Biology subjects

Sankararaman, S.

Publications and source records attributed to Sankararaman, S..

At least 19 recordsLinked to original sources

bioRxiv2026

Genome-wide LD scores enable robust heritability and genetic correlation estimation from summary statistics across genetic ancestries

Our knowledge of human complex trait architecture across diverse populations remains limited. Estimating SNP heritability and genetic correlation from GWAS summary statistics - successful in revealing genetic architecture in homogeneous cohorts - is challenging in structured and admixed populations due, in part, to the difficulty of representing long-range LD. We present SUMMIT, a statistical framework that combines genome-wide LD scores, efficiently estimated using random projections, with GWAS summary statistics to estimate genome-wide and annotation-specific heritability and genetic correlation. In simulations that explore diverse aspects of genetic architecture across European (EUR), South Asian (SAS), and admixed African-European cohorts (AFR) in the UK Biobank (UKB), SUMMIT had increased accuracy and reduced bias compared to prior methods, with the largest gains in non-European ancestries reflecting improved modeling of long-range LD. Across 40 UKB traits, SUMMIT produced estimates of genome-wide common-SNP heritability (h_g^2) and heritability enrichments in genomic annotations that were broadly concordant across ancestries with the SUMMIT h_g^2 estimates more closely matching recent whole-genome sequencing based estimates. SUMMIT substantially improved genetic correlation analysis recovering more significant findings across ancestries (23 SAS and 3 AFR at FDR < 0.05 compared with none and two for cov-LDSC) with the non-European discoveries showing concordant effects in Europeans suggesting that, like heritability, global patterns of pleiotropy are largely shared across ancestries. By combining accurate genome-wide LD and efficient moment estimation, SUMMIT enables investigation into how heritability and genetic correlation vary across populations and genomic annotations, including in structured and admixed cohorts, at the genome-wide, annotation-specific, and cross-trait levels.

genomics↗

bioRxiv2026

A biobank-scale method for learning modulators of gene-environment interaction underlying human complex traits from multiple environmental exposures

It is increasingly recognized that genetic effects on complex traits and diseases are shaped by environmental context. Biobanks that measure diverse environmental exposures alongside genotypes and phenotypes at scale enable systematic study of gene-environment (GxE) interactions. Existing approaches, however, are limited in their ability to accurately model polygenic GxE involving many exposures across genome-wide genetic variants. It is unclear which exposure combinations are relevant for a given trait while distinguishing true interactions from environment-dependent heteroskedastic noise. To address these challenges, we develop Efficient multi-eNvironmental Gene-environment Interaction iNference Estimator (ENGINE), a supervised variance-component framework that learns an embedding that combines multiple environmental exposures while jointly estimating additive, GxE, and heteroskedastic noise components. To enable biobank-scale inference, ENGINE makes a single pass over the genotype matrix to cache genotype-dependent summaries, then assembles normal-equation components and gradients at each iteration. In simulations, ENGINE controls type I error rates, achieves high power, and accurately recovers the environmental embedding while remaining efficient at biobank-scale. Applied to five complex traits paired with lifestyle exposures in N = 291,273 unrelated white British individuals and M = 454,207 common SNPs (MAF> 0.01) from the UK Biobank, ENGINE recovered GxE variance that was on average 1.4-fold larger than that captured by a single exposure and 5.5-fold larger than that captured by the first principal component of the exposures.

genetics↗

bioRxiv2026

keju: powerful and accurate inference in Massively Parallel Reporter Assays

Massively Parallel Reporter Assays (MPRAs) interrogate the regulatory function of thousands of designed genetic elements in parallel through linked DNA and RNA readouts using an engineered construct and attached minimal reporter. Given the complexity of MPRA experimental designs, several different sources of uncertainty complicate inference. We show that previous methods do not account for substantial differences in uncertainty levels between the DNA and RNA counts and between batches. Accordingly, we present keju, a hierarchical statistical model that estimates candidate transcription rate, differential activity between conditions, and effects from promoter composition for MPRA data. To maximize statistical power and improve false positive rate control, keju conditions on the DNA counts to model batch-specific and modality-specific uncertainty in the RNA counts. keju shows vastly improved sensitivity (59%) in simulations compared to previous methods (31% for MPRAnalyze and 9% for BCalm), and also has lower, more robust false positive rates, calling only 6.8% of unlabeled negative controls significant in real data (compared to 34% for MPRAnalyze and 12% for BCalm).

