Discussion 是「對話」，不是「重述」
新手最常見的錯誤：把 Discussion 寫成 Results 的擴大版。Discussion 的真正任務是「把你的結果放回領域既有知識的脈絡裡」，回答四個問題：
- 主要發現是什麼？(1 段，不要超過)
- 與既有文獻一致嗎？不一致的話為什麼？
- 可能機制與意義？
- 限制與未來方向？
Beginner's most common mistake: writing Discussion as an expanded Results. The real job of Discussion is to put your results into the field's existing context, answering four questions:
- What is the main finding? (1 paragraph max)
- Is it consistent with prior literature? If not, why?
- Possible mechanism & significance?
- Limitations & future directions?
一、Discussion 的標準六段
主要發現摘要
1-3 句話濃縮核心結果，不要重貼數字。例：「本研究在 18 例 TNBC 中首次描繪 chemotherapy 後的轉錄組重塑，鑑定 3 個化療抗性相關細胞態。」
1-3 sentences distilling core findings — no number copy-paste. Eg: "We profiled the post-chemo transcriptional remodeling of 18 TNBC samples and identified 3 chemo-resistance-associated cell states."
與既有文獻對話
「我們的結果與 X 一致 (Smith 2023)」「我們延伸了 Y 的觀察」「我們的數據顛覆了 Z 的結論——可能是因為 cohort/method 差異」。具體引用，不要籠統說 "many studies"。
"Our results agree with X (Smith 2023)"; "We extend Y's observation"; "Our data contradicts Z — possibly due to cohort / method differences." Cite specifically; don't say "many studies."
可能機制與生物意義
提出可能的解釋機制，明確標示「speculative」。可引用既知的 pathway / 細胞類型，但不要當成已證實。
Propose plausible mechanisms — flag them as speculative. You may cite known pathways / cell types but don't present them as proven.
臨床/應用意義
對醫師、患者、政策、後續研究有什麼意義？避免 overclaim：「these findings may inform」「warrant further investigation」比「will revolutionize」好得多。
Implications for clinicians, patients, policy, follow-up research. Avoid overclaim: "these findings may inform" or "warrant further investigation" beats "will revolutionize."
限制 (Limitations)
誠實列出 3-5 個真正的限制（不是套話）。寫法：「Limitation + 為何重要 + 我們做了什麼緩解 + 未來該怎麼補」。
Honestly list 3-5 real limitations (not boilerplate). Format: "Limitation + why it matters + what we did to mitigate + future fix."
結論與未來方向
2-3 句結論 + 2-3 個具體 next steps。結論要與 Introduction 的 aim 呼應。
2-3 sentence conclusion + 2-3 concrete next steps. Conclusion must echo the Introduction's aim.
二、Limitations 三層寫法
新手把 limitations 寫成「樣本太少、未來會做更多」，審稿人秒看穿。專業作法分三層：
Beginners write "small sample, will do more next time" — reviewers see through it instantly. Professional version has three layers:
| 層級 | 範圍 | 範例 |
|---|---|---|
| ① 內部 | 研究設計本身的限制：sample size、selection bias、measurement errorStudy design limits: sample size, selection bias, measurement error | 「樣本來自單一醫學中心，可能限制推論至其他族群」"Single-center cohort may limit generalizability" |
| ② 外部 | 推廣/應用上的限制：generalizability, real-world translationGeneralizability and translation limits | 「我們僅納入治療前樣本，無法評估治療反應的縱向變化」"We only included pre-treatment samples, so longitudinal treatment response cannot be assessed" |
| ③ 方法學 | 技術/演算法/工具的限制Technique / algorithm / tool limits | 「scRNA-seq dropouts 可能低估低表達基因；CITE-seq 蛋白驗證有助降低此風險」"scRNA-seq dropouts may underestimate low-expression genes; CITE-seq protein validation could mitigate" |
① 「樣本不夠」（為什麼不夠？n 多少才夠？）
② 「未來會做 in vivo 驗證」（沒講為什麼這次不做）
③ 「結果還需要更多研究確認」（廢話，所有研究都需要）
① "Sample size is small" (why is it small? what n is needed?)
