Classification of Single-Cell Gene Expression Trajectories from Incomplete and Noisy Data

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Gene expression dynamics are crucial for understanding cellular states, distinguishing between healthy and diseased conditions like cancer.
• Boolean networks with perturbation (BNps) offer a framework for modeling gene regulatory network dynamics at a discrete state level.

Purpose of the Study:

• To develop a computational method for classifying gene-expression trajectories from healthy and cancerous cells.
• To compare the efficacy of single-cell versus multiple-cell averaged expression data for classification.

Main Methods:

• Modeling gene expression dynamics using Boolean networks with perturbation (BNps) for two distinct classes.
• Employing a Gaussian observation model for gene expression values based on hidden binary states.
• Utilizing Expectation-Maximization (EM) algorithm for learning BNps and model parameters.
• Developing a plug-in Bayes classifier for trajectory classification, accommodating missing data.
• Proposing a distinct model for multiple-cell expression data based on the central limit theorem.

Main Results:

• The Expectation-Maximization (EM) algorithm was successfully adapted with closed-form updates for parameter learning.
• The developed classifier effectively handles gene expression trajectories, including those with missing data points.
• Simulations indicated that single-cell gene expression trajectory data yields lower classification error compared to multiple-cell averaged data, particularly under high-noise conditions.
• Performance was validated using data from a mammalian cell-cycle network, including wild-type and mutated (p27) scenarios.

Conclusions:

• Boolean networks with perturbation (BNps) provide a viable model for gene expression dynamics in healthy and cancerous states.
• Single-cell gene expression trajectory analysis is a more robust approach for classification than using averaged data, especially in noisy biological systems.
• The proposed computational framework enhances the ability to classify cellular states based on gene expression patterns.