bioinformatics↗

bioRxiv2026

Choice of phenotype scale is critical in biobank-based GxE tests

The importance of gene-environment interactions (GxE) for complex human traits is heavily debated. Recently, biobank-based GWAS have revealed many statistically significant GxE signals, though most lack clear evidence of biological significance. Here, we partly explain this discrepancy by showing that many GxE signals simplify to additive effects on a different phenotype scale, a classical concern that is currently underappreciated. Our results clearly distinguish GxSex effects on height, which vanish on the log scale, from GxSex effects on testosterone, where the log scale uncovers biologically meaningful female-specific effects. Across 32 phenotypes in UK Biobank, we find that scaling by a power transformation can explain 46% of PGSxSex interactions, and that simple log transformation can explain 23%, with similar results for other environments. We also show that phenotype scale can substantially impact GWAS discovery and the construction and evaluation of polygenic scores. Finally, we provide a set of guidelines to consider and choose phenotype scale in modern genetic studies.

genetics↗

bioRxiv2025

Comprehensive gene heritability estimation reveals the genetic architecture of rare coding variants underlying complex traits

Whole-exome sequencing (WES) enables high-resolution interrogation of the contribution of rare coding variants to complex trait variation. However, existing methods for heritability estimation attributed to rare-coding variants are often limited by the effects of linkage disequilibrium (LD) and by the sparse nature of rare variant data. We introduce FLEX (Fast, LD-aware Estimation of eXome-wide and gene-level heritability), a scalable and flexible framework for estimating and partitioning heritability across genes or sets of genes using WES data. FLEX integrates all coding variants- from common to ultra-rare - within a unifled model and corrects for LD-induced effects to improve the accuracy of heritability estimates. In addition, FLEX supports both individual-level and summary statistic data and is computationally efflcient for biobank-scale datasets. Through extensive simulations, we show that FLEX is well-calibrated while providing accurate heritability estimates. We applied FLEX to WES data across N = 153, 351 unrelated European ancestry individuals and 20 quantitative traits in the UK Biobank. We identifled 64 gene-trait pairs with signiflcant gene-level heritability (p < 0.05/18, 624 accounting for the number of protein-coding genes tested), among which rare coding variants explained 38% of gene-level heritability, on average. Compared to heritability estimates from genome-wide imputed SNPs, incorporation of rare and ultra-rare coding variants led to a 24.8% increase in heritability on average, while effect sizes at rare and ultra-rare variants are substantially larger ({approx} 18x on average). Partitioning across variant effect annotations, we flnd that predicted loss-of-function variants had stronger individual effects than missense variants (24% on average) while missense variants accounted for a greater share of rare coding heritability. Together, FLEX provides an adaptable and accurate approach for quantifying gene-level heritability, advancing our understanding of the genetic architecture of complex traits, and facilitating the discovery of trait-relevant genes.

genetics↗

bioRxiv2025

Neanderthal introgressed ancestry reveals human genomic regions enriched with recessive deleterious mutations

Negative selection on deleterious mutations plays a key role in shaping human genetic variation, yet the dominance effects of these mutations remain poorly understood because existing statistical methods often cannot distinguish dominance effects from the overall selective effects. In this work, we take a fundamentally different approach by leveraging the distribution of Neanderthal ancestry across the human genome. Simulations show that recessive deleterious mutations can increase archaic introgressed ancestry through heterosis in the absence of positive selection, in contrast to the depletion expected under additive effects. We use this signal to develop DominL, a machine learning classifier trained on simulations of human demography with Neanderthal introgression to identify megabase-scale genomic windows enriched with recessive mutations. DominL demonstrates robust accuracy, with particularly high power in exon-dense regions. Applied to 7 non-African populations from the 1000 Genomes Project, DominL identifies approximately 3-9% of the genome as enriched for recessive mutations, with most high-confidence regions shared across populations. Predicted regions show patterns consistent with expected signatures from recessive mutations, including weakened background selection, depletion of runs of homozygosity, and enrichment of non-additive trait-associated variants. These regions also contain genes associated with metabolic and immune-related functions.

evolutionary biology↗

bioRxiv2024

Identifying novel genetic and phenotypic associations to genomic features by leveraging off-target reads in exome sequencing data

