NEMix: single-cell nested effects models for probabilistic pathway stimulation.

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Nested effects models (NEMs) are used for inferring subcellular networks from RNA interference (RNAi) data.
• Single-cell RNAi screens generate high-dimensional data with significant cell-to-cell variation, complicating network inference due to weak average phenotypes.
• Unobserved cellular states, like pathway activation, can confound network structure learning.

Purpose of the Study:

• To develop an advanced nested effects model that explicitly accounts for unobserved pathway activation in single-cell RNAi screens.
• To improve the accuracy of subcellular network inference in the presence of cellular heterogeneity and hidden pathway stimulation.
• To apply the novel model to analyze human rhinovirus infection data from single-cell imaging.

Main Methods:

• Proposed NEMix, a nested effects mixture model extending the NEM framework for probabilistic combinatorial knock-downs.
• Developed a parameter inference scheme using the Expectation Maximization algorithm.
• Analyzed model identifiability and performance through extensive simulation studies.

Main Results:

• NEMix significantly improves pathway structure learning compared to classical NEMs when hidden pathway stimulation is present.
• Application to human rhinovirus infection data demonstrated high accuracy in network inference.
• The inferred NEMix network outperformed the classical NEM in accuracy, particularly with uncertain pathway stimulation.

Conclusions:

• NEMix effectively addresses cellular heterogeneity and unobserved pathway activation in single-cell RNAi screens.
• The model provides a more accurate approach to subcellular network inference from complex biological data.
• NEMix is available as part of the R/Bioconductor package 'nem'.