From correlation to causation using directed topological overlap matrix: Applications in genomics.

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Causal discovery algorithms often rely on the local causal Markov condition, which is frequently violated in biological systems.
• Genomic data presents challenges such as measurement errors, averaging effects, and feedback loops, undermining standard causal discovery assumptions.
• Existing methods typically require very large sample sizes, which are often impractical for biological studies.

Purpose of the Study:

• To develop a more flexible and robust causal discovery approach for biological data.
• To address the limitations of the local causal Markov condition in the presence of real-world data complexities.
• To introduce the directed topological overlap matrix (DTOM) as an alternative causal discovery tool.

Main Methods:

• Relaxing the local causal Markov condition and employing Reichenbach's common cause principle.
• Developing the directed topological overlap matrix (DTOM) for causal inference.
• Validating DTOM using three real gene expression datasets: cattle myostatin mutation, yeast gene deletion, and Alzheimer's disease progression.

Main Results:

• DTOM demonstrates robustness against measurement errors, averaging effects, and feedback loops.
• The method is significantly more sample-efficient compared to traditional causal discovery algorithms.
• DTOM successfully identified causal relationships in diverse biological contexts, including distinguishing gene functions and implicating genes in Alzheimer's disease progression.

Conclusions:

• DTOM offers a flexible and powerful alternative for causal discovery in genomics and other biological fields.
• The approach effectively handles common challenges in biological data, leading to more reliable causal inference.
• DTOM's application in Alzheimer's disease research highlights its potential for uncovering disease mechanisms.