Sparse Regression Based Structure Learning of Stochastic Reaction Networks from Single Cell Snapshot Time Series.

Area of Science:

• Systems Biology
• Computational Biology
• Biochemistry

Background:

• Biological systems rely on stochastic chemical reaction networks for dynamics and variability.
• Experimental limitations often result in incomplete characterization of these network topologies.
• Existing methods for topology refinement are computationally intensive and limited to small systems.

Purpose of the Study:

• To develop a computational method for large-scale structure learning of reaction networks.
• To address limitations of current topology inference methods.
• To enable the discovery of reaction network structures with partial prior knowledge.

Main Methods:

• Proposed the reactionet lasso, a computational procedure based on the Chemical Master Equation.
• Employed a stepwise sparse regression approach to implicitly consider numerous topology variants.
• Assessed performance on synthetic data for the TRAIL-induced apoptosis signaling cascade (70 reactions).

Main Results:

• The reactionet lasso efficiently recovers reaction network structures ab initio with high sensitivity and specificity.
• Achieved < 1% false discoveries, identifying 45% of true reactions among > 6000 possibilities.
• Successfully analyzed over 102,000 network topologies.

Conclusions:

• reactionet lasso enables efficient, large-scale structure learning for biological reaction networks.
• This method can be integrated with single-cell technologies for applications in areas like cancer biology.
• Facilitates the detection of pathological alterations in reaction networks and provides accessible software.