DeepCausality

DeepCausality infers and estimates causal factors from free-text documents using AI-powered language models and Judea Pearl's Do-calculus to support causal analysis in biomedical and scientific text.

Key Features:

• AI-powered language models: Leverages advanced language models to extract and identify causal terms embedded within large volumes of unstructured text.
• Named Entity Recognition (NER): Employs NER techniques to identify and classify entities within text to facilitate determination of potential causal relationships.
• Judea Pearl's Do-calculus: Integrates Do-calculus to apply formal mathematical principles for causal inference.
• Knowledge-based causal tree generation: Produces a knowledge-based causal tree used for patient stratification in domain-specific applications.
• Performance metrics: In application to idiosyncratic drug-induced liver injury (DILI) from the LiverTox database, reported accuracy is 0.92, F-score is 0.84, 90% of identified causal terms matched American College of Gastroenterology (ACG) guidelines, and concordance with domain expert iDILI severity scores was 0.91.

Scientific Applications:

• LiverTox / DILI causal term estimation: Applied to the LiverTox database to estimate causal terms related to idiosyncratic drug-induced liver injury (DILI).
• Patient stratification: Generated a knowledge-based causal tree for patient stratification in biomedical research.
• iDILI severity scoring and validation: Produced iDILI severity scores with a reported concordance of 0.91 against domain expert evaluations and 90% consistency with ACG guideline terms.

Methodology:

Combines AI-powered language models with Named Entity Recognition (NER) and Judea Pearl's Do-calculus within a unified computational framework.