Cell Genomics

Tandem repeats (TRs) are implicated in over 70 Mendelian disorders and likely contribute to the "missing heritability" of complex traits and diseases, yet TR variations in Asian populations remain poorly characterized. Here, we constructed an Asian-specific SG10K-TR catalog by leveraging the SG10K_Health Dataset, comprising 916,274 autosomal TR loci genotyped in 9,490 individuals of Chinese (5,528), Malay (1,824), Indian (2,108), and other ancestries (30). Using a novel integrative measure for both repeat length and frequency variations, TRDDS, we found that population-level TR variations are selectively constrained in coding and promoter regions, whereas the enrichment of TRs with high population diversity was observed in regulatory sites with low chromatin accessibility and pathways related to neuronal functions. We also identified candidate TRs under selection that predominantly targets neuronal and synaptic architecture. Analysis of linkage disequilibrium (LD) patterns revealed that TRs are often poorly tagged by small variants, although we identified 123 candidate functional TRs that may underlie association signals previously attributed to nearby noncoding SNPs. Finally, TR-based GWAS of six anthropometric and lipid traits identified ten loci with genome-wide significant associations, including two novel loci for BMI (LINC02817) and height (UNC45B), and a TR variant as causal candidate for a known GWAS locus at HMGCR for LDL. Together, this study establishes a critical Asian-specific TR resource and highlights the fundamental role of TR diversity in driving evolutionary neuroplasticity and shaping the genetic architecture of complex traits.

Circulating plasma proteins are key biomarkers and therapeutic targets, now measurable at scale through high-throughput technologies, yet whether expanding proteomics platforms beyond the classical plasma secretome enhances genetic discovery and causal inference remains poorly understood. Here, we use an expanded SomaScan 7k platform to map the genetic architecture of a broader segment of the plasma proteome and to evaluate how proteome expansion affects pQTL discovery, causal inference and therapeutic target prioritisation. After quality control, we analysed 7,144 aptamers targeting 6,267 proteins in the harmonised dataset of two European cohorts: INTERVAL (n = 9,251 participants) and CHRIS (n = 4,194), and conducted genome-wide pQTL association analyses followed by meta-analysis. We identified 7,870 significant pQTLs (P-value < 1.26 x 10E-11; 1,784 cis, 6,086 trans), of which 2,704 (34%) associations were not reported in five prior large-scale pQTL studies. Newly assessed proteins, which accounted for 53% (1,422/2,704) of the novel associations, were less likely to harbour cis-pQTLs associations (15%) than those in the previous platform version (28%), consistent with their lower expected plasma concentrations and predominantly intracellular localisation. Colocalization analyses revealed widespread sharing of genetic signals across proteins and characterised 22 pleiotropic trans-regulatory hotspots accounting for 68% of all trans-pQTLs. Through two-sample Mendelian randomization analyses on 2,003 phenotypes from the Million Veteran Program, UK Biobank, and FinnGen (combined N > 1.2 million), we identified 6,340 genetically supported protein-trait associations, highlighting disease mechanisms and potential therapeutic opportunities beyond currently drug-targeted circulating proteins. Together, these findings provide a systematic view of the genetic architecture of the expanded plasma proteome and demonstrate that plasma proteome expansion reveals genetically anchored disease biology beyond the classical secretome, while exposing inherent biological and technical constraints of studying low-abundance intracellular proteins in circulation.

Genetic prediction of complex phenotypes typically relies on additive linear models, which scale well but cannot capture non-additive effects or deeply integrate molecular and clinical data. Domain-specific neural networks have driven advances in images, text, and other modalities, but genome-scale neural networks remain challenging because genotypes are sparse and high-dimensional, effective sample sizes are limited, and generic architectures lack interpretability. Here, we introduce the omnigenic neural network, a biologically structured architecture inspired by the omnigenic model of complex traits. The model learns hierarchical representations of biological processes, accommodates multimodal inputs, supports transfer learning, and enables multitask prediction. Models trained in the UK Biobank and evaluated in the All of Us cohort for ischemic heart disease, type 2 diabetes, and schizophrenia outperformed published PGS Catalog and PRS-CSx scores. A multitask model trained across 36 cardiovascular endpoints further outperformed corresponding single-phenotype models and baselines. The architecture provides systems-level interpretability by quantifying the contributions of biological processes, which were consistent with established disease mechanisms. It also captures non-linear interactions between variants. Analysis of these interactions using Integrated Hessians revealed patterns concordant with previously reported epistatic associations. Together, these findings establish the omnigenic neural network as a flexible framework for interpretable, multimodal, and multitask genomic prediction.

