Voting-based integration algorithm improves causal network learning from interventional and observational data: An...

Area of Science:

• Computational Biology
• Network Inference
• Causal Discovery

Background:

• Pooling data from multiple experiments enhances statistical power for causal network inference.
• Uncertainty in intervention effects during data pooling can lead to false causal discoveries.

Purpose of the Study:

• To introduce a novel method, "Learn and Vote," for robust causal interaction inference from multi-experiment datasets.
• To address the challenge of false causal discoveries arising from pooled multi-experiment data.

Main Methods:

• The "Learn and Vote" method involves learning experiment-specific causal networks.
• These networks are then combined using weighted averaging to form a consensus network.

Main Results:

• Empirical studies on synthetic and real-world datasets were conducted.
• The "Learn and Vote" method demonstrated superior accuracy compared to state-of-the-art approaches on larger network datasets.

Conclusions:

• The "Learn and Vote" method offers a more accurate approach to causal network inference from multi-experiment data.
• This method mitigates the risk of false causal discoveries when dealing with pooled experimental data.