Learning causal networks with latent variables from multivariate information in genomic data.

Area of Science:

• Genomics
• Systems Biology
• Computational Biology

Background:

• Learning causal networks from genomic data is difficult without time-series or perturbation data.
• Observational genomic datasets often contain unobserved latent variables that complicate causal inference.

Purpose of the Study:

• To develop an information-theoretic method capable of learning causal graphical models from purely observational genomic data.
• To account for the influence of unobserved latent variables in causal network reconstruction.
• To provide a robust algorithm (miic) that outperforms existing methods.

Main Methods:

• The miic algorithm starts with a complete graph and iteratively removes edges by identifying indirect information paths.
• Edge confidence is assessed through data randomization.
• Remaining edges are oriented using causal signatures present in observational data.

Main Results:

• The miic approach successfully reconstructs causal networks from observational data, including latent variable effects.
• Performance benchmarks show miic surpasses previous methods across various network types.
• The method is validated on biological networks at different scales, from single-cell gene regulation to vertebrate evolution.

Conclusions:

• Miic offers a powerful new approach for inferring causal relationships in genomics using only observational data.
• The method's ability to handle latent variables and its superior performance make it valuable for diverse biological applications.
• Miic is publicly available, facilitating its adoption in systems biology research.