Human immune responses remain poorly characterized across Indigenous South American (ISA) populations, limiting our understanding of human immune variation. Here, we generated a single-cell transcriptional atlas of 1.4 million peripheral blood mononuclear cells from 345 individuals representing 13 South American populations, profiled at baseline and following stimulation with lipopolysaccharide, influenza A virus, or Trypanosoma cruzi. Genetic ancestry was associated with the expression of a large fraction of the transcriptome, often only after stimulation, and explained more variation than age or sex. On average, cis-eQTLs accounted for one-quarter of these ancestry-associated differences. Thousands of regulatory variants were common in ISA populations but rare or absent in European reference datasets. Integrating eQTLs with GWAS revealed stimulation-specific disease-associated regulatory variants and an enrichment of ancestry-associated expression differences among genes with colocalizing signals. Our findings highlight the importance of studying diverse populations across immune contexts to understand the genetic architecture of human immune responses.

Pregnancy involves profound changes to maternal immunity throughout gestation to simultaneously tolerate the fetus and protect both mother and baby from infection. Dysregulation of this process can lead to pregnancy complications and adverse outcomes. However, how the maternal peripheral immune system adapts throughout an uncomplicated pregnancy remains poorly understood, limiting our ability to identify and interpret alterations associated with pregnancy complications. Here, we longitudinally profiled maternal whole-blood transcriptomes of 1,305 nulliparous women at 12, 20, 28 and 36 weeks of gestation (n=3,776 samples) from the Pregnancy Outcome Prediction Study 2 (POPs2). Differential expression analysis revealed that the maternal immune system underwent its most pronounced remodeling in early pregnancy (between 12 and 20 weeks gestation). Co-expressed gene modules demonstrated immune-dominant programs were highly active in early stages, shifting towards increased cellular and metabolic activity in later pregnancy, reflecting maternal adaptation to increasing physiological demands. Through cis-eQTL mapping we found that genetic regulatory effects increasingly diverged from the non-pregnant state as pregnancy advanced. Colocalising with GWAS revealed putative mechanisms underlying genetic predisposition to pregnancy complications, with 58% of these colocalizations detectable only in the context of pregnancy. Finally, we identified eQTL that exhibit significant interactions with gestational age, implicating dynamic regulation of innate immune and metabolic pathways. These findings show that pregnancy is not only a state of changing gene expression, but also a dynamic regulatory environment that reshapes the effects of genetic variation. By characterizing these adaptations across gestation, this work provides a reference for interpreting maternal immune variation and genetic risk across gestation.

East Asian populations, representing over 20% of the global population, remain critically underrepresented in human genomic studies, limiting our understanding of population-stratified genetic variation and its implications for health and disease. Here we present the first phase of the Asian Pan-Genome project (APG), comprising 320 nearly complete, fully phased haploid genome assemblies from 160 East Asian individuals. These assemblies achieve unprecedented quality, with an average contig N50 of 144.3 megabase pairs and an average quality value of 64.5. Leveraging these superior assemblies, we reveal previously uncharacterized diversity in human repeatome, including population-stratified patterns in centromere satellites and rDNA arrays. Compared to existing global human genome assemblies, the newly generated genomes supplement 152 million base pairs of novel sequences, 355 gene gains, 18,300 structural variation loci and 26 large euchromatic inversions missing from current human pangenomes. We perform population stratification analyses of structural variations, and further resolve the structural haplotypes of complex genomic regions such as Major Histocompatibility Complex and Survival Motor Neuron loci across global pangenomes, exemplifying tandem-duplicate and inversion-rich complex locus architectures in the human genome, respectively. This resource provides a critical foundation for human genetic studies, especially for East Asian populations, promoting more accurate variant discovery, reducing bias, and ultimately advancing the equity and efficacy of genomic medicine.

