PubMed and Beyond
AI for Evidence Utilization · Retrieve

PubMed and beyond: biomedical literature search in the age of artificial intelligence

Qiao Jin, Robert Leaman, Zhiyong Lu
eBioMedicine 2024
PDF Journal
These slides were generated with the help of AI and may contain errors.
01Motivation

Biomedical search in the age of AI

  • Biomedical knowledge doubles every few years.
  • AI has transformed search — from ranking to embeddings to LLM assistants.
  • The tool landscape has outgrown what most clinicians and researchers can track.
  • Clinicians raise a question for roughly every other patient and leave about half unanswered (Del Fiol et al., 2014).
02Overview

The review surveys more than 30 tools across five information needs

The landscape of biomedical search, organized by input type and information need — from eviden
The landscape of biomedical search, organized by input type and information need — from evidence-based medicine to literature mining. Fig. 1.
03Scenario 1

Evidence-based medicine

  • PICO-structured clinical questions.
  • Queries are expanded and ranked to surface the best available evidence for a bedside decision.
A search engine for evidence-based medicine.
A search engine for evidence-based medicine. Fig. 2.
04Scenario 2

Precision medicine & genomics

  • Gene- and variant-centric retrieval.
  • A genomic query is mapped to its many synonyms first, so variant nomenclature doesn’t fragment results.
Variant-aware literature retrieval.
Variant-aware literature retrieval. Fig. 3.
05Scenario 3

Semantic search

  • Queries and sentences are embedded by meaning.
  • Equivalent phrasings (“heart attack” / “myocardial infarction”) retrieve each other.
Embedding-based semantic search.
Embedding-based semantic search. Fig. 4.
06Scenario 4

Literature recommendation

  • Topic- and article-based recommenders.
  • Learn from a set of seed papers to surface related and follow-up work.
Topic- and article-based recommendation.
Topic- and article-based recommendation. Fig. 5.
07Scenario 5

Literature mining

  • Extract entities and their relations across many papers.
  • Assemble a searchable knowledge graph.
Mining entity associations into a knowledge graph.
Mining entity associations into a knowledge graph. Fig. 6.
08Summary

PubMed and Beyond at a glance

BackgroundClinicians now have far more than PubMed for finding biomedical evidence.
ProblemThe AI-era search-tool landscape has outgrown what any clinician can track.
ApproachA practical field guide surveying 30+ tools.
ScopeFive information needs — EBM, precision medicine, semantic search, recommendation & mining.
ConclusionA practitioner’s map to the right AI-era search tool for each need.