nem

nem reconstructs signaling networks de novo from time-course perturbation data to elucidate protein interactions and regulatory pathways relevant to cancer research.

Key Features:

• Input Requirements: Accepts pathway components subjected to perturbations together with phenotypic readouts such as gene or protein expression levels.
• Output Generation: Produces a directed graph representing the hierarchical structure of phenotypic responses and inferred signaling influences.
• Network Modeling: Employs network modeling to represent molecular interactions and signaling dependencies.
• State Transition Modeling: Models signaling dynamics by defining active and passive states for each protein across discrete time points and generating possible state transitions under a fixed signal propagation scheme.
• Likelihood Scoring: Calculates a likelihood score representing the probability of observed measurements given temporal protein states.
• Optimization Algorithms: Uses a Hidden Markov Model to determine optimal state sequences and a Genetic Algorithm to search network structures by optimizing overall likelihood across candidate networks.
• Data Types: Applied to time-course experimental data including phosphorylated protein abundances measured by reverse phase protein arrays (RPPA).
• Benchmarking: Demonstrated performance advantages over dynamical Bayesian network approaches on human breast cancer cell line data, including recovery of ERBB pathway signaling cascades.
• Implementation: Implemented in the R programming language.

Scientific Applications:

• De novo signaling reconstruction: Reconstruction of signaling networks from perturbation time-course experiments in systems biology.
• Cancer signaling analysis: Analysis of protein interaction dynamics and regulatory pathways in cancer research, including human breast cancer cell lines.
• Phosphoproteomics interpretation: Interpretation of phosphorylated protein abundance data from reverse phase protein arrays to infer pathway activity.
• Pathway cascade identification: Identification of signaling cascades such as components of the ERBB pathway.
• Therapeutic hypothesis generation: Supporting the identification of regulatory aberrations that can inform potential therapeutic targets.

Methodology:

Reconstructs signaling networks from time-course experimental data following external perturbations; defines active/passive protein states across discrete time points and generates possible state transitions under a fixed signal propagation scheme; computes likelihood scores of observed measurements given temporal states; applies a Hidden Markov Model to infer optimal state sequences and a Genetic Algorithm to search network structures by maximizing overall likelihood; applied to RPPA-measured phosphorylated protein abundance datasets; implemented in R.