Foundation models trained on electronic healthcare records (EHRs) have gained traction with the aim to transform personalised medicine. However, their interpretability is bound to redescribing the records the models were trained on, missing implicitly learned concepts and biases. Here, we show that human genetics provides an orthogonal layer to surface implicitly learned biological concepts and otherwise hidden risk factors. Re-implementing the generative transformer Delphi-2M in >500,000 UK Biobank participants, we performed genome-wide association testing on its 120 learned embeddings and identified 434 genome-wide-significant signals across 151 independent loci and 98 embeddings, revealing a heritable structure that feature-attribution methods cannot recover. Effector-gene mapping implicated cholesterol metabolism and an IL-1-family epithelial-alarmin pathway, supported by strong (>50-fold) enrichment for variants previously associated with blood lipids, body-mass index, and asthma. Loci recovered the targets of essentially all approved lipid-lowering and severe-asthma therapies, and another twelve drugs not obvious from genetic results based on single ICD-10 GWAS. Yet, embeddings poorly explained variation in pleiotropic risk factors, while still retaining most of their predictive value. Substantial improvements in predictive performance were hence confined to a minority of common diseases by adding specific diagnostic or organ-derived markers. Our findings suggest that human genetics might be most powerful as an orthogonal explanatory or regularising layer to train the next generation of EHR-based foundation models that likely benefit most from the addition of targeted biomarkers to advance personalised medicine.

Systemic lupus erythematosus (SLE) is a chronic autoimmune disease characterized by over 200 risk variants identified through genome-wide association studies. While the majority of these are non-coding variants with unresolved functions, elucidating their mechanisms is critical for prioritizing therapeutic targets with increased clinical success. Several SLE risk loci span genes involved in the IL-12 signaling pathway. However, for most of them the causal variants, their definitive target genes, and their cellular consequences remain unestablished. Concurrently the expansion of double-negative 2 (DN2) B cells is a hallmark of SLE, but whether IL-12 and/or genetic risk functionally drive DN2 cells is unclear. In this study, we integrated candidate risk variants at the 3q25.33 risk locus with regulatory maps of a B cell line, identifying risk variant rs485499 located within a putative enhancer 39kb downstream of IL12A, and overlapping an open chromatin region in primary B cells stimulated with a DN2-skewing cocktail. Using CRISPR-based tools in a B cell line, we validated this region as an enhancer, rs485499 as a likely causal variant and established IL12A as its definitive target gene. Individuals homozygous for the rs485499 risk allele exhibited elevated IL12A production in naive B cells and presented with an expanded DN2 population in peripheral blood, compared to non-risk allele carriers. Mechanistically, we found the transcription factor IRF4 preferentially binds the rs485499 risk allele, driving IL12A upregulation. In vitro recombinant IL-12A promoted DN2 differentiation an effect that is abrogated by IL-12 inhibition with ustekinumab, establishing a causal IL-12-driven DN2 B cell expansion axis. Finally, we reveal that DN2 B cells inherently possess a previously unrecognized cytotoxic function that is potentiated by the IL-12 signaling axis. This cytotoxic profile is further supported by SLE patient data, identifying it as a bona fide effector state of DN2 B cells. Collectively, these findings identify rs485499 as a likely causal variant within this locus and establish a functional link between genetic risk and DN2 B cell expansion, validating this subset as a key pathogenic driver armed with a newly identified cytotoxic program. Furthermore, we identify the IL-12-IFNy-DN2 axis as a promising therapeutic target, providing a mechanistic rationale for future subset-specific interventions in SLE.

