VaxLLM

VaxLLM (Vaccine Large Language Model) is a fine-tuned Large Language Model (LLM) designed to classify and annotate PubMed articles related to Brucella vaccines. It automatically extracts structured vaccine-related metadata from abstracts and full-texts, enabling downstream integration into databases such as VIOLIN. The VaxLLM annotation involves extracting the itemized information of vaccine antigen, vaccine formulation, host species used as animal model, and experiment performed to investigate the vaccine.

Key Features

• Abstract Classification
  Determines whether a given PubMed abstract contains sufficient vaccine formulation details to be considered vaccine-related.

• Structured Annotation
  Extracts key fields:

  • Vaccine Introduction
  • Vaccine Type
  • Vaccine Antigen
  • Vaccine Formulation
  • Host Species Used as Laboratory Animal Model
  • Experiment Used to Investigate the Vaccine
  • Vaccine Efficacy
  • Immume Response
  • Article Results

• Full-text Extraction Support
  Enhances abstract-level annotations with additional details from full-text articles. https://github.com/Hegroup-Bioinformatics/PaperToData

• PubTator Integration
  Adds gene, disease, and other biomedical annotations for enriched metadata.

• Export to Excel
  Clean, structured outputs ready for manual review or database ingestion.

Workflow

End-to-End Pipeline:

Abstract Retrieval → PubMed keyword search (PMIDs, metadata, abstracts)

Abstract Classification → VaxLLM determines if article is vaccine-related

Annotation → Extracts vaccine details into structured fields

Full-text Retrieval (optional) → PMC full-text fetching

Full-text Extraction → Enrich annotations with additional details

PubTator Integration → Adds biomedical entities (genes, diseases, etc.)

Data Harmonization → Standardize fields via Data Harmonizer

Export to database → Convert Excel sheet to database format

Model Details

• Developed by: Xingxian Li at He Group, University of Michigan
• License: MIT
• Finetuned from model: Meta-Llama-3-8B-Instruct
• Task Supported: Text Generation
• Specialized Domain: Brucella vaccine-related content
• Model Type: Fine-tuned Language Model

Sample Codes

Sample code to use VaxLLM: use_VaxLLM

Sample data provided: brucella_articles.txt

Sample code to clean VaxLLM results: clean_result.ipynb

Sample output: VaxLLM_results.xlsx