Upwards of 40% of reads in sequencing datasets may be unmapped and discarded by standard protocols. Recent work has shown the utility of re-analyzing these unmapped reads to construct meaningful features, such as immune diversity repertoires or copy number variation in mtDNA and rDNA. While previous analyses of these features have produced significant correlations with diverse traits, they have generally been limited to analyses of RNA-sequencing data in phenotype-specific cohorts. Here, we explore whether associations can be identified using population-scale, whole-exome sequencing data in the UK BioBank. Using recently developed tools, we constructed multiple features including T-cell receptor diversity metrics, microbial load, and mtDNA and rDNA copy numbers for nearly 50,000 individuals in the UK BioBank. We first verify the validity of our method by showing that GWAS on these constructed traits results in replication of associations from studies in which the phenotypes were explicitly measured. Next, across several GWAS, we identified 21 novel independent significant loci in 11 genes, most of them in genes implicated in the innate immune response. Finally, we further analyzed the read-constructed features by establishing correlations to other population-level biobank traits such as immune disorders, metabolic disorders, neuropsychiatric disorders, and blood cell counts. Our results suggest that existing tools for feature construction from unmapped reads can offer novel information at the population level, and that these features can be used to establish novel genetic associations.

bioinformatics↗

bioRxiv2024

Single-cell DNA methylome and 3D genome atlas of the human subcutaneous adipose tissue

Human subcutaneous adipose tissue (SAT) contains a diverse array of cell-types; however, the epigenomic landscape among the SAT cell-types has remained elusive. Our integrative analysis of single-cell resolution DNA methylation and chromatin conformation profiles (snm3C-seq), coupled with matching RNA expression (snRNA-seq), systematically cataloged the epigenomic, 3D topology, and transcriptomic dynamics across the SAT cell-types. We discovered that the SAT CG methylation (mCG) landscape is characterized by pronounced hyper-methylation in myeloid cells and hypo-methylation in adipocytes and adipose stem and progenitor cells (ASPCs), driving nearly half of the 705,063 detected differentially methylated regions (DMRs). In addition to the enriched cell-type-specific transcription factor binding motifs, we identified TET1 and DNMT3A as plausible candidates for regulating cell-type level mCG profiles. Furthermore, we observed that global mCG profiles closely correspond to SAT lineage, which is also reflected in cell-type-specific chromosome compartmentalization. Adipocytes, in particular, display significantly more short-range chromosomal interactions, facilitating the formation of complex local 3D genomic structures that regulate downstream transcriptomic activity, including those associated with adipogenesis. Finally, we discovered that variants in cell-type level DMRs and A compartments significantly predict and are enriched for variance explained in abdominal obesity. Together, our multimodal study characterizes human SAT epigenomic landscape at the cell-type resolution and links partitioned polygenic risk of abdominal obesity to SAT epigenome.

genomics↗

bioRxiv2024

Simple models of non-random mating and environmental transmission bias standard human genetics statistical methods

There is recognition among human complex-trait geneticists that not only are many common assumptions made for the sake of statistical tractability (e.g., random mating, independence of parent/offspring environments) unlikely to apply in many contexts, but that methods reliant on such assumptions can yield misleading results, even in large samples. Investigations of the consequences of violating these assumptions so far have focused on individual perturbations operating in isolation. Here, we analyze widely used estimators of genetic architectural parameters, including LD-score regression and both population-based and within-family GWAS, across a broad array of perturbations to classical assumptions, such as multivariate assortative mating and vertical transmission (parental effects on offspring phenotypes not mediated by genetic inheritance). We find that widely-used statistical approaches are unreliable across a broad range of perturbations, and that structural sources of confounding often operate synergistically to distort conclusions. For example, mild multivariate assortative mating and vertical transmission together can dramatically inflate heritability estimates and GWAS false positive rates. Further, GWAS will become progressively more polluted by off-target associations as sample sizes increase. Given these challenges, we introduce xftsim, a forward time simulation library capable of modeling a wide range of genetic architectures, mating regimes, and transmission dynamics, to facilitate the systematic comparison of existing approaches and the development of robust methods. Together, our findings illustrate the importance of comprehensive sensitivity analysis and present a valuable tool for future research.

genetics↗

bioRxiv2024

Fast variance component analysis using large-scale ancestral recombination graphs

Recent algorithmic advancements have enabled the inference of genome-wide ancestral recombination graphs (ARGs) from genomic data in large cohorts. These inferred ARGs provide a detailed representation of genealogical relatedness along the genome and have been shown to complement genotype imputation in complex trait analyses by capturing the effects of unobserved genomic variants. An inferred ARG can be used to construct a genetic relatedness matrix, which can be leveraged within a linear mixed model for the analysis of complex traits. However, these analyses are computationally infeasible for large datasets. We introduce a computationally efficient approach, called ARG-RHE, to estimate narrow-sense heritability and perform region-based association testing using an ARG. ARG-RHE leverages a method for computing genotype-matrix products from genealogical data in sublinear time, along with scalable randomized algorithms. This enables fast estimation of variance components and their statistical significance, supports parallel analysis of multiple quantitative traits, and facilitates other linear mixed-model analyses. We conduct extensive simulations to verify the computational efficiency, statistical power, and robustness of this approach. We then apply it to detect associations between 21,159 genes and 52 blood-related traits, using an ARG inferred from genotype data of 337,464 individuals from the UK Biobank. In these analyses, combining ARG-based and imputation-based testing yields 8% more gene-trait associations than using imputation alone, suggesting that inferred genome-wide genealogies may effectively complement genotype imputation in the analysis of complex traits.

genetics↗

bioRxiv2024

A Scalable Adaptive Quadratic Kernel Method for Interpretable Epistasis Analysis in Complex Traits

Our knowledge of the contribution of genetic interactions (epistasis) to variation in human complex traits remains limited, partly due to the lack of efficient, powerful, and interpretable algorithms to detect interactions. Recently proposed approaches for set-based association tests show promise in improving power to detect epistasis by examining the aggregated effects of multiple variants. Nevertheless, these methods either do not scale to large numbers of individuals available in Biobank datasets or do not provide interpretable results. We, therefore, propose QuadKAST, a scalable algorithm focused on testing pairwise interaction effects (also termed as quadratic effects) of a set of genetic variants on a trait and quantifying the proportion of phenotypic variance explained by these effects. We performed comprehensive simulations and demonstrated that QuadKAST is well-calibrated. Additionally, QuadKAST is highly sensitive in detecting loci with epistatic signal and accurate in its estimation of quadratic effects. We applied QuadKAST to 53 quantitative phenotypes measured in {approx} 300, 000 unrelated white British individuals in the UK Biobank to test for quadratic effects within each of 9, 515 protein-coding genes (after accounting for linear additive effects). We detected 32 trait-gene pairs across 17 traits that demonstrate statistically significant signals of quadratic effects ([Formula] accounting for the number of genes and traits tested). Our method enables the detailed investigation of epistasis on a large scale, offering new insights into its role and importance.

genetics↗

bioRxiv2024

Scalable summary statistics-based heritability estimation method with individual genotype level accuracy

SNP heritability, the proportion of phenotypic variation explained by genotyped SNPs, is an important parameter in understanding the genetic architecture underlying various diseases and traits. Methods that aim to estimate SNP heritability from individual genotype and phenotype data are limited by their ability to scale to Biobank-scale datasets and by the restrictions in access to individual-level data. These limitations have motivated the development of methods that only require summary statistics. While the availability of publicly accessible summary statistics makes them widely applicable, these methods lack the accuracy of methods that utilize individual genotypes. Here we present a SUMmary statistics-based Randomized Haseman-Elston regression (SUM-RHE), a method that can estimate the SNP heritability of complex phenotypes with accuracies comparable to approaches that require individual genotypes, while exclusively relying on summary statistics. SUM-RHE employs Genome-Wide Association Study (GWAS) summary statistics and statistics obtained on a reference population, which can be efficiently estimated and readily shared for public use. Our results demonstrate that SUM-RHE obtains estimates of SNP heritability that are substantially more accurate compared to other summary statistic methods and on par with methods that rely on individual-level data.

genomics↗

bioRxiv2023

A scalable and robust variance components method reveals insights into the architecture of gene-environment interactions underlying complex traits

