為什麼方法寫作決定了論文壽命
Introduction 與 Discussion 會因為時代改變而過時，但方法是論文唯一不會被時間打折的部分——只要寫得夠詳細，10 年後仍能被引用、被重做、被擴充。
2026 年的標準是：(1) 對應的 EQUATOR reporting checklist 必須 cited 在文末；(2) 軟體/套件版本必須明確；(3) Code 與 Data 必須有 availability statement，無論是公開還是 reasonable request。
Introductions and Discussions age. Methods are the only timeless part of a paper — written well, they can be cited, replicated, and extended a decade later.
The 2026 standard: (1) the relevant EQUATOR reporting checklist must be cited; (2) software / package versions must be explicit; (3) code & data must have an availability statement — whether deposited publicly or "available upon reasonable request."
一、依研究類型挑選 checklist
EQUATOR Network 維護著 500+ 種報告指引，目的是讓不同研究類型有對應的「最低資訊清單」。投稿前先抓 checklist 並逐項核對，比起寫完再回頭補要省時很多。
The EQUATOR Network maintains 500+ reporting guidelines — a minimum-information checklist per study type. Download the checklist before writing, not after.
| Checklist | 適用研究 | 重點項目 |
|---|---|---|
| CONSORT 2025 | 隨機對照試驗 (RCT)Randomized controlled trials | 隨機分派、盲化、CONSORT flow diagramRandomization, blinding, flow diagram |
| STROBE | 觀察性研究 (cohort / case-control / cross-sectional)Observational studies (cohort / case-control / cross-sectional) | study setting、sampling、bias control、confoundersSetting, sampling, bias control, confounders |
| PRISMA 2020 | 系統性回顧、meta-analysisSystematic reviews, meta-analyses | PROSPERO 註冊、搜尋策略、PRISMA flow diagramPROSPERO registration, search strategy, flow diagram |
| ARRIVE 2.0 | 動物實驗Animal studies | 3R、IACUC、樣本量計算、隨機化、blinding3Rs, IACUC, sample size, randomization, blinding |
| SPIRIT | 臨床試驗 protocolClinical trial protocols | trial design、endpoint hierarchy、stopping rulesTrial design, endpoint hierarchy, stopping rules |
| STARD | 診斷準確度研究Diagnostic accuracy | 參考標準、樣本選擇、cut-off 設定Reference standard, sample selection, cut-off determination |
| TRIPOD-AI | 機器學習預測模型ML prediction models | 資料來源、特徵工程、外部驗證、calibrationData sources, features, external validation, calibration |
| CARE | 病例報告Case reports | 時間軸、患者觀點、informed consentTimeline, patient perspective, informed consent |
二、Methods 的 6 個子段
Study design & setting
研究設計（cohort / RCT / cross-sectional）、時間（recruitment dates）、地點（醫院、實驗室）、是否多中心。第一句就要清楚交代，避免讀者腦補。
Design (cohort / RCT / cross-sectional), period (recruitment dates), site (hospital, lab), single- or multi-center. Make the first sentence unambiguous.
Participants / samples
Inclusion / exclusion criteria 必須完整列出。臨床研究：年齡、性別、診斷準則。生資/Omics：物種、組織、批次資訊、樣本來源（GEO accession、collaborator）。
Inclusion / exclusion criteria — fully spelled out. Clinical: age, sex, diagnostic criteria. Bioinformatics/omics: species, tissue, batch info, accession (GEO/SRA) or collaborator source.
Variables / measurements
每個變項：定義、單位、量測工具、測量者、reliability。Primary outcome 與 secondary outcomes 分開列。
Per variable: definition, unit, instrument, measurer, reliability. List primary outcome and secondary outcomes separately.
Sample size & power
α、β、effect size、預期 dropout、最後算出 n。引用所用軟體（G*Power、PASS、R pwr）。RCT 與 ML 模型尤其必要。
α, β, effect size, expected dropout, final n. Cite software (G*Power, PASS, R pwr). Mandatory for RCTs and ML models.
Statistical / computational analysis
所用統計檢定 + 假設檢驗條件、效應量、CI、缺失值處理、敏感度分析。生資：完整 pipeline、所有套件版本、隨機種子。
Statistical tests + assumptions, effect sizes, CIs, missing-data handling, sensitivity analyses. Bioinformatics: full pipeline, all package versions, random seeds.
Ethics, code & data availability
IRB/IACUC 編號、informed consent、利益衝突；data deposit (GEO / Zenodo / Figshare) + DOI；code repository (GitHub) + commit hash + license。
IRB/IACUC number, informed consent, COI; data deposit (GEO / Zenodo / Figshare) + DOI; code repo (GitHub) + commit hash + license.
