R.ROSETTA

R.ROSETTA implements interpretable rule-based classification and statistical characterization of models for bioinformatics datasets using rough set theory.

Key Features:

• Rule-Based Classification Models: Builds interpretable, non-linear classifiers by inducing rules from data using rough set theory.
• Rough Set Theory: Applies rough set theory to handle vagueness and uncertainty in decision-table formatted data.
• Combinatorial Statistics: Computes combinatorial statistics for rules to quantify support and statistical characteristics of model components.
• Data Balancing (Undersampling): Performs data balancing through undersampling to address class imbalance.
• Rule P-value Estimation: Estimates P values for individual rules to assess statistical significance.
• Support Set Retrieval: Retrieves support sets associated with induced rules.
• Model Merging: Merges rule-based models to consolidate rule sets across analyses.
• Prediction of External Classes: Predicts class labels for external or new samples using induced rule sets.
• Visualization Tools: Provides visualization routines to inspect rules and model structure.
• Synthetic Data Generation: Generates synthetic data for analysis and testing.
• Statistical Properties Reporting: Reports statistical properties for models and their components to aid interpretation.

Scientific Applications:

• Life sciences interpretability: Applied where interpretable machine learning is required to explain predictive mechanisms in biological data.
• Transcriptome analysis (autism case-control): Used on transcriptome datasets from autism case-control studies to generate hypotheses on co-predictive feature mechanisms that distinguish phenotype classes and to identify neurodevelopmental and autism-related genes.

Methodology:

The methodology uses rough set theory and rule-based modeling on decision-table formatted data, with explicit computational procedures including combinatorial statistics for rule evaluation, undersampling for data balancing, estimation of rule P values, retrieval of rule support sets, model merging, prediction of external classes, visualization routines, and synthetic data generation.