How to Give Your AI Agent Access to PubMed Data

import alterlab
from alterlab import Client

client = Client("YOUR_API_KEY")

# Define schema for PubMed article structure
schema = {
    "title": "string",
    "authors": "string",
    "journal": "string",
    "pub_date": "string",
    "abstract": "string",
    "doi": "string"
}

# Extract structured data from a PubMed article URL
result = client.extract(
    url="https://pubmed.ncbi.nlm.nih.gov/34567890/",
    schema=schema
)

# Result.data is a dict, ready for LLM context or RAG pipeline
print(result.data)

curl -X POST https://api.alterlab.io/api/v1/extract \
  -H "X-API-Key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://pubmed.ncbi.nlm.nih.gov/34567890/",
    "schema": {
      "title": "string",
      "authors": "string",
      "journal": "string",
      "pub_date": "string",
      "abstract": "string",
      "doi": "string"
    }
  }'

import alterlab
from alterlab import Client

client = Client("YOUR_API_KEY")

# Search PubMed for recent articles on cancer immunotherapy
search_params = {
    "query": "cancer immunotherapy 2024",
    "site": "pubmed.ncbi.nlm.nih.gov",
    "num_results": 10
}

response = client.search(**search_params)

# Response contains structured list of articles
for article in response.data:
    print(f"{article['title']} - {article['journal']}")

curl -X POST https://api.alterlab.io/api/v1/search \
  -H "X-API-Key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "cancer immunotherapy 2024",
    "site": "pubmed.ncbi.nlm.nih.gov",
    "num_results": 10
  }'

import alterlab
from alterlab import Client
import hashlib
import json
from datetime import datetime, timedelta

# Initialize client (in production, load API key from secure vault)
client = Client("YOUR_API_KEY")

# Track seen articles to avoid duplicates
SEEN_ARTICLES_FILE = "seen_articles.json"

def load_seen_articles():
    try:
        with open(SEEN_ARTICLES_FILE) as f:
            return set(json.load(f))
    except FileNotFoundError:
        return set()

def save_seen_articles(seen_set):
    with open(SEEN_ARTICLES_FILE, "w") as f:
        json.dump(list(seen_set), f)

def monitor_diabetes_research():
    seen = load_seen_articles()
    
    # Search for new diabetes articles from last 7 days
    seven_days_ago = (datetime.now() - timedelta(days=7)).strftime("%Y/%m/%d")
    search_query = f"type 2 diabetes treatment {seven_days_ago}[Date - Publication] : 3000[Date - Publication]"
    
    search_response = client.search(
        query=search_query,
        site="pubmed.ncbi.nlm.nih.gov",
        num_results=20
    )
    
    new_articles = []
    for article in search_response.data:
        # Create unique ID from PMID or DOI
        article_id = article.get("pmid") or article.get("doi") or hashlib.md5(article["title"].encode()).hexdigest()
        
        if article_id not in seen:
            seen.add(article_id)
            
            # Extract full structured data for new article
            extract_result = client.extract(
                url=article["url"],
                schema={
                    "title": "string",
                    "authors": "string",
                    "journal": "string",
                    "pub_date": "string",
                    "abstract": "string",
                    "doi": "string"
                }
            )
            
            new_articles.append(extract_result.data)
    
    # Update knowledge base with new articles (pseudo-code)
    if new_articles:
        update_knowledge_base(new_articles)
        save_seen_articles(seen)
        print(f"Added {len(new_articles)} new diabetes research articles to knowledge base")
    else:
        print("No new articles found")

def update_knowledge_base(articles):
    # In practice: embed abstracts and store in vector DB (e.g., Pinecone, Weaviate)
    pass

if __name__ == "__main__":
    monitor_diabetes_research()