationships between the exposome—defined as the totality of exposures throughout an individual's life—and health outcomes, facilitating preventative strategies. This review examines the capabilities and limitations of AI in environmental health and safety, providing insights into its judicious and effective use for environmental management and healthcare.

Background: It is well-established that males have a higher mortality risk than females. Immune cells and their function are known to undergo characteristic changes during aging, and immune cells are known to have sex differences. Immune cells and their function have been linked to mortality risk, but no studies have investigated to what degree, if at all, Immune Cell Biomarkers (ICBs) contribute to the known differences in mortality risk by sex. Methods: Using participant data from the Health and Retirement Study (n = 8,822), we applied multivariable linear regressions adjusting for age, cytomegalovirus (CMV) serostatus, sex, and race/ethnicity to identify differences by sex in 48 immune cell biomarker (“ICB”, e.g. T cells, B cells, Monocytes, etc.) percentages and counts (measured in 2016). We studied how the associations between ICBs and mortality risk differ by sex using stratified Cox Proportional Hazard (CPH) models. We estimated how inclusion of sex explained the relationship between ICBs and all-cause mortality, and conversely, how inclusion of individual and all ICBs combined explain the relationship between sex and all-cause mortality using multivariable modeling approaches. Results: Differences in ICBs by sex range between 2-38% (39/48 statistically significant). 9 ICBs were significantly associated with mortality risk in the entire sample. While different ICBs were significantly associated with mortality risk in the stratified analyses, particularly with respect to monocyte, B cell, and NK cell populations, adjusting for sex modestly influenced the hazard ratios of the ICBs (sex: 8 ICBs, percent change <5.4%). Furthermore, individual and cumulative contributions of ICBs in explaining the differences in mortality risk by sex were not significant. Conclusion: ICBs vary by sex in the older population; however, their role in explaining mortality risk by sex are modest.