Unsupervised Machine Learning Reveals Temporal Components of Gene Expression in HeLa Cells Following Release from...

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Dynamic biological phenomena, like growth factor responses, involve parallel processes that are difficult to disentangle.
• Analyzing time-series gene expression data requires methods to resolve concurrent biological activities.

Purpose of the Study:

• To apply unsupervised machine learning for deconvolving time-dependent gene expression data into distinct temporal components.
• To identify and characterize underlying biological processes within complex time-series datasets.

Main Methods:

• Utilized publicly available RNA-sequencing (RNAseq) data from synchronized HeLa cells at multiple time points post-cell cycle arrest.
• Employed Fourier analysis and Topic modeling to analyze temporal gene expression patterns.
• Deconvolved time-series data to identify underlying temporal components and their contributions.

Main Results:

• Identified three distinct temporal components within the gene expression data.
• Discovered two oscillatory components corresponding to the G1-S and G2-M phases of the cell cycle.
• Revealed a third transient component linked to immediate early response genes, cell proliferation, and cervical cancer.

Conclusions:

• Unsupervised machine learning effectively identifies hidden temporal dynamics in biological systems.
• This methodology offers potential for early disease detection and monitoring of biological recovery processes.
• The deconvolution of gene expression data provides insights into complex cellular behaviors over time.