Understanding the contribution of gene-environment interactions (GxE) to complex trait variation can provide insights into mechanisms underlying disease risk, explain sources of heritability, and improve the accuracy of genetic risk prediction. While biobanks that collect genetic and deep phenotypic data over large numbers of individuals offer the promise of obtaining novel insights into GxE, our understanding of the architecture of GxE in complex traits remains limited. We introduce a method that can estimate the proportion of trait variance explained by GxE (GxE heritability) and additive genetic effects (additive heritability) across the genome and within specific genomic annotations. We show that our method is accurate in simulations and computationally efficient for biobank-scale datasets. We applied our method to{approx} 500, 000 common array SNPs (MAF[≥] 1%), fifty quantitative traits, and four environmental variables (smoking, sex, age, and statin usage) measured across{approx} 300, 000 unrelated white British individuals in the UK Biobank. We found 69 trait-environmental variable pairs with significant genome-wide GxE heritability (p < 0.05/200 correcting for the number of trait-E pairs tested) with an average ratio of GxE to additive heritability{approx} 6.8% that include BMI with smoking (ratio of GxE to additive heritability = 6.3 {+/-} 1.1%), WHR (waist-to-hip ratio adjusted for BMI) with sex (ratio = 19.6 {+/-} 2%), LDL cholesterol with age (ratio = 9.8 {+/-} 3.9%), and HbA1c with statin usage (ratio = 11 {+/-} 2%). Analyzing nearly 8 million common and low-frequency imputed SNPs (MAF[≥] 0.1%), we document an increase in genome-wide GxE heritability of about 28% on average over array SNPs. We partitioned GxE heritability across minor allele frequency (MAF) and local linkage disequilibrium values (LD score) of each SNP to observe that analogous to the relationship that has been observed for additive allelic effects, the magnitude of GxE allelic effects tends to increase with decreasing MAF and LD. Testing whether GxE heritability is enriched around genes that are highly expressed in specific tissues, we find significant tissue-specific enrichments that include brain-specific enrichment for BMI and Basal Metabolic Rate in the context of smoking, adipose-specific enrichment for WHR in the context of sex, and cardiovascular tissue-specific enrichment for total cholesterol in the context of age. Our analyses provide detailed insights into the architecture of GxE underlying complex traits.

genetics↗

bioRxiv2023

A multivariate outcome test of covariance

Multivariate analysis is becoming central in studies investigating high-throughput molecular data, yet, some important features of these data are seldom explored. Here, we present MANOCCA (Multivariate Analysis of Conditional CovAriance), a powerful method to test for the effect of a predictor on the covariance matrix of a multivariate outcome. The proposed test is by construction orthogonal to tests based on the mean and variance, and is able to capture effects that are missed by both approaches. We first compare the performances of MANOCCA with existing correlation-based methods and show that MANOCCA is the only test correctly calibrated in simulation mimicking omics data. We then investigate the impact of reducing the dimensionality of the data using principal component analysis when the sample size is smaller than the number of pairwise covariance terms analysed. We show that, in many realistic scenarios, the maximum power can be achieved with a limited number of components. Finally, we apply MANOCCA to 1,000 healthy individuals from the Milieu Interieur cohort, to assess the effect of health, lifestyle and genetic factors on the covariance of two sets of phenotypes, blood biomarkers and flow cytometry-based immune phenotypes. Our analyses identify significant associations between multiple factors and the covariance of both omics data.

bioinformatics↗

bioRxiv2023

A biobank-scale test of marginal epistasis reveals genome-wide signals of polygenic epistasis

The contribution of epistasis (interactions among genes or genetic variants) to human complex trait variation remains poorly understood. Methods that aim to explicitly identify pairs of genetic variants, usually single nucleotide polymorphisms (SNPs), associated with a trait suffer from low power due to the large number of hypotheses tested while also having to deal with the computational problem of searching over a potentially large number of candidate pairs. An alternate approach involves testing whether a single SNP modulates variation in a trait against a polygenic background. While overcoming the limitation of low power, such tests of polygenic or marginal epistasis (ME) are infeasible on Biobank-scale data where hundreds of thousands of individuals are genotyped over millions of SNPs. We present a method to test for ME of a SNP on a trait that is applicable to biobank-scale data. We performed extensive simulations to show that our method provides calibrated tests of ME. We applied our method to test for ME at SNPs that are associated with 53 quantitative traits across {approx} 300 K unrelated white British individuals in the UK Biobank (UKBB). Testing 15, 601 trait-loci associations that were significant in GWAS, we identified 16 trait-loci pairs across 12 traits that demonstrate strong evidence of ME signals (p-value [Formula]). We further partitioned the significant ME signals across the genome to identify 6 trait-loci pairs with evidence of local (within-chromosome) ME while 15 show evidence of distal (cross-chromosome) ME. Across the 16 trait-loci pairs, we document that the proportion of trait variance explained by ME is about 12x as large as that explained by the GWAS effects on average (range: 0.59 to 43.89). Our results show, for the first time, evidence of interaction effects between individual genetic variants and overall polygenic background modulating complex trait variation.

