pyTWMR: transcriptome-wide Mendelian randomization in python

Area of Science:

• Genetics
• Statistical Genetics
• Computational Biology

Background:

• Mendelian randomization (MR) is a key method for inferring causal relationships between gene expression and complex traits.
• Existing MR approaches have limitations in utilizing multi-SNP instruments and multi-trait exposures for robust causal inference.
• Transcriptome-wide summary statistics-based Mendelian Randomization (TWMR) offers an advanced framework for causal inference.

Purpose of the Study:

• To present a novel, efficient, and robust Python implementation of the TWMR and its reverse version (revTWMR) algorithms.
• To enhance the computational speed and applicability of TWMR for large-scale genetic studies.

Main Methods:

• Developed a Python-based software package, pyTWMR, implementing TWMR and revTWMR.
• Integrated GPU computational support for accelerated data processing.
• Incorporated a robust computation mode to handle highly correlated gene expressions and genetic variants.

Main Results:

• The pyTWMR implementation significantly speeds up TWMR analyses through GPU acceleration.
• The robust computation mode ensures reliable causal inference even with correlated genetic data.
• The software facilitates the use of multiple genetic variants (SNPs) and gene expression traits as exposures.

Conclusions:

• pyTWMR provides an efficient and robust computational tool for performing transcriptome-wide Mendelian randomization.
• This implementation advances the capability to perform causal inference between gene expression and complex traits using observational data.
• The open-source availability of pyTWMR promotes wider adoption and further development in genetic research.