Causal machine learning for single-cell genomics.

Area of Science:

• Genomics
• Machine Learning
• Systems Biology

Background:

• Single-cell '-omics' technologies provide high-resolution transcriptional data from individual cells.
• Large-scale perturbation screens reveal gene effects on transcriptomes.
• Integrating these methods offers opportunities to elucidate gene causality in biological processes.

Purpose of the Study:

• To delineate the application of causal machine learning (ML) to single-cell genomics.
• To identify and discuss challenges and potential solutions in this emerging field.

Main Methods:

• Review and synthesis of current causal ML models used in single-cell biology.
• Identification of key challenges including model generalization, interpretability, and learning cell dynamics.

Main Results:

• The commonly applied causal model in single-cell biology is presented.
• Three critical open problems are discussed: lack of generalization to novel conditions, model interpretation complexity, and difficulties in learning cell dynamics.

Conclusions:

• Causal ML holds significant promise for advancing single-cell genomics research.
• Addressing the identified challenges is crucial for unlocking the full potential of these integrated approaches.