import gradio as gr import requests import json import pandas as pd from datetime import datetime, timedelta import re from typing import List, Dict, Tuple import xml.etree.ElementTree as ET from collections import Counter import plotly.express as px import plotly.graph_objects as go from transformers import pipeline import numpy as np class CancerResearchLiteratureMiner: def __init__(self): # Initialize NLP pipelines try: self.summarizer = pipeline("summarization", model="facebook/bart-large-cnn") self.classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli") except Exception as e: print(f"Warning: Could not load transformers models: {e}") self.summarizer = None self.classifier = None # Research categories for classification self.research_categories = [ "drug discovery", "immunotherapy", "chemotherapy", "radiation therapy", "biomarkers", "diagnostics", "metastasis", "tumor microenvironment", "animal models", "preclinical studies", "toxicity", "pharmacokinetics" ] # Animal model keywords self.animal_keywords = [ "mouse", "mice", "rat", "rats", "xenograft", "orthotopic", "transgenic", "knockout", "immunodeficient", "nude mice", "SCID", "NOD", "PDX", "patient-derived xenograft", "syngeneic", "canine", "dog", "feline", "cat" ] def search_pubmed(self, query: str, max_results: int = 50) -> List[Dict]: """Search PubMed for cancer research papers""" # Enhance query with animal model terms enhanced_query = f"({query}) AND (animal model OR mouse OR mice OR rat OR xenograft OR preclinical)" # Search PubMed search_url = "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi" search_params = { "db": "pubmed", "term": enhanced_query, "retmax": max_results, "retmode": "json", "sort": "relevance" } try: search_response = requests.get(search_url, params=search_params) search_data = search_response.json() if "esearchresult" not in search_data or not search_data["esearchresult"]["idlist"]: return [] # Get detailed information ids = search_data["esearchresult"]["idlist"] fetch_url = "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi" fetch_params = { "db": "pubmed", "id": ",".join(ids), "retmode": "xml" } fetch_response = requests.get(fetch_url, params=fetch_params) # Parse XML response papers = self._parse_pubmed_xml(fetch_response.text) return papers except Exception as e: return [{"error": f"Search failed: {str(e)}"}] def _parse_pubmed_xml(self, xml_content: str) -> List[Dict]: """Parse PubMed XML response""" papers = [] try: root = ET.fromstring(xml_content) for article in root.findall(".//PubmedArticle"): paper = {} # Extract basic info medline = article.find(".//MedlineCitation") if medline is not None: pmid = medline.find(".//PMID") paper["pmid"] = pmid.text if pmid is not None else "N/A" # Extract title title = article.find(".//ArticleTitle") paper["title"] = title.text if title is not None else "N/A" # Extract abstract abstract_elem = article.find(".//Abstract/AbstractText") paper["abstract"] = abstract_elem.text if abstract_elem is not None else "N/A" # Extract authors authors = [] for author in article.findall(".//Author"): fname = author.find(".//ForeName") lname = author.find(".//LastName") if fname is not None and lname is not None: authors.append(f"{fname.text} {lname.text}") paper["authors"] = ", ".join(authors[:3]) + ("..." if len(authors) > 3 else "") # Extract journal and date journal = article.find(".//Journal/Title") paper["journal"] = journal.text if journal is not None else "N/A" pub_date = article.find(".//PubDate/Year") paper["year"] = pub_date.text if pub_date is not None else "N/A" papers.append(paper) except Exception as e: return [{"error": f"XML parsing failed: {str(e)}"}] return papers def analyze_papers(self, papers: List[Dict]) -> Dict: """Analyze the retrieved papers for insights""" if not papers or papers[0].get("error"): return {"error": "No papers to analyze"} analysis = { "total_papers": len(papers), "year_distribution": {}, "animal_models": {}, "research_categories": {}, "key_findings": [], "drug_mentions": [], "methodology_trends": {} } # Analyze each paper for paper in papers: # Year distribution year = paper.get("year", "Unknown") analysis["year_distribution"][year] = analysis["year_distribution"].get(year, 0) + 1 # Analyze abstract for animal models and categories abstract = paper.get("abstract", "").lower() title = paper.get("title", "").lower() full_text = f"{title} {abstract}" # Animal model detection for animal in self.animal_keywords: if animal in full_text: analysis["animal_models"][animal] = analysis["animal_models"].get(animal, 0) + 1 # Extract drug mentions (simple regex for common drug patterns) drugs = re.findall(r'\b[A-Z][a-z]*(?:mab|nib|ine|ole|cin|tin)\b', paper.get("abstract", "")) analysis["drug_mentions"].extend(drugs) # Classify research category if classifier is available if self.classifier and abstract != "n/a": try: result = self.classifier(abstract[:512], self.research_categories) top_category = result["labels"][0] analysis["research_categories"][top_category] = analysis["research_categories"].get(top_category, 0) + 1 except Exception: pass # Process drug mentions drug_counter = Counter(analysis["drug_mentions"]) analysis["drug_mentions"] = dict(drug_counter.most_common(10)) return analysis def generate_summary(self, papers: List[Dict], analysis: Dict) -> str: """Generate a comprehensive summary of findings""" if not papers or papers[0].get("error"): return "No papers found or error in retrieval." summary = f""" # Literature Mining Summary ## Overview - **Total Papers Found**: {analysis['total_papers']} - **Search Date**: {datetime.now().strftime('%Y-%m-%d')} ## Key Insights ### Animal Models Used """ # Top animal models if analysis["animal_models"]: top_models = sorted(analysis["animal_models"].items(), key=lambda x: x[1], reverse=True)[:5] for model, count in top_models: summary += f"- **{model.title()}**: {count} papers\n" summary += "\n### Research Focus Areas\n" # Research categories if analysis["research_categories"]: top_categories = sorted(analysis["research_categories"].items(), key=lambda x: x[1], reverse=True)[:5] for category, count in top_categories: summary += f"- **{category.title()}**: {count} papers\n" summary += "\n### Frequently Mentioned Drugs\n" # Drug mentions if analysis["drug_mentions"]: for drug, count in list(analysis["drug_mentions"].items())[:5]: summary += f"- **{drug}**: {count} mentions\n" summary += "\n### Recent Highlights\n" # Recent papers (last 2 years) current_year = datetime.now().year recent_papers = [p for p in papers if p.get("year", "").isdigit() and int(p["year"]) >= current_year - 2] for paper in recent_papers[:3]: summary += f"- **{paper.get('title', 'N/A')}** ({paper.get('year', 'N/A')})\n" summary += f" *{paper.get('journal', 'N/A')}*\n\n" return summary def create_visualizations(self, analysis: Dict): """Create visualization plots""" plots = {} # Year distribution if analysis["year_distribution"]: years = list(analysis["year_distribution"].keys()) counts = list(analysis["year_distribution"].values()) fig_year = px.bar( x=years, y=counts, title="Publication Year Distribution", labels={"x": "Year", "y": "Number of Papers"} ) plots["year_dist"] = fig_year # Animal models if analysis["animal_models"]: models = list(analysis["animal_models"].keys())[:10] model_counts = [analysis["animal_models"][m] for m in models] fig_models = px.bar( x=model_counts, y=models, orientation='h', title="Most Common Animal Models", labels={"x": "Number of Papers", "y": "Animal Model"} ) plots["animal_models"] = fig_models # Research categories if analysis["research_categories"]: categories = list(analysis["research_categories"].keys()) cat_counts = list(analysis["research_categories"].values()) fig_categories = px.pie( values=cat_counts, names=categories, title="Research Focus Distribution" ) plots["categories"] = fig_categories return plots def create_gradio_interface(): """Create the Gradio interface""" miner = CancerResearchLiteratureMiner() def search_and_analyze(query, max_results): """Main function to search and analyze literature""" if not query.strip(): return "Please enter a search query.", None, None, None, None # Search papers papers = miner.search_pubmed(query, max_results) if not papers or papers[0].get("error"): error_msg = papers[0].get("error", "No papers found") if papers else "No papers found" return f"Error: {error_msg}", None, None, None, None # Analyze papers analysis = miner.analyze_papers(papers) # Generate summary summary = miner.generate_summary(papers, analysis) # Create visualizations plots = miner.create_visualizations(analysis) # Create papers dataframe papers_df = pd.DataFrame([ { "PMID": p.get("pmid", "N/A"), "Title": p.get("title", "N/A")[:100] + "..." if len(p.get("title", "")) > 100 else p.get("title", "N/A"), "Authors": p.get("authors", "N/A"), "Journal": p.get("journal", "N/A"), "Year": p.get("year", "N/A") } for p in papers ]) return ( summary, papers_df, plots.get("year_dist"), plots.get("animal_models"), plots.get("categories") ) # Create interface with gr.Blocks(title="Cancer Research Literature Mining Agent", theme=gr.themes.Soft()) as interface: gr.Markdown(""" # 🔬 Cancer Research Literature Mining Agent This AI agent searches and analyzes scientific literature related to cancer research in animal models. It automatically extracts insights about animal models used, research focus areas, and emerging trends. **Features:** - PubMed literature search with animal model focus - Automatic categorization of research areas - Drug mention extraction - Publication trend analysis - Interactive visualizations """) with gr.Row(): with gr.Column(scale=2): query_input = gr.Textbox( label="Research Query", placeholder="e.g., 'breast cancer immunotherapy', 'lung cancer biomarkers', 'pancreatic cancer treatment'", lines=2 ) max_results = gr.Slider( minimum=10, maximum=100, value=50, step=10, label="Maximum Results" ) search_btn = gr.Button("🔍 Search & Analyze Literature", variant="primary") with gr.Column(scale=1): gr.Markdown(""" ### Tips for Better Results: - Use specific cancer types (e.g., "breast cancer", "melanoma") - Include treatment modalities (e.g., "immunotherapy", "chemotherapy") - Add animal model terms (e.g., "mouse model", "xenograft") """) with gr.Tabs(): with gr.TabItem("📊 Summary & Insights"): summary_output = gr.Markdown(label="Analysis Summary") with gr.TabItem("📋 Papers Found"): papers_output = gr.Dataframe( headers=["PMID", "Title", "Authors", "Journal", "Year"], label="Retrieved Papers" ) with gr.TabItem("📈 Visualizations"): with gr.Row(): year_plot = gr.Plot(label="Publication Timeline") models_plot = gr.Plot(label="Animal Models") with gr.Row(): categories_plot = gr.Plot(label="Research Categories") # Connect the search function search_btn.click( search_and_analyze, inputs=[query_input, max_results], outputs=[summary_output, papers_output, year_plot, models_plot, categories_plot] ) # Add examples gr.Examples( examples=[ ["breast cancer immunotherapy mouse model", 50], ["lung cancer biomarkers xenograft", 30], ["pancreatic cancer treatment PDX", 40], ["melanoma drug resistance animal model", 35] ], inputs=[query_input, max_results] ) gr.Markdown(""" ### About This Agent This literature mining agent is specifically designed for cancer research in animal models. It searches PubMed for relevant papers and provides automated analysis of research trends, commonly used animal models, and emerging therapeutic approaches. **Data Sources:** PubMed/NCBI databases **Last Updated:** June 2025 **Supported Research Areas:** All cancer types and animal models """) return interface # Create and launch the interface if __name__ == "__main__": interface = create_gradio_interface() interface.launch( server_name="0.0.0.0", server_port=7860, share=True )