STEP 9 / 9

臨床判讀 (ACMG/AMP)

28 個證據碼、5 級分類、1 套規則——把 annotated VCF 變成可發報告的 P / LP / VUS / LB / B 判定。

28 evidence codes, 5 classification tiers, 1 rule set — turn an annotated VCF into reportable P / LP / VUS / LB / B classifications.

一、為什麼需要 ACMG 標準?

2015 年之前,每個臨床實驗室對「這個變異致病嗎」的判定都靠專家經驗,結果同一變異可能被 A 實驗室判 Pathogenic、B 實驗室判 VUS、C 實驗室判 Benign。這不只是學術差異——病人會因此接受不同治療。

2015 年 ACMG (American College of Medical Genetics) 與 AMP (Association for Molecular Pathology) 聯合發表 Standards and Guidelines for the Interpretation of Sequence Variants(Richards et al. 2015),定義了 28 個標準化證據碼與組合規則,把判讀從「主觀經驗」推向「結構化推理」。

之後 ClinGen 持續發布基因專屬與證據專屬的精修指引(如 PVS1 細化、PP3/BP4 計分校準、SVI working groups),是目前全球臨床診斷的事實標準。

Before 2015, every clinical lab judged variant pathogenicity by expert opinion — the same variant might be Pathogenic at lab A, VUS at lab B, Benign at lab C. Not just an academic discrepancy — patients received different treatments based on this.

In 2015 the ACMG (American College of Medical Genetics) and AMP (Association for Molecular Pathology) jointly published Standards and Guidelines for the Interpretation of Sequence Variants (Richards et al. 2015), defining 28 standardized evidence codes and combining rules — pushing interpretation from "subjective expertise" toward "structured reasoning."

ClinGen has since continuously released gene-specific and evidence-specific refinements (e.g., PVS1 specification, PP3/BP4 calibration, SVI working groups) and is now the de facto global standard.

二、五級分類

P
致病
Pathogenic
LP
可能致病
Likely Pathogenic
VUS
意義未明
Uncertain
LB
可能良性
Likely Benign
B
良性
Benign

LPP 在臨床上常一視同仁(> 90% 機率致病),可作為診斷依據。VUS 不可作為臨床決策依據——這是 ACMG 報告中最常見也最棘手的類別,常需家系驗證或 functional study。

LP and P are typically treated equivalently in clinical settings (>90% probability of pathogenicity) and may serve as diagnostic basis. VUS cannot be used for clinical decision-making — the most common and trickiest category in ACMG reports, often requiring family validation or functional studies.

三、28 個證據碼總覽

致病性證據(PVS / PS / PM / PP)

代碼權重證據類型說明
PVS1極強LoF + 機制吻合nonsense / frameshift / splice ±1,2 / start-loss / 全外顯子 deletion,且該基因 LoF 為已知致病機制
PS1蛋白變異等同已知致病不同 nucleotide change 但產生相同 amino acid 變化
PS2de novo + 親子關係確認父母均為陰性 (paternity/maternity 已確認)
PS3功能性研究well-established functional assays 顯示功能受損
PS4病例對照患者中顯著富集 (OR > 5, CI 不跨 1.0)
PM1變異位於熱點位於已知 mutational hotspot 或 functional domain
PM2中→支持人群中極罕gnomAD 中缺席或極低 AF;ClinGen 後續降為 Supporting
PM3隱性疾病的 trans 證據與已知 pathogenic 變異 in trans 出現於患者
PM4蛋白質長度變化非重複區的 in-frame indel 或 stop-loss 改變蛋白質長度
PM5新 missense 但同位置已知致病同 codon 之前已有不同的 pathogenic missense
PM6疑似 de novo父母 negative 但 paternity 未驗證
PP1-5支持家系共分離 / missense gene / in silico / 表型一致 / 信譽資料庫5 個輔助證據碼,影響力較弱

良性證據(BA / BS / BP)

Code權重說明
BA1獨立 (Stand-alone)gnomAD popmax AF > 5% (ClinGen 已細化到不同基因有不同閾值)
BS1AF > 預期疾病盛行率
BS2健康成年 homozygous / hemizygous 觀察到
BS3功能研究顯示無影響
BS4家系不共分離 (在患病者中缺席)
BP1-7支持LoF 基因的 missense / in silico benign / synonymous + 無 splice 預測 ...等 7 項

四、證據如何組合成最終分類

判定所需證據
Pathogenic1 PVS1 + (1 PS) 1 PVS1 + (≥2 PM) 1 PVS1 + 1 PM + 1 PP 1 PVS1 + ≥2 PP ≥2 PS 1 PS + (≥3 PM) ...
Likely Pathogenic1 PVS1 + 1 PM 1 PS + (1-2 PM) 1 PS + ≥2 PP ≥3 PM 2 PM + ≥2 PP 1 PM + ≥4 PP
Likely Benign1 BS + 1 BP ≥2 BP
Benign1 BA1 (獨立) ≥2 BS
VUS不符合上述任一組合,或證據互相衝突

五、互動模擬:ACMG 分類器

勾選你蒐集到的證據碼,自動算出最終 ACMG 分類。注意這是教學示意,實務臨床判讀需要對該變異的全面文獻、家系、功能研究綜合評估。

Tick the evidence codes you've gathered to auto-compute the final ACMG classification. This is an illustrative tool; real clinical interpretation requires comprehensive literature, family, and functional study integration.

VUS

六、自動化判讀工具與資料庫

InterVar

第一個 ACMG 自動化工具(Wang lab, AJHG 2017)。整合 ANNOVAR + 28 規則。輸出建議分類,但無法完全取代手動判讀。

The first ACMG auto-classification tool (Wang lab, AJHG 2017). Wraps ANNOVAR + the 28 rules. Outputs suggested classification but cannