FOntCell

FOntCell merges complementary biological ontologies to fuse semantic and structural information for improved cell classification and characterization in high-throughput single-cell RNA-Seq datasets.

Key Features:

• Automatic Ontology Fusion: Automates fusion of multiple classification ontologies via name-matching of class label names and structural mappings between ontology elements.
• Graph Convolution Techniques: Applies graph convolution methods including vectorial structure matching to compute similarities between graph-representing vectors of different sizes and a constraint-based structure matching adapted from the Blondel method.
• Semantic and Structural Integration: Combines semantic attributes and graph structure by processing ontology attributes to extract relations and class synonyms for informed mapping and merging.
• Output Formats and Iterative Adaptation: Emits fused ontologies in OBO format and supports iterative reuse to incorporate new data and updates such as those from the Human Cell Atlas.

Scientific Applications:

• Cell Ontology Unification: Merges existing cell ontologies such as CELDA, LifeMap, and LungMAP Human Anatomy to produce unified ontologies for improved cell classification.
• Discovery of New Cell Types: Supports discovery and classification of novel cell types by integrating diverse ontology sources and enriching annotations for high-throughput datasets.

Methodology:

Implemented in Python, FOntCell performs parallel-computed fusion using name-matching searches on class label names and structural mappings between ontology elements; it leverages graph convolution techniques (vectorial structure matching and constraint-based Blondel adaptation), processes ontology attributes to extract relations and class synonyms, and outputs fused ontologies in OBO format.