② "Future work will include in vivo validation" (no rationale for omitting now)
③ "Findings need further confirmation" (useless — every study needs that)
三、Discussion 段落對照
❌ Overclaim
「我們的結果證明 Gene X 是 TNBC 化療抗性的核心驅動因子，必將成為新藥標靶。」
(從相關性跳到因果，從一個 cohort 跳到「必將」——審稿人會用紅筆畫滿。)
"Our results prove Gene X is the central driver of chemo resistance in TNBC and will become a new drug target."
(Jumps from correlation to causation, from one cohort to "will" — reviewer will redline this.)
✅ 校準語氣
「我們的數據顯示 Gene X 表達與化療抗性顯著相關 (HR 1.8, 95% CI 1.2-2.7)，與 Smith 等 (2023) 的細胞株發現一致。然而，本研究為觀察性設計，無法確立因果關係；功能性驗證 (e.g. Gene X knockdown in PDX models) 仍是必要的下一步。」
"Our data show Gene X expression is significantly associated with chemoresistance (HR 1.8, 95% CI 1.2–2.7), consistent with cell-line findings by Smith et al. (2023). However, this observational design cannot establish causality; functional validation (e.g., Gene X knockdown in PDX models) remains a necessary next step."
❌ 假 limitations
「本研究有一些限制：樣本數較少、結果還需更多研究確認、未來會做更多分析。」
"This study has limitations: small sample size, results need confirmation, future work will do more analysis."
✅ 真 limitations
「本研究有三個主要限制。第一，n=18 對於 cell-state 分群已足夠，但對 subgroup 分析 (如 PD-L1+ vs PD-L1−) power 不足；我們透過 bootstrap (n=1000) 估計 stability。第二，所有樣本來自台大醫院 2024-2025 cohort，可能不代表其他種族與化療方案族群；驗證 cohort (TCGA-BRCA, METABRIC) 顯示一致趨勢。第三，scRNA-seq 對低表達 transcription factor 的偵測有 dropout 限制，未來可結合 ATAC-seq 或 CITE-seq 提升解析度。」
"This study has three main limitations. First, n=18 is adequate for cell-state clustering but underpowered for subgroup analysis (e.g., PD-L1+ vs PD-L1−); we estimated stability via bootstrap (n=1000). Second, samples are from a single 2024-2025 NTUH cohort and may not generalize across ethnicities or chemotherapy regimens; validation in TCGA-BRCA and METABRIC showed consistent trends. Third, scRNA-seq dropout limits detection of low-expression TFs; future work could integrate ATAC-seq or CITE-seq for finer resolution."
四、Hedging：不過度也不退縮
學術寫作講究「校準的不確定性」。Hedging 過度→沒立場、沒說服力；Hedging 不足→overclaim、被審稿人抓。
Academic writing requires "calibrated uncertainty." Over-hedge → no stance, no persuasion. Under-hedge → overclaim, reviewer catches it.
| 證據強度 | 動詞 | 副詞 / 形容詞 |
|---|---|---|
| 非常強 | demonstrate, establish, show | conclusively, robustly, clearly |
| 強 | indicate, support, reveal | strongly, consistently |
| 中等 | suggest, imply, propose | likely, plausibly, often |
| 弱 | may, might, could | possibly, potentially, in principle |
五、Discussion 檢查清單
🌳 送審前的 7 項檢查
📝 自我檢測
1. Discussion 結構與 Introduction 的關係是？
1. Discussion's structural relationship to Introduction is?
2. 觀察性 cohort 研究發現 Gene X 與化療抗性顯著相關，最恰當的 Discussion 動詞是？
2. Observational cohort finds Gene X significantly associated with chemoresistance. Most appropriate verb in Discussion?
3. 哪一句屬於「真 limitations」？
3. Which is a "real limitation"?