Simpati

Simpati classifies patients using pathway-specific patient similarity networks to identify signature biological pathways associated with disease phenotypes.

Key Features:

• Pathway-Specific Patient Similarity Networks: Constructs pathway-specific patient similarity networks that organize patient molecules into pathway-centered networks for interpretable comparison.
• Interconnected Data Propagation: Applies a propagation algorithm to standardize high-throughput datasets and infer new activity scores that capture interconnected cellular molecule behavior.
• Novel Similarity Measure: Implements a unique similarity measure that quantifies how patient pairs coordinate within pathways by analyzing topological properties of patient networks.
• Cohesive Subgroup Detection Algorithm: Detects cohesive subgroups of patients to handle atypical pathway activity and improve robustness of classification.
• Recommender System for Classification: Uses a recommender-system approach to classify unknown patients by assessing similarity to known patients and identifying outliers.
• Performance and Efficiency: Demonstrates improved performance across five cancer datasets using two biological omics, with advantages in sparse data scenarios and reduced computational requirements relative to compared classifiers.

Scientific Applications:

• Patient Classification: Classifies patient cohorts based on pathway-specific similarity patterns to distinguish clinical or phenotypic groups.
• Signature Pathway Identification: Identifies signature biological pathways that characterize disease phenotypes by detecting coordinated pathway perturbations.
• Mechanistic Insight into Disease: Reveals altered biological mechanisms underlying phenotypes by highlighting pathways that are differentially involved between groups.
• Upinvolved and Downinvolved Pathway Distinction: Distinguishes upinvolved pathways, where signaling cascades are more active in diseased patients, from downinvolved pathways with reduced activity.

Methodology:

Organizes patient molecules into pathway-specific patient similarity networks, applies a propagation algorithm to standardize data and infer activity scores, computes a novel similarity measure via topological analysis, detects cohesive patient subgroups, and classifies unknown patients using a recommender-system approach.