Machine Learning of Stem Cell Identities From Single-Cell Expression Data via Regulatory Network Archetypes.

Area of Science:

• Stem cell biology
• Systems biology
• Computational biology

Background:

• The molecular regulatory network governing stem cell pluripotency is well-modeled for average cells.
• Significant cell-to-cell variability exists, suggesting dynamic network activity within individual stem cells.
• This variability impacts environmental signal processing and cell fate determination.

Purpose of the Study:

• To infer regulatory network patterns within individual stem cells using single-cell expression data.
• To identify distinct network configurations and their association with specific cell states.
• To demonstrate the application of machine learning in understanding single-cell biology.

Main Methods:

• Adaptation of a machine learning method originally developed for face recognition.
• Analysis of single-cell gene expression data from cultured mouse embryonic stem cells.
• Identification and classification of regulatory network configurations.

Main Results:

• Three distinct regulatory network configurations were identified in mouse embryonic stem cells.
• These configurations correspond to naïve pluripotent, formative pluripotent, and early primitive endoderm states.
• Network configurations were associated with specific regulatory activity archetypes governing cell response, cell cycle, and information processing.

Conclusions:

• Cellular identity variability arises naturally from altered regulatory network dynamics.
• Machine learning methods can effectively elucidate single-cell biology and cell community dynamics.
• Understanding network variability is crucial for deciphering stem cell behavior and differentiation.