Background Substantial cardiovascular and kidney risk persists after successful weight loss. Whether features that genetics implicates in cardiovascular-kidney-metabolic (CKM) disease show larger short-term responses to weight-loss and cardiometabolic interventions than other features is unknown. Methods Across four molecular layers (proteome, transcriptome, methylome, metabolome), we applied layer-specific cis-Mendelian randomization with colocalization or summary-data shared-signal filtering against eight CKM genome-wide association studies. Within 13 omics-by-intervention analyses (diet, bariatric surgery, empagliflozin, behavioral weight loss), we compared response magnitude between nominated and adequately-instrumented non-nominated features using rank-based Cliff's {delta} ; methylation was baseline-variance-matched and directional analyses exploratory. Results Here we show genetic nomination is not consistently associated with larger observed response magnitude: estimates are small (|{delta}| < 0.08 in all 13; median |{delta}| = 0.03), near zero (descriptive pooled {delta} = -0.003, 95% CI -0.016 to +0.010; I2 = 0%), none significant by label-permutation, though smaller proteomic and transcriptomic analyses remain compatible with modest differences. Nominated proteins and metabolites frequently change (48--71% respond, like comparison features), not universal non-response. Nominated CpG sites have lower baseline inter-individual variance (Mann--Whitney P = 1.7 x 10-6 to 4 x 10-3), and apparent methylation persistence attenuates after variance-matching. Conclusions Under the definitions and datasets studied, genetic target support does not consistently predict pharmacodynamic responsiveness and should not be read as a treatment-response or reversibility biomarker; baseline dynamic range should be assessed and controlled when comparing molecular change across selected features. Nominated features are not shown unchanged or to explain residual risk; same-subject longitudinal studies are needed to test prognostic or therapeutic value.

The multiomic Variant-to-Gene (mV2G, https://mv2g.hbliulab.org) is a comprehensive atlas that integrates diverse functional genomic evidence to prioritize tissue-specific variant-to-gene (V2G) associations. While genome-wide association studies (GWAS) have identified millions of associations between genetic variants and diseases, translating these findings into biological mechanisms remains challenging because >90% of variants reside in noncoding regions. Existing V2G resources provide complementary regulatory evidence but are fragmented and often lack tissue-specific interpretation. To address this challenge, we constructed the mV2G atlas by integrating 24 types of functional genomic evidence across 50 human tissues, including molecular quantitative trait loci, enhancer-gene predictions, three-dimensional chromatin interactions, and experimental validation. The atlas contains 188,634,118 evidence-supported V2G pairs involving 13,618,039 variants and 69,521 genes. We further developed a unified tissue-specific V2G prioritization framework and prioritized 1,530,420 high-confidence functional V2G pairs involving 1,131,316 unique variants, with 87% exhibiting tissue-specificity. The mV2G atlas provides searchable variant- and gene-centered interfaces, an interactive browser for visualizing variants, target genes, cis-regulatory elements, and chromatin states, as well as downloadable datasets. By integrating complementary regulatory evidence into a unified framework, mV2G provides an accessible resource for interpreting the functional and phenotypic impact of genomic variation in relevant tissues for human diseases. GRAPHICAL ABSTRACT O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=78 SRC="FIGDIR/small/743997v1_ufig1.gif" ALT="Figure 1"> View larger version (28K): org.highwire.dtl.DTLVardef@ac8319org.highwire.dtl.DTLVardef@1d31920org.highwire.dtl.DTLVardef@169225org.highwire.dtl.DTLVardef@1d4dc02_HPS_FORMAT_FIGEXP M_FIG C_FIG

The major histocompatibility complex (MHC) is one of the most polymorphic and disease-relevant regions of the human genome, yet the downstream effects of MHC variation on circulating protein networks across ancestries remain incompletely defined. Here, we conducted multi-ancestry protein quantitative trait locus (pQTL) mapping in the UK Biobank (n = 43,762) and China Kadoorie Biobank (n = 3,977) to characterise how genetic variation within the MHC region shape plasma protein abundance. Among 2,920 plasma proteins tested, we identified significant MHC associations (13 cis and 606 trans), with most signals showing consistent effects across ancestries, while also identifying an ancestry-enriched association with EDAR in East Asians. Local genetic correlation analyses identified distinct co-regulated protein networks associated with HLA class I and class II regions, consistent with their respective known immune functions. We also conducted HLA-specific colocalisation, and found multiple HLA trans-regulated proteins associations which colocalised with immune-mediated traits at the same classical HLA gene, including colocalisation between B2M and multiple sclerosis at HLA-A (PP = 0.991). Together, these findings describe the principal downstream proteomic consequences of MHC variation and provide a framework for linking disease-associated HLA genes to immune protein networks.

