Learning Pathway Dynamics from Single-Cell Proteomic Data: A Comparative Study.

Area of Science:

• Systems biology
• Computational biology
• Proteomics

Background:

• Single-cell technologies enable large-scale snapshot observations, revealing intercellular heterogeneity.
• Trajectory inference algorithms are established for studying cell proliferation and differentiation dynamics.
• Applying trajectory inference to protein signaling system dynamics remains an underexplored challenge.

Purpose of the Study:

• To assess the efficacy of trajectory inference algorithms for learning protein signaling system dynamics.
• To develop novel, general-purpose metrics for evaluating trajectory inference performance in this context.
• To identify suitable algorithms for inferring signal transduction progression from proteomic temporal data.

Main Methods:

• Tested existing trajectory inference algorithms on four proteomic temporal datasets.
• Designed new evaluation metrics assessing biological meaning, trajectory consistency, robustness, data correlation, and parameter smoothness.
• Compared performance of Scorpius and a novel Diffusion Maps/Principal Curves approach.

Main Results:

• Experimental time alone is insufficient to determine protein order in signal transduction.
• Inferred trajectories provide richer insights into underlying biological dynamics.
• Established methods struggle with high-dimensional, small-sample, slow-dynamics, or complex signaling data; Scorpius and the Diffusion Maps/Principal Curves approach showed adequate performance but varied across metrics.

Conclusions:

• Trajectory inference can enhance understanding of protein signaling dynamics, but limitations exist.
• Novel evaluation metrics are crucial for assessing performance in signaling contexts.
• No single trajectory inference method is universally applicable; further research is needed to address remaining challenges.