Benchmate Modules

This is the documentation for usage instructions for all modules in the ccm_benchmate package. For a technical API reference, please see the API Reference.

Modules can be used independently or unified under the Project meta-module:

• APIs: Public biological database query clients (UniProt, NCBI, Ensembl, StringDB, IntAct, etc.).
• Literature: Paper search, PDF parsing, chunking, and vision-language figure/table interpretation.
• Sequence: Representation and mutation of DNA, RNA, protein, and 3Di sequences.
• Structure: PDB and AlphaFold 3D structure handling.
• Molecule: Small molecule handling with RDKit SMILES parsing and ECFP4/FCFP4 fingerprints.
• Genome: Genomic feature annotation and sequence retrieval.
• Alignment: High-speed sequence and structure homology searches (BLAST, MMseqs2, Foldseek, Folddisco).
• Ranges: Genomic interval operations.
• Variant: Sequence, structural, and tandem repeat variant models.
• KnowledgeBase: PostgreSQL (17+) schema management with pgvector and rdkit.
• Project: Meta-module unifying database creation, entity persistence (to_kb), item listing (list_items), exact object retrieval (from_kb), and multimodal search (project.search).

For a complete tutorial combining all modules into an integrated workflow, see Project Workflow & Usage Examples.

Configuration

Benchmate relies on AI models specified in config.yaml. The default models (e.g. Qwen vision-language and embedding models) are selected to ensure accuracy while operating within standard single-GPU VRAM limits (<40GB).

---

Table of contents

• APIs
• Literature
• Genome
• Sequence
• Structure
• Molecule
• Alignment
• Ranges and GenomicRanges
• Variants
• Inference
• Project Meta Module
• Knowledge Base