Essential Regression: A generalizable framework for inferring causal latent factors from multi-omic datasets

Area of Science:

• Computational Biology
• Systems Immunology
• Machine Learning

Background:

• High-dimensional biological data requires advanced analytical methods for multi-omic integration.
• Current methods struggle with data dimensionality, distribution differences, and inferring causality.

Purpose of the Study:

• To develop an interpretable machine learning approach for integrating multi-omic datasets.
• To identify latent factors and their causal relationships with system-wide outcomes.

Main Methods:

• Essential Regression (ER): a novel latent-factor-regression approach.
• Integration of multi-omic datasets without structural or distributional assumptions.
• Coupling ER with probabilistic graphical modeling for enhanced causal inference.

Main Results:

• ER effectively integrates diverse multi-omic data.
• Outperforms state-of-the-art methods in predictive accuracy.
• Demonstrates utility in system immunology, identifying novel causal inferences.

Conclusions:

• ER provides a robust framework for causal inference from multi-omic data.
• Applicable to various biological contexts, including immunosenescence and immune dysregulation.
• Facilitates discovery of molecular and cellular mechanisms underlying complex diseases.