Finding semantic patterns in omics data using concept rule learning with an ontology-based refinement operator

Area of Science:

• Bioinformatics
• Computational Biology
• Data Mining

Background:

• Identifying meaningful patterns in omics data is crucial for understanding biological systems.
• Gene Set Enrichment Analysis is a common method but has limitations in pattern evaluation.
• A new framework is needed for discovering complex patterns in 2D binary omics data.

Purpose of the Study:

• To introduce a novel tool/framework for inducing complex patterns in 2D binary omics data.
• To discover and describe semantically coherent biclusters.
• To reveal interpretable hidden rules in omics data that capture semantic differences between classes.

Main Methods:

• A new rapid method called sem1R is presented, inspired by the CN2 rule learner.
• It employs a novel refinement operator that leverages prior knowledge from ontologies.
• The operator includes Redundant Generalization and Redundant Non-potential reduction procedures to prune the rule space.

Main Results:

• The sem1R method reveals interpretable hidden rules in omics data.
• The ontology-based refinement operator significantly speeds up the rule induction process.
• The method effectively captures semantic differences between target and non-target classes.

Conclusions:

• The efficiency and effectiveness of the ontology-based refinement operator were validated on three real gene expression datasets.
• The sem1R algorithm drastically speeds up pattern induction.
• The C++ implementation is available as an R package.