Causal discovery using compression-complexity measures.

Area of Science:

• Causal inference
• Sequence analysis
• Computational biology

Background:

• Causal inference is a fundamental scientific challenge.
• Inferring causal direction from discrete symbolic sequences is complex.
• Existing methods often require temporal structures.

Purpose of the Study:

• To develop a novel framework for inferring causal direction between two discrete symbolic sequences.
• To utilize lossless compressors and context-free grammars (CFGs) for causal discovery.
• To apply the framework to genome sequence analysis, specifically SARS-CoV-2.

Main Methods:

• Framework based on lossless compressors to infer CFGs from sequence pairs.
• Quantification of grammar-based compression of one sequence by another.
• Proposed three models using Compression-Complexity Measures (CCMs) like Lempel-Ziv (LZ) complexity and Effort-To-Compress (ETC).

Main Results:

• Models infer causal directions without requiring temporal structures.
• Empirical evaluation on synthetic and real-world benchmarks shows competitive performance.
• Demonstrated successful application in inferring causal information from SARS-CoV-2 genome sequences.

Conclusions:

• The proposed framework provides a novel approach to causal inference from symbolic sequences.
• CCMs offer a powerful tool for discovering causal relationships in biological sequences.
• This method opens new avenues for studying viral evolution, virulence, and pathogenicity.