Inference of Boolean networks under constraint on bidirectional gene relationships

Area of Science:

• Systems Biology
• Computational Biology
• Network Inference

Background:

• Coefficient of determination (CoD) infers Boolean networks (BNs) from steady-state data, prioritizing predictive gene relationships.
• Unconstrained CoD inference can lead to spurious attractor cycles due to lack of temporal data and resulting bidirectional gene relationships.
• These spurious cycles create unaccounted steady-state probability mass, indicating poor inference relative to network dynamics.

Purpose of the Study:

• To characterize how bidirectional gene relationships impact the attractor structure of Boolean networks.
• To develop and evaluate a constrained CoD inference algorithm that mitigates spurious attractor cycles.
• To compare the performance of the constrained algorithm against unconstrained CoD inference using a melanoma-based network.

Main Methods:

• Characterization of the effect of bidirectional gene relationships on Boolean network attractor structure.
• Development of a constrained coefficient of determination (CoD) inference algorithm.
• Comparative performance analysis of constrained and unconstrained CoD inference algorithms on a melanoma-based network.

Main Results:

• Bidirectional gene relationships were characterized for their impact on Boolean network attractor structure.
• The proposed constrained CoD inference algorithm demonstrated superior performance in avoiding spurious non-singleton attractors compared to unconstrained CoD.
• The constrained algorithm showed improved inference accuracy relative to the attractor structure.

Conclusions:

• Constraining the coefficient of determination (CoD) inference process effectively reduces spurious attractor cycles in Boolean networks.
• The developed constrained CoD algorithm offers a more accurate network inference from steady-state data, particularly in avoiding artifacts like spurious attractors.
• This improved inference method has implications for understanding gene regulatory networks, as demonstrated by its application to a melanoma-based network.