genetics↗

bioRxiv2023

VISTA: An integrated framework for structural variant discovery

Structural variation (SV), refers to insertions, deletions, inversions, and duplications in human genomes. With advances in whole genome sequencing (WGS) technologies, a plethora of SV detection methods have been developed. However, dissecting SVs from WGS data remains a challenge, with the majority of SV detection methods prone to a high false-positive rate, and no existing method able to precisely detect a full range of SVs present in a sample. Previous studies have shown that none of the existing SV callers can maintain high accuracy across various SV lengths and genomic coverages. Here, we report an integrated structural variant calling framework, VISTA (Variant Identification and Structural Variant Analysis) that leverages the results of individual callers using a novel and robust filtering and merging algorithm. In contrast to existing consensus-based tools which ignore the length and coverage, VISTA overcomes this limitation by executing various combinations of top-performing callers based on variant length and genomic coverage to generate SV events with high accuracy. We evaluated the performance of VISTA on using comprehensive gold-standard datasets across varying organisms and coverage. We benchmarked VISTA using the Genome-in-a-Bottle (GIAB) gold standard SV set, haplotype-resolved de novo assemblies from The Human Pangenome Reference Consortium (HPRC)1,2, along with an in-house PCR-validated mouse gold standard set. VISTA maintained the highest F1 score among top consensus-based tools measured using a comprehensive gold standard across both mouse and human genomes. VISTA also has an optimized mode, where the calls can be optimized for precision or recall. VISTA-optimized is able to attain 100% precision and the highest sensitivity among other variant callers. In conclusion, VISTA represents a significant advancement in structural variant calling, offering a robust and accurate framework that outperforms existing consensus-based tools and sets a new standard for SV detection in genomic research.

bioinformatics↗

bioRxiv2023

Tractable and Expressive Generative Models of Genetic Variation Data

Generative models play an important role in population genetics, being used to generate artificial genomes (AGs) that help benchmark methods, test evolutionary hypotheses, and construct reference panels for imputation while working around data-sharing restrictions. Existing generative models of genetic variation, however, struggle to faithfully express dependencies in the data while retaining tractability and preserving privacy. We introduce Genetic Probabilistic Circuits (GPC), a deep generative model for genetic variation data based on hidden Chow-Liu trees represented as probabilistic circuits. GPC generalizes traditional hidden Markov models by allowing arbitrary tree structures over latent variables, enabling it to capture long-range dependencies among SNPs. GPC is tractable, supporting exact computation of marginal and conditional probabilities that enables both AG generation and direct genotype imputation (avoiding the need for simulating AGs). We show that GPC generates AGs that accurately reproduce population structure and linkage disequilibrium patterns across a range of length scales. Compared to other deep generative approaches, GPC consistently improves imputation accuracy, with particularly strong gains for low-frequency variants and in populations that are not well-represented in public reference panels. Finally, we show that GPC better preserves privacy of the training data, thereby providing a practical framework for AG generation in settings with limited data access.

genetics↗

bioRxiv2023

Phenotypic subtyping via contrastive learning

Defining and accounting for subphenotypic structure has the potential to increase statistical power and provide a deeper understanding of the heterogeneity in the molecular basis of complex disease. Existing phenotype subtyping methods primarily rely on clinically observed heterogeneity or metadata clustering. However, they generally tend to capture the dominant sources of variation in the data, which often originate from variation that is not descriptive of the mechanistic heterogeneity of the phenotype of interest; in fact, such dominant sources of variation, such as population structure or technical variation, are, in general, expected to be independent of subphenotypic structure. We instead aim to find a subspace with signal that is unique to a group of samples for which we believe that subphenotypic variation exists (e.g., cases of a disease). To that end, we introduce Phenotype Aware Components Analysis (PACA), a contrastive learning approach leveraging canonical correlation analysis to robustly capture weak sources of subphenotypic variation. In the context of disease, PACA learns a gradient of variation unique to cases in a given dataset, while leveraging control samples for accounting for variation and imbalances of biological and technical confounders between cases and controls. We evaluated PACA using an extensive simulation study, as well as on various subtyping tasks using genotypes, transcriptomics, and DNA methylation data. Our results provide multiple strong evidence that PACA allows us to robustly capture weak unknown variation of interest while being calibrated and well-powered, far superseding the performance of alternative methods. This renders PACA as a state-of-the-art tool for defining de novo subtypes that are more likely to reflect molecular heterogeneity, especially in challenging cases where the phenotypic heterogeneity may be masked by a myriad of strong unrelated effects in the data. Code AvailabilityPACA is available as an open source R package on GitHub: https://github.com/Adigorla/PACA

genomics↗