📄 genetic and genomic medicine

Conditional polygenic enrichment distinguishes causal from tagging disease-critical cell populations in single-cell RNA-seq

The paper introduces scDRS-FM, a novel method that integrates single-cell RNA-seq and GWAS data to distinguish causal disease-critical cell populations from correlated tagging populations by modeling conditional polygenic enrichment, thereby enabling more accurate fine-mapping of disease-relevant cellular contexts across diverse traits and cell types.

Original authors: Turcan, A., Hou, K., Lin, K. Z., Pfenning, A., Sakaue, S., Zhang, M. J.

Published 2026-08-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Turcan, A., Hou, K., Lin, K. Z., Pfenning, A., Sakaue, S., Zhang, M. J.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Geneticists have long held a map of the human genome dotted with thousands of tiny markers that point to where disease risk lives. These markers, found by scanning the DNA of hundreds of thousands of people, act like street signs indicating a neighborhood where something is wrong. However, knowing the neighborhood is not the same as knowing the house. For decades, scientists have struggled to pinpoint exactly which cells in the body are responsible for these genetic risks. A disease might be linked to a specific type of immune cell, but within that broad category, there are many subtypes that look and act almost identically. When researchers try to find the culprit, they often get a list of suspects that includes both the true cause and innocent bystanders that simply hang out in the same neighborhood and share similar traits. This confusion makes it difficult to understand how a disease actually starts or how to stop it.

To solve this, a team of researchers has developed a new method called scDRS-FM, designed to separate the true causes of disease from the innocent bystanders in the body's cellular landscape. The researchers combined two massive sets of data: genetic information from people with various diseases and detailed snapshots of gene activity from millions of individual cells. By looking at these cells one by one, rather than in large groups, they could see which specific cells were actually driving the disease risk. The key innovation was a statistical technique that allowed them to ask a specific question for every cell: "Is this cell showing signs of disease risk even after we account for the fact that it looks a lot like its neighbors?" This approach filters out the noise of cells that are only associated with a disease because they share a common language with the true culprits, leaving behind a much clearer picture of the actual biological mechanisms at work.

The researchers tested their method on a vast array of biological data, including genetic information from 75 different diseases and complex traits, such as body mass index and cholesterol levels. They paired this with genetic snapshots from over 5.8 million individual cells, covering 580 different types of cells in both humans and mice. In the past, older methods would often flag dozens of cell types as being involved in a single disease, creating a confusing list where it was impossible to tell who was responsible. For example, when looking at brain-related diseases, previous tools might suggest that almost every type of brain cell was involved. The new method, however, acted like a filter, reducing that long list to a much shorter, more precise set of suspects. It successfully distinguished between cells that were truly driving the disease and those that were merely tagging along because they shared similar genetic programs.

When applied to immune diseases, the method revealed specific subgroups of immune cells that were driving conditions like inflammatory bowel disease. It found that certain activated T cells, which are part of the body's defense system, were the primary drivers, while other similar-looking cells were not. The researchers could even see that these dangerous cells were characterized by a specific pattern of activity, producing a mix of chemical signals that fueled inflammation. Similarly, in the brain, the method identified specific populations of immune cells called microglia and certain types of neurons that were linked to Alzheimer's disease. It showed that the risk was not spread evenly across the brain but was concentrated in specific regions, such as the midtemporal gyrus and the entorhinal cortex, and was tied to a loss of normal, healthy maintenance functions in those cells.

The study also explored how different diseases are related to one another. By comparing the cellular activity patterns across the 75 diseases, the researchers found connections that genetic maps alone had missed. Two diseases might not share many of the same genetic risk markers, yet they could still be driven by the same types of cells working in the same way. This suggests that different diseases can converge on the same biological pathways, offering new ways to think about how they might be treated. The researchers confirmed their findings by testing them against other datasets and using different statistical approaches, showing that their results were robust and reliable. They also demonstrated that their method worked well in computer simulations, where they knew exactly which cells were supposed to be the cause, proving that the tool could accurately find the needle in the haystack.

This work represents a significant step forward in understanding the cellular roots of human disease. By moving beyond broad categories and identifying the precise cellular populations responsible for genetic risk, scientists can now focus their efforts on the right targets. The method does not just list which cells are involved; it explains how they are involved, distinguishing between the active drivers of disease and the passive observers. This clarity is essential for developing new therapies that can intervene at the source of a problem rather than just treating the symptoms. As researchers continue to gather more detailed data on the human body, tools like this will be crucial for turning the vast complexity of our biology into a clear, actionable understanding of health and disease.

Problem Statement
Integrating single-cell RNA sequencing (scRNA-seq) with genome-wide association studies (GWAS) offers a pathway to identify the specific cell types and states underlying heritable diseases. However, existing methods primarily rely on marginal association testing, which struggles to distinguish between causal cell populations and "tagging" populations. Tagging occurs when non-causal cells exhibit significant disease associations simply because they share gene expression programs with causal cells (analogous to linkage disequilibrium in variant-level GWAS). This issue is pervasive, particularly among closely related cell states (e.g., different T cell subsets or neuronal populations across brain regions), leading to inflated associations that obscure true biological mechanisms. Current approaches lack the statistical power and resolution to disentangle these signals at the single-cell level while accounting for sequencing noise and cell-cell correlations.

Methodology: scDRS-FM
The authors introduce scDRS-FM, a framework designed to fine-map disease-associated individual cells by evaluating conditional polygenic enrichment. The method generalizes the marginal scDRS approach to a conditional setting through three primary steps:

Denoising and Gene Set Construction: scRNA-seq data is denoised using the MAGIC algorithm to improve signal-to-noise ratios. A putative disease gene set is constructed from GWAS summary statistics, selecting the top 1,000 genes ranked by MAGMA gene-level association statistics.
Marginal Association Testing: Similar to scDRS, the method computes a marginal disease score for each cell by assessing the weighted expression of putative disease genes, using mean- and variance-matched control genes to generate marginal P-values.
Conditional Association Testing: This is the core innovation. scDRS-FM tests for conditional associations by modeling the relationship between a focal cell's gene expression and the disease signal while conditioning on the expression profiles of all other cell populations in the dataset. This is achieved via a weighted multivariate regression ( $Z \sim X_c + \sum_{c' \neq c} X_{c'}$ ), where $Z$ is the vector of gene-level disease associations and $X$ represents gene expression. To address multicollinearity among highly correlated cell populations, the method applies a partial ridge penalty with data-adaptive regularization parameters to the background cells. It further aggregates cells into "meta-cells" (pseudobulk) to reduce noise and computes conditional P-values using stratified permutation.

The framework outputs both marginal and conditional disease scores and P-values, enabling downstream analyses such as identifying independent disease-critical populations, computing cross-disease functional correlations, and decomposing disease gradients into cellular phenotypes.

Key Contributions and Results
The authors validated scDRS-FM through extensive simulations and real-data analyses involving 75 diseases/traits and 9 scRNA-seq datasets (spanning ~5.8 million cells and 580 cell types).

Simulation and Calibration: In causal simulations with tagging, scDRS-FM maintained well-calibrated false discovery rates (FDR), whereas the marginal method (scDRS) showed inflated false positives. scDRS-FM demonstrated substantially higher statistical power (measured by AUPRC) in identifying causal cells compared to scDRS and scPagwas, effectively removing spurious tagging associations.
Cell Type-Level Analysis (TMS Atlas): Applied to 75 diseases across 120 cell types, scDRS-FM reduced the number of significant cell type-disease associations by filtering out tagging signals. For example, it distinguished causal pancreatic $\beta$ -cell associations with BMI from tagging associations with bipolar disorder and schizophrenia. The results showed strong concordance with orthogonal evidence (e.g., LDSC-SEG) and previous fine-mapping studies, confirming the method's ability to isolate independent causal contributions.
Fine-Grained Resolution (CD4+ T Cells): In a high-resolution analysis of CD4+ T cell subpopulations associated with immune diseases (e.g., IBD), scDRS-FM identified independent disease-critical populations that spanned multiple predefined subtypes. It revealed distinct functional programs, such as a "Multi-Cytokine" phenotype enriched in specific IBD-associated populations, which previous methods failed to resolve due to high correlation between cell states.
Brain Disease Fine-Mapping: Applied to Alzheimer's disease (AD) and other brain traits using the SEA-AD dataset, scDRS-FM fine-mapped associations to specific brain regions and layers. It identified independent populations of microglia (depleted of homeostatic programs) and excitatory neurons (specifically in the midtemporal gyrus) associated with AD, distinguishing them from tagging populations.
Cross-Disease Functional Correlation: The authors demonstrated that scDRS-FM-derived functional correlations capture disease relationships based on shared gene and pathway activity that are not fully reflected by genetic correlations alone, revealing convergence in disease mechanisms beyond simple genetic overlap.

Significance and Claims
The paper claims that scDRS-FM provides a principled and powerful framework for fine-mapping disease-relevant cellular contexts. Its primary significance lies in its ability to:

Disentangle Causal from Tagging: By modeling cell-cell correlations, it distinguishes true causal drivers from cells that merely share expression programs, a limitation of all previous marginal approaches.
Enhance Statistical Power: Through denoising and conditional modeling, it achieves higher power to detect causal cells, particularly in datasets with high heterogeneity or limited sample sizes.
Enable Multi-Resolution Analysis: It allows for the characterization of disease associations not just at the broad cell type level, but across subtypes, spatial regions, and continuous phenotypes, providing a more nuanced view of disease architecture.

The authors emphasize that while the method provides strong statistical evidence for prioritizing cell populations, it does not directly establish biological causality, which requires experimental validation. They also note limitations, including the use of mouse data for human disease inference in some analyses and the computational cost of memory usage due to the denoising step. Nevertheless, the framework represents a significant advancement in resolving the cellular basis of polygenic diseases.

More like this