Sets2Networks: network inference from repeated observations of sets

Area of Science:

• Complex Systems Research
• Network Science
• Computational Biology

Background:

• Complex systems are often modeled as networks, but inferring these networks from direct interactions is challenging.
• Existing network inference methods primarily focus on quantitative data, overlooking co-occurrence data.
• Co-occurrence data is prevalent in systems biology and other complex systems research, highlighting a need for specialized inference methods.

Purpose of the Study:

• To present a general method for network inference using repeated observations of related entity sets.
• To infer underlying networks by generating an ensemble of networks consistent with observed data.
• To interpret link frequency in the ensemble as the probability of link presence in the real network.

Main Methods:

• Utilized exponential random graphs to generate and sample an ensemble of consistent networks.
• Employed an algorithmic approach for numerical execution of the network inference method.
• Validated the method on synthetic data before application to real-world problems.

Main Results:

• Successfully applied the method to infer protein-protein interactions from proteomics data.
• Integrated diverse datasets (Chip-seq, gene expression) to infer regulatory networks in pluripotency.
• Extracted a network of cancer drugs and adverse events from FDA AERS data.
• Constructed a co-authorship network of researchers at Mount Sinai School of Medicine.

Conclusions:

• The presented network inference method is versatile and applicable to various network types.
• The approach is valuable for systems biology, systems pharmacology, and other research fields.
• Online software is available for creating networks from entity-set libraries.