Biomedical Manuscript Polishing Assistant

Problem

Biomedical writing often gets stuck when the science is clear but the English does not read like a manuscript. Chinese notes, rough English, experimental results, and figure legends may lack comparison groups, endpoints, statistical phrasing, or section-level tone. A naive rewrite can also add missing sample sizes, mechanisms, or citations.

How It Works

• Task routing: identify background, research gap, objective, hypothesis, methods, two-group comparison, multi-group comparison, statistical expression, result reporting, association, mechanism, conclusion, significance, limitations, or future work.
• Knowledge-base lookup: prefer local data/ resources such as pattern_library.csv, collocation_index.csv, and lab_lexicon.csv; use scripts like search_terms.py for terminology, collocations, and sentence patterns.
• Controlled rewriting: do not invent experiments, statistics, controls, sample sizes, or citations; require permission and label external wording; prefer suggest, indicate, or support for indirect evidence, and avoid overclaiming terms such as prove.

Sentence-level output includes a polished version, alternatives, a pattern note, and cautions when needed. Paragraph-level output explains sentence functions. Figure legends start with what the figure shows, then name the model, treatment, control, assay, endpoint, and statistics.

Boundaries

It is best for polished manuscript paragraphs, figure legends, and abstract work when the experimental content and evidence strength are clear. If the experimental subject, comparison group, intervention, endpoint, or statistical result is missing and a safe rewrite is not possible, it should ask for the missing detail. The bundled knowledge base is an expanded subset; complete local search requires replacing it with full CSV exports, while update_data.py is only for updating the local knowledge base.