PathBinder

PathBinder extracts biomolecular interactions from biomedical literature, including PubMed, using machine learning and empirical text analysis to identify interaction descriptions and quantify interaction likelihoods.

Key Features:

• Automatic Interaction Extraction: Automatically extracts interaction statements from free-text biomedical literature, including PubMed abstracts and articles, by identifying syntactic and semantic properties.
• Empirical Text Analysis: Implements "text empirics" via manual examination of syntactic and semantic properties to inform algorithm design for detecting interaction-describing passages.
• Probabilistic Methods: Applies probabilistic methods to synthesize evidence across sources and compute likelihoods of interactions between specified biomolecules.
• Biomolecular Interaction Network Construction: Builds networks of biomolecular interactions with edges weighted by computed interaction likelihoods to represent metabolic and regulatory relationships.

Scientific Applications:

• Metabolic Pathway Analysis: Annotates metabolic interactions to support reconstruction and analysis of biochemical pathways, including plant metabolism.
• Regulatory Interaction Studies: Annotates regulatory interactions relevant to gene expression regulation and signal transduction.
• Literature Mining and Data Integration: Enables large-scale literature mining to integrate interaction evidence from disparate sources for reviews and meta-analyses.

Methodology:

Combines manual examination of syntactic and semantic properties ("text empirics") with machine learning techniques to identify interaction-describing passages in PubMed texts, uses probabilistic methods to synthesize evidence from multiple sources, and constructs biomolecular interaction networks weighted by inferred interaction likelihoods.