Medical Literature Analyzer

Problem

Reading a medical paper often scatters conclusions, statistics, and reproducible methods across sections. Supplementary evidence can blur into the core paper, and experimental or translational paths are hard to derive from a single source. This skill treats one medical document as input, first performs a text-grounded analysis, then searches related references around key drugs, diseases, or mechanisms, and finally produces reviewable reports and graph data.

How It Works

The workflow reads PDF, DOCX, URL, or pasted text, then extracts title, authors, DOI, PMID, study design, evidence level, key findings, P values, 95% CI, HR/OR/RR, and sample size. It performs relationship-driven entity extraction, keeping only entity pairs supported by the text, across diseases, drugs, genes, symptoms, pathways, outcomes, and related types, then writes them to entities.json. It must run build_knowledge_graph.py and visualize_graph.py to generate an interactive HTML graph. Core capabilities include active supplementary literature search, short-, medium-, and long-term translational path inference, experimental method summarization, and 1-3 actionable protocols, saved as analysis_report.md, graph JSON, and supplementary reference data.

Boundaries

It fits research articles, mechanism studies, clinical trials, and method-heavy papers. If retrieval fails, entities are empty, or the topic is atypical, results need human review. Generated translational paths and protocols are uncertain; key conclusions should follow the original paper, and experiments require ethics review.