AI Co-Scientist for Knowledge Synthesis in Medical Contexts: A Proof of Concept
Arya Rahgozar, Pouria Mortezaagha
arXiv (Cornell University)
Problems Identified (5)
Biomedical research waste: Biomedical research waste is attributed to redundant studies, incomplete reporting, and limited scalability of evidence synthesis workflows.
Limited scalability of evidence synthesis: Traditional evidence synthesis workflows have limited scalability in biomedical science.
Evidence redundancy and gaps: Biomedical corpora contain thematic redundancy and underexplored research areas that need to be identified.
Biomedical research waste: Biomedical research waste is attributed to redundant studies, incomplete reporting, and limited scalability of evidence synthesis workflows.
Limited scalability of evidence synthesis: Traditional evidence synthesis workflows have limited scalability in biomedical science.
Proposed Solutions (5)
PICOS-aware AI co-scientist: The paper proposes an AI co-scientist for scalable and transparent knowledge synthesis based on explicit PICOS formalization.
Hybrid vector-graph knowledge platform: The platform combines relational storage, vector-based semantic retrieval, and a Neo4j knowledge graph.
PubMedBERT multi-task classifier: A transformer-based multi-task classifier fine-tuned from PubMedBERT is used for study design classification from titles and abstracts.
Bi-LSTM PICOS classifier: A Bidirectional Long Short-Term Memory baseline is used for automated PICOS compliance detection from titles and abstracts.
Hybrid retrieval-augmented generation: Full-text synthesis uses retrieval-augmented generation with hybrid vector and graph retrieval.
Results (3)
High study design classification accuracy:
PICOS compliance detection accuracy:
RAG improves structured synthesis queries:
Research Domain
Biomedical evidence synthesis / biomedical NLP