MRSL: a causal network pruning algorithm based on GWAS summary data.

Area of Science:

• Genetics
• Causal Inference
• Bioinformatics

Background:

• Observational data analysis is crucial for understanding biological systems.
• Genetic variants offer complementary insights for causal structure learning.
• Mendelian randomization (MR) studies have identified numerous marginal causal relationships.

Purpose of the Study:

• To develop a novel causal network pruning algorithm, MRSL (MR-based structure learning algorithm).
• To leverage marginal causal relationships from MR studies for enhanced structure learning.
• To infer conditional causal structures using only genome-wide association studies (GWAS) summary statistics.

Main Methods:

• MRSL integrates graph theory with multivariable MR.
• The algorithm employs topological sorting for improved structure learning precision.
• MRSL introduces MR-separation and candidate separating sets, replacing traditional d-separation.

Main Results:

• Simulations show MRSL achieves up to a 2-fold higher F1 score and is 100 times faster than competing methods.
• Application to UK Biobank GWAS data for 26 biomarkers and 44 diseases identified expected and novel causal links.
• The identified links possess biological interpretations and are supported by existing literature or clinical reports.

Conclusions:

• MRSL is an efficient and precise algorithm for causal discovery from GWAS summary statistics.
• The method effectively identifies biologically relevant causal relationships between traits and diseases.
• MRSL advances the field of causal inference by integrating genetic data and advanced graph algorithms.