MRdb: a comprehensive database of univariate and multivariate Mendelian randomization with large-scale GWAS summary data

Qian Liu, Yujie Zhang, Houxing Li, Jiatong Li, Mengyu Xin, Rui Sun, Yifan Dai, Xinxin Shan, Yuting He, Bochao Xu, Shangwei Ning, Peng Wang, Qiuyan GuoDatabase 2025 · DOI

TL;DR

MRdb is a new, comprehensive database that makes it easier for researchers to perform both simple (univariable) and complex (multivariable) Mendelian randomization analyses. By integrating a vast collection of genetic association data for exposures and outcomes, MRdb aims to accelerate the discovery of causal relationships in complex diseases.

What Problem Does This Paper Address?

Understanding the intricate mechanisms behind complex diseases and developing effective public health interventions relies heavily on large-scale causal inference. Mendelian randomization (MR) is a powerful statistical technique that uses genetic variants as instrumental variables to infer causal links between exposures and outcomes, helping to overcome limitations of traditional observational studies like confounding and reverse causation. However, existing MR tools often struggle with flexibility, particularly for multivariable analyses (where multiple exposures are considered simultaneously), and with effectively integrating diverse, large-scale genetic datasets. This fragmentation and lack of comprehensive integration hinders researchers from fully exploring complex causal pathways.

Methods

MRdb addresses these challenges by providing a unified platform for both univariable and multivariable MR analyses. Key methodological aspects include:

  • Data Standardization and Preprocessing: A rigorous process to harmonize data from various sources. This involves:
  • Normalizing genomic locations to GRCh38.
  • Standardizing key genetic parameters (SNP ID, effect allele, beta, standard error, P-value).
  • Employing statistical imputation for missing values.
  • Calculating beta from odds ratios (OR) using the natural logarithm.
  • Estimating standard errors (SE) when missing, using P-values and chi-square distributions.
  • Deriving P-values from beta and SE using a Z-test approach.
  • Quality Control for Instrumental Variables (IVs):
    • Filtering SNPs based on a P-value threshold (P < 1e-5) for significant association with exposures.
  • Performing Linkage Disequilibrium (LD) clumping to ensure SNP independence.
  • Filtering weak instruments by retaining SNPs with an F-statistic > 10 to mitigate bias.
  • Univariable MR Analysis: Leverages the TwoSampleMR R package, supporting methods like Inverse Variance Weighted (IVW), MR-Egger regression, and Wald ratio. It includes comprehensive sensitivity analyses for heterogeneity (Cochran's Q test) and horizontal pleiotropy (MR-Egger intercept, MR-PRESSO), along with leave-one-out analysis.
  • Multivariable MR (MVMR) Analysis: Utilizes the MVMR R package, supporting IVW and MR-Egger methods to analyze multiple exposures simultaneously. It incorporates LASSO feature selection for IV selection and MR-PRESSO for pleiotropy correction. Sensitivity analyses are also integrated.

Datasets

MRdb integrates a substantial collection of publicly available GWAS summary data:

  • Exposure Data: Encompasses 12 distinct categories from high-quality GWAS sources, including:
  • 19,126 expression quantitative trait loci (eQTL) genes.
  • 4,907 plasma proteins.
  • 1,400 plasma metabolites.
  • Other categories like immune cell types, blood metabolites, microbiome components, plasma lipids, circulating proteins, blood cell types, inflammatory factors, and thyroid hormones.
  • Outcome Data: Integrates 48,507 disease outcomes sourced from:
  • FinnGen R10 (approximately 5% of outcomes).
  • The IEU Open GWAS Project (approximately 95% of outcomes).

Collectively, MRdb contains over 27,000 exposures and 48,000 outcomes, representing data from more than 1.14 million individuals.

Key Insights and Ideas

  • Bridging the Gap in Multivariable MR: The primary innovation is the creation of a user-friendly database that specifically supports multivariable MR. This allows researchers to move beyond single-exposure analyses and investigate complex interactions between multiple risk factors for a given disease.
  • Comprehensive Data Integration: By consolidating diverse exposure types (genetics, proteins, metabolites) with a wide array of disease outcomes, MRdb provides a powerful resource for hypothesis generation and testing across various biological domains.
  • Emphasis on Robustness and Sensitivity: The inclusion of extensive preprocessing, quality control for IVs, and a suite of sensitivity analyses (heterogeneity, pleiotropy, leave-one-out) is crucial for ensuring the reliability and validity of MR findings. This helps researchers identify and mitigate potential biases.
  • User-Friendly Interface: The database is designed with user-friendliness in mind, offering a web interface and downloadable datasets, making advanced MR analyses more accessible.

Key Results

The paper demonstrates the utility of MRdb through a technical validation example, investigating the relationship between interleukin-10 (IL-10) and gestational diabetes mellitus (GDM).

  • MRdb estimated a causal effect between IL-10 gene expression and GDM, with an odds ratio (OR) of 1.09 (95% CI: 1.01-1.28) and a P-value of 1.22e-3.
  • The platform successfully applied both IVW and MR-Egger regression methods, showcasing its ability to identify and adjust for potential horizontal pleiotropy.
  • Various sensitivity analyses, including heterogeneity testing, funnel plots, forest plots, and leave-one-out analyses, were performed to assess the robustness of the findings.

Limitations and Future Work

  • Data Coverage: While MRdb is extensive, the authors acknowledge that the continuous generation of biomedical big data means the database will require regular updates to incorporate the latest GWAS findings.
  • Future Work: The authors suggest that the complementary strengths of MRdb with other existing databases (like PhenoScanner, MR-Base, GWAS Catalog, DMRdb) will collectively advance the field of MR. Future development will likely focus on expanding data coverage and potentially incorporating more advanced analytical methods. The platform is designed to be regularly updated, implying ongoing efforts to enhance its scope and capabilities.

Summary updated

Abstract

Recent advancements highlight the importance of large-scale causal inference in elucidating disease mechanisms and guiding public health strategies. Mendelian randomization (MR) has become a cornerstone method for identifying causal relationships by leveraging genetic variants as instrumental variables. However, existing tools lack flexibility for multivariable analyses and fail to integrate diverse datasets effectively. To address these challenges, we introduce MRdb, a comprehensive database designed for conducting both univariable and multivariable MR analyses. MRdb encompasses 12 distinct categories of exposure data, including but not limited to 19 126 expression quantitative trait loci genes, 4907 plasma proteins, and 1400 plasma metabolites. Additionally, it integrates 48 507 disease outcomes sourced from FinnGen R10 and the IEU Open GWAS Project. MRdb offers robust data preprocessing features, including handling missing statistics, harmonizing datasets, and selecting instrumental variables to ensure high-quality analyses. Collectively, MRdb bridges the gaps in existing tools by integrating diverse datasets with user-friendly functionalities, empowering researchers to explore complex causal mechanisms.
BibTeX
@inproceedings{liu2025mrdb,
  author = {Qian Liu and Yujie Zhang and Houxing Li and Jiatong Li and Mengyu Xin and Rui Sun and Yifan Dai and Xinxin Shan and Yuting He and Bochao Xu and Shangwei Ning and Peng Wang and Qiuyan Guo},
  title = {{MRdb: a comprehensive database of univariate and multivariate Mendelian randomization with large-scale GWAS summary data}},
  booktitle = {Database},
  year = {2025},
  doi = {10.1093/database/baaf054},
}
Figure