三、生資/Omics 方法寫作對照
❌ 無法重現
「我們使用 Seurat 分析 scRNA-seq 數據，過濾低品質細胞並進行 clustering 與 differential expression。」
"We analyzed scRNA-seq data with Seurat, filtered low-quality cells, then clustered and ran differential expression."
✅ 可重現版
「Cellranger v8.0 (refdata-gex-GRCh38-2024) 處理原始 fastq 後，以 Seurat v5.1.0 (R 4.4.1) 過濾 nFeature_RNA 200–6000 且 percent.mt < 15% 的細胞 (n=18 432)。SCTransform v2 標準化後取前 30 個 PCs，以 Louvain (resolution=0.6) 聚類，UMAP n.neighbors=30, min.dist=0.3。差異表達採 Wilcoxon (FindMarkers, only.pos=TRUE, logfc.threshold=0.25)，BH 校正，FDR<0.05。完整程式：github.com/charlene/tnbc-scrna (commit a3f9c12)，原始矩陣 GEO: GSE298765。」
"Raw fastq processed with Cellranger v8.0 (refdata-gex-GRCh38-2024). Seurat v5.1.0 (R 4.4.1) filtered cells with nFeature_RNA 200–6000 and percent.mt < 15% (n=18,432). SCTransform v2 normalization, top 30 PCs, Louvain clustering (resolution=0.6), UMAP (n.neighbors=30, min.dist=0.3). Differential expression: Wilcoxon (FindMarkers, only.pos=TRUE, logfc.threshold=0.25), BH-adjusted FDR<0.05. Full code: github.com/charlene/tnbc-scrna (commit a3f9c12); raw matrices GEO: GSE298765."
四、可重現範本
# Statistical analysis paragraph template Continuous variables are presented as [mean (SD)] if normally distributed or [median (IQR)] otherwise; categorical variables as n (%). Between-group comparisons used [Student's t-test / Mann-Whitney U] for continuous and [chi-squared / Fisher's exact] for categorical variables. Multivariable adjustment used [logistic / Cox] regression adjusting for [a priori confounders: age, sex, ...]. Two-sided p < 0.05 was considered significant. Multiple-testing was controlled with [Benjamini-Hochberg FDR / Bonferroni]. All analyses used [R 4.4.1 / Python 3.12] with package versions in Supplementary Table S1.
# Data Availability Statement Raw sequencing data have been deposited in [GEO under accession GSEXXXXXX] and are publicly available as of the date of publication. Processed Seurat objects, intermediate matrices, and figure source data are available at Zenodo (DOI: 10.5281/zenodo.XXXXXXX). De-identified clinical data supporting the findings of this study are available from the corresponding author upon reasonable request, subject to [IRB approval].
# Code Availability Statement All analysis code is available at github.com/yourgroup/projectname (commit hash: abc1234) under MIT license. A frozen Zenodo snapshot of the repository at time of publication is archived at DOI: 10.5281/zenodo.XXXXXXX. The pipeline is containerised (Docker image: ghcr.io/yourgroup/proj:v1.0.0) and managed by Snakemake v8.20. A workflow diagram is in Supplementary Figure S1.
# Ethics & Consent Statement This study was approved by the [Institutional Review Board of XXX Hospital] (approval no. XXX-2025-001) and conducted in accordance with the Declaration of Helsinki (2024 revision). Written informed consent was obtained from all participants prior to enrollment. Animal experiments were approved by the IACUC of [Institution] (protocol no. XXX-A-2025-007) and reported per the ARRIVE 2.0 guidelines.
① 寫「Seurat 進行分析」但沒寫版本——5 年內 Seurat 已從 v3 改到 v5，演算法差很大。
② 「significant at p<0.05」但沒交代 multiple testing 校正。
③ "data available upon request" 但作者離職、實驗室解散後就拿不到——deposit 才是負責任做法。
① "We used Seurat" without version — Seurat v3→v5 algorithms differ substantially.
② "Significant at p<0.05" without mentioning multiple-testing correction.
③ "Available upon request" fails when authors move or labs close — depositing is the responsible approach.
五、寫 Methods 之前的 5 個問題
🌳 開始寫之前先問自己
sessionInfo()、Python 跑 pip freeze，存成 supplementary。sessionInfo() in R or pip freeze in Python; save as supplementary.📝 自我檢測
1. 你做的是觀察性 cohort 研究（追蹤 5 年發病率），最該對照哪個 EQUATOR checklist？
1. You ran an observational cohort study (5-year incidence). Which EQUATOR checklist applies?
2. 關於 code/data availability，最負責任的作法是？
2. The most responsible approach to code/data availability is?
3. 寫 scRNA-seq 方法時，下列何項可以省略？
3. Which item is OK to omit from a scRNA-seq Methods section?