Genotype imputation remains essential for large-scale human genetics studies, but its performance is limited by the size and ancestral diversity of available reference panels, reducing accuracy for rare variants and underrepresented populations. Here, we present a cloud-based imputation service built on a multi-ancestry reference panel derived from 515,579 jointly phased genomes from the All of Us (N=414,830) and National Human Genome Research Institute's Analysis, Visualization, and Informatics Lab-space (AnVIL, N=100,749) datasets. The All of Us + AnVIL reference panel is highly diverse and includes 261,163 participants most genetically similar to non-European reference populations, spanning 665,398,839 high-quality autosomal sites, representing a nearly 50% increase over TOPMed, the previous largest imputation service. Across multiple ancestry groups, the panel enables accurate imputation (empirical R2 0.8) for variants with allele frequencies as low as 0.2%, extending reliable imputation into the rare-variant frequency spectrum, including allele frequencies down to 0.002% and 0.006% for samples with European ancestry and African ancestry in the United States, respectively. Compared with TOPMed, the panel improves imputation accuracy across all ancestry groups except Africans, and recovers additional trait-associated variants not represented in existing reference panels. To facilitate broad community access while preserving participant privacy, we deploy the panel through a secure cloud-based imputation platform using privacy-preserving recombined haplotypes. This resource establishes a new foundation for genome-wide association studies (GWAS) and fine-mapping, especially in previously underrepresented populations.

Transcriptome-wide association studies (TWAS) can identify genes where genetically predicted gene expression is associated with disease risk, but translating those signals into therapeutic opportunities remains time-consuming, manual, and difficult to reproduce. We developed TRACE (TWAS-driven Repurposing through AI-assisted Curation of Evidence), a gene- and phenotype-agnostic computational pipeline that accepts a TWAS gene and effect-size direction, normalizes the gene symbol, retrieves FDA-approved drug-gene candidates from four online resources, collects related peer-reviewed literature from PubMed, and uses a fine-tuned biomedical language model to classify whether the literature supports a direct drug-gene relationship, the mechanism of action, and the direction of effect. The pipeline then compares the drug-derived direction with the direction implied by the TWAS effect estimate to rank candidate therapeutic pairs and flag potential drug safety concerns. The local classifier, built on BiomedBERT, was trained using pipeline-derived labels, BioCreative VI ChemProt gold-standard chemical-protein relation examples, and author-reviewed active-learning cases, reaching a held-out macro F1 of 0.809 across three simultaneous classification tasks. We validated the pipeline against a manually curated endometriosis gold standard of 43 drug-gene pairs spanning six TWAS-identified genes, developed through S-PrediXcan analysis of endometriosis GWAS summary statistics, manual querying of four drug-gene interaction databases for each gene, literature review of drug-gene mechanistic evidence, and Mendelian randomization validation of candidate pairs. External validation used two independently published genetically informed drug-repurposing studies in metabolic dysfunction-associated steatotic liver disease (MASLD) and type 2 diabetes (T2D). The pipeline recovered 90.7% of endometriosis pairs, 88.2% of MASLD pairs, and 92.9% of T2D pairs that were present in at least one queried database. Applied to 99 endometriosis-associated TWAS genes, the pipeline identified 1,089 FDA-approved drug-gene pairs, 32 candidate therapeutic pairs, and 77 potential safety concerns, including independent recovery of leuprolide acetate, an established endometriosis therapy. This framework provides a scalable, literature-grounded bridge from TWAS discovery to prioritized therapeutic hypotheses, while preserving uncertainty through manual-review flags and requiring downstream Mendelian randomization, electronic health record-based validation, and experimental follow-up before clinical interpretation.

Genome-wide association studies (GWAS) have identified thousands of loci associated with complex traits and diseases, and extensive efforts are underway to translate these variant-level signals into biological mechanism. A widely applied approach is pathway enrichment analysis, which tests whether genetic associations concentrate within biological pathways beyond background polygenic expectations. However, pathway databases contain extensive sharing of genes across pathways (pathway overlap), an underappreciated source of bias that creates structural dependencies in enrichment statistics and obscures pathway-specific genetic signal. Moreover, the degree of pathway overlap is increasing as pathway resources expand. Here, we introduce Gene Swap Randomization (GSR), an empirical framework that preserves pathway size and multi- pathway gene membership in the null model, enabling explicit adjustment for pathway overlap. Applying GSR to enrichment results from the Molecular Signatures Database (MSigDB) across twelve complex traits and four pathway analysis approaches (MAGMA, PascalX, GSA-MiXeR, and PRSet), we show that pathway overlap can produce enrichment under polygenicity even in the absence of pathway-specific biology. GSR improves prioritization of biologically relevant pathways supported by independent gene-disease associations (Open Targets, Malacards), regulatory interactions (DoRothEA), and tissue-specific expression patterns (GTEx). GSR improves concordance with external benchmarks in 60.8% of comparisons overall and 79.3% disease- association benchmarks, corresponding to improvement in 10 of 16 aggregated method-validation framework comparisons. We demonstrate that pathway overlap is a key source of bias in GWAS pathway enrichment, that pathway-specific disease enrichment persists after conditioning on overlap, and that GSR improves biological insight by distinguishing pathway-specific genetic signal from enrichment driven by pathway overlap.

Sex differences in human health are typically modeled as binary comparisons between males (XY) and females (XX). However, sex-linked molecular traits vary continuously and are shaped by environmental factors. Here, we introduce the Proteomic Sex Index (PSI), a continuous, plasma proteomics-based measure that quantifies the degree to which an individual's proteome aligns with their chromosomal sex, with higher PSI values indicating profiles less typical for their sex. Leveraging data from 51,310 individuals from six demographically heterogeneous cohorts, we show that PSI captures broad, biologically meaningful variation in sex-enriched traits and diseases. Individuals with proteomic profiles shifted away from their chromosomal sex exhibited trait values less characteristic of that sex. For example, males with higher PSI were shorter, had lower grip strength, and showed reduced severity of COVID-19 (OR=0.59, p=3.0x10-7). Analyses of individuals undergoing gender-affirming hormone therapy showed that, although PSI correlated with circulating sex hormones, it also captured additional variation independent of hormone levels. PSI is partially heritable and genetically influenced. Individuals with Klinefelter syndrome (XXY) had markedly elevated PSI (OR=13.7, p=1.1x10-13), and genome-wide association studies of PSI identified sex-biased cis-pQTLs with evidence of causal links to sex-specific disease risk. PSI provides a nuanced, individualized measure of sex-related biology that enables the study of within-sex molecular variation and its links to health and disease.

Motivation: Interpreting noncoding pharmacogenomic variants requires identifying the regulatory program that may be perturbed and the direction of the effect, information not supplied by population frequency or general regulatory annotations. Results: We developed AetherXeno, an allele-resolved atlas that scores all substitutions across 8,691 experimentally supported elements for PXR, FXR and AhR and connects 17.27 million variants to evidence matched by exact allele. Across the six paired comparisons, the gain in Spearman correlation ranged from 0.039 to 0.334. Every regional bootstrap interval excluded zero. Homology exclusion preserved these findings, while motif core and liver expression enrichment supported biological organization of the score landscape. The browser, programmatic interface and bulk files expose the same versioned records from individual variants to cohort analyses. Availability and implementation: The web resource is available at https://aetherxeno.org and the versioned data at doi:10.5281/zenodo.20753827.

The completion of telomere-to-telomere (T2T) human genomes has expanded the accessible landscape of human genetic variation, yet benchmark resources remain limited to conventional high-confidence regions defined by existing reference frameworks. Here, we generated a near-perfect diploid T2T genome (T2T-LIN) from a Chinese individual and established assembly-based truth sets by comparison with T2T-YAO, an ancestry-matched near-perfect T2T reference genome. The benchmark showed a heterozygous/homozygous SNV ratio of ~2, consistent with expectations under Hardy-Weinberg equilibrium, and enabled genome-wide evaluation of reference-dependent biases. We found that reference choice substantially influences genome interpretation: ancestry-matched linear T2T references provided the most faithful representation of individual genomic variation and enabled more accurate genome reconstruction than unmatched linear, diploid and graph-based references. Benchmarking previously inaccessible repetitive and structurally complex regions revealed substantial limitations of current variant callers that were masked by conventional metrics. The T2T-LIN and YAO-LIN benchmarks establish a T2T-era framework for evaluating reference-dependent genome interpretation and variant discovery across nearly the complete human genome.

The retinal pigment epithelium and choroid are critical for supporting the function and maintaining the homeostasis of the outer retina, and their dysfunction underlies a range of inherited and complex ocular diseases. To comprehensively characterize the cellular, transcriptomic, and epigenomic heterogeneity and dynamics within these tissues, we assembled an integrated multi-omics reference atlas comprising 719,813 single-cell/single-nucleus transcriptomes and 234,007 snATAC-seq profiles from 102 ancestrally diverse donors spanning 0 to 99 years of age, including cells from both the macula and periphery. This atlas resolves 48 distinct cell types or states and catalogs 448,567 open chromatin regions. Specifically, we resolved five distinct RPE subpopulations organized along a central-to-peripheral spatial axis, alongside two distinct stress/senescence states. We reconstructed the transcriptomic and epigenetic zonation of endothelial cells and expanded choroidal stromal heterogeneity by characterizing 11 fibroblast and two pericyte types. Age-associated compositional analysis revealed a significant fractional depletion of melanocytes, PI16+ fibroblasts, and venule endothelial cells with age, alongside a modest relative loss of central RPE and a corresponding increase in far-peripheral RPE. Cell-type-specific aging transcriptomics uncovered shared pathways related to inflammatory responses alongside distinct cell-type-specific signatures. Notably, significant age-associated epigenetic changes concentrated in the macula during the transition from early-to-middle adulthood and remained stable into old age, with transcription factors from the AP-1/bZIP family emerging as the dominant enriched motifs. Finally, integrating this atlas with AMD GWAS data provides novel variant-to-gene evidence implicating LIPG and COL4A3 in AMD pathogenesis. Together, this multi-omics atlas serves as both an invaluable community reference and a powerful discovery engine that translates genetic risk signals into localized target cells and candidate mechanisms, laying a foundation for understanding RPE/choroid biology in health and disease.

Medical records and omics data are rapidly becoming standard in healthcare settings, which characterize the whole-person from dysfunctional molecules to phenotypes, and thus offer potential for precise disease diagnosis and target discovery. Whereas, it remains an open problem to systematically integrate and interprete medical record and omics data of individual patients. In this study, for the first time, we propose a novel graph AI model framework, Graph in Graph (GiG), to integrate and interpret the whole-person medical and omic datasets. Specifically, the medical record data is modeled using a person-phenotype graph, followed by omics signaling graphs of invidival patients, which enables the integration of information learned from omic-signaling graph and medical phenotype features to characterize individual patients and to prioritize important omic biomarkers and phenotypes. As an exploratory study, we applied and evaluated the GiG model to study the type 2 diabetes (T2D) and pre-T2D vs healthy using the Long Life Family Study (LLFS) cohort, which enrolls families with exceptional longevity to uncover biological mechanisms of healthy aging with medical and omics data. The evaluation results showed that GiG not only achieve a high prediction but also can interpret the prediction by ranking the essential clinical and omic biomarkers. The GiG framework can be applied to other studies by effectively integrating and interpreting medical and omic datasets for disease diagnosis and pathogenesis discovery.

Most genes produce multiple non-coding and coding RNA transcripts with potentially divergent biological functions, yet their contribution to cancer remains largely unexplored. Here, we built an integrated, transcript-resolution RNA sequencing atlas of prostate cancer spanning 1,140 clinical samples from 10 independent studies, covering the full range of disease progression from normal tissue to primary, hormone-sensitive, and castration-resistant disease. Cross-study integration removes technical batch effects while retaining the biological heterogeneity that distinguishes these disease states, a property we validate against long-read sequencing and leverage to place newly incorporated external cohorts within this disease-progression context. We describe transcript patterns of oncogenic and tumor-suppressive long non-coding RNAs along disease progression and identify 41 protein-coding genes with significant isoform switches in cancer-relevant pathways, including WNT2B, RNF43, RELA, BRCA1, ROCK2, and TRIM11. Finally, we provide the Prostate Cancer Atlas (www.prostatecanceratlas.org), an interactive web resource that lets researchers contextualize their own transcriptomic data against this atlas.