Knowledge Base Module

The Knowledge Base (KnowledgeBase) is the relational and vector database abstraction layer for ccm_benchmate. Built on PostgreSQL 17+, it leverages pgvector for high-dimensional vector embeddings (literature text chunks, figures, tables) and rdkit for small molecule structure indices and fingerprint similarity.

The database tables are organized around the central project table, associating stored entities with a specific project ID.

Usually, end users do not query KnowledgeBase directly, but interact with it seamlessly through the Project meta-module and project.search. For step-by-step examples of storing and querying entities in the Knowledge Base, see Project Workflow Examples.

Below is the relational database schema:

Database Extensions & Requirements:

• PostgreSQL: Version 17+
• pgvector Extension: Enables <-> vector similarity operations for literature embeddings.
• rdkit Extension: Enables chemical SMILES indexing and Tanimoto similarity operations.

Note for Developers:

The KnowledgeBase layer abstracts SQLAlchemy table creation (_create_kb) and session management for the Project module. Direct interaction is managed via project.sequence.to_kb(), project.paper.to_kb(), project.molecule.to_kb(), project.apis.call_class.to_kb(), and project.sequence_variant.to_kb().