Are we ready for causal discovery in biological systems using deep learning?

Area of Science:

• Causal inference and network biology
• Machine learning for biological systems

Background:

• Causal discovery has advanced significantly over 30 years.
• Applying causal discovery to large-scale, self-regulating biological systems presents challenges.
• Traditional methods often assume global acyclicity, limiting their biological applicability.

Purpose of the Study:

• To highlight emerging neural methods for inferring pairwise causal relationships in biological data.
• To discuss the limitations of current causal discovery approaches for complex biological networks.
• To identify key technological hurdles impeding the application of causal discovery in biology.

Main Methods:

• Leveraging efficient and scalable neural network approaches.
• Inferring pairwise causal relationships directly from data.
• Moving beyond the assumption of global acyclicity in causal models.

Main Results:

• Emerging neural methods offer a path beyond traditional acyclicity assumptions.
• These methods can infer pairwise causal relationships directly from biological data.
• Significant technological challenges remain for comprehensive biological network inference.

Conclusions:

• Advanced neural methods show promise for biological causal discovery.
• Overcoming five key technological hurdles is crucial for realizing the potential of biological causal networks.
• Deeper understanding and stronger inference of biological causal networks are attainable with further development.