How jiaxianyan/biominer works
BioMiner turns scientific-paper PDFs into protein–ligand–bioactivity records, with a related structure-to-bioactivity annotation task. The documented workflow covers document preprocessing, chemical-structure extraction and recognition, bioactivity extraction, and result postprocessing; the UI supports reviewing and editing results. Edges reflect the README’s stated stages and the sampled interface/API evidence. Several implementation details are unsampled, so uncertain internal calls and data stores are omitted.
Main entry points: BioMiner/interface.py BioMiner orchestrator · biovista_component_evaluate.py BioVista benchmark · process_pdf.py PDF preprocessing · molscribe/interface.py Structure recognition
Components 22
Extraction workflow
- BioMiner orchestrator
BioMiner/interface.py - PDF preprocessing
BioMiner/commons/process_pdf.py - MinerU integration
BioMiner/commons/mineru_pdf.py
Structure processing
- Molecule detection
BioMiner/commons/mol_detection.py - Structure recognition
BioMiner/MolScribe/molscribe/interface.py - MolScribe model
BioMiner/MolScribe/molscribe/model.py - Structure utilities
BioMiner/commons/process_mol.py - Coreference and Markush
BioMiner/interface.py
Bioactivity extraction
- Bioactivity extraction
BioMiner/runner/mllm.py
Results and review
- Result extraction
BioMiner/runner/extractor.py - Protein utilities
BioMiner/commons/protein_utils.py - Review API
BioMiner_UI/backend/app.py - Review interface
BioMiner_UI/frontend/static/app.js - Extraction outputs
Benchmark tasks
- BioVista benchmark
biovista_component_evaluate.py - Detection task
BioVista/component_infernece/molecule_detection/moldetect_inference_batch.py - OCSR task
BioVista/component_infernece/ocsr/molscribe_inference_batch.py - Coreference task
BioVista/component_infernece/full_coreference/coreference_inference_batch_with_index_split_image_layout.py - Markush task
BioVista/component_infernece/markush_enumerate/markush_zip_inference_batch_with_index_layout.py
Other components
- Researcher
- Paper PDFs
- Language model API
Connections 21
- Researcher → BioMiner orchestrator: submits papers
- Paper PDFs → BioMiner orchestrator: provides PDFs
- BioMiner orchestrator → PDF preprocessing: preprocesses papers
- PDF preprocessing → MinerU integration: extracts pages
- BioMiner orchestrator → Molecule detection: detects molecules
- BioMiner orchestrator → Structure recognition: recognizes structures
- Structure recognition → MolScribe model: predicts structures
- BioMiner orchestrator → Coreference and Markush: resolves structures
- Coreference and Markush → Structure utilities: processes chemistry
- BioMiner orchestrator → Bioactivity extraction: extracts measurements
- Bioactivity extraction → Language model API: requests extraction
- BioMiner orchestrator → Result extraction: postprocesses results
- Result extraction → Protein utilities: uses protein data
- BioMiner orchestrator → Extraction outputs: writes results
- Researcher → Review interface: reviews results
- Review interface → Review API: calls API
- Review API → Extraction outputs: reads and edits
- BioVista benchmark → Detection task: evaluates detection
- BioVista benchmark → OCSR task: evaluates OCSR
- BioVista benchmark → Coreference task: evaluates coreference
- BioVista benchmark → Markush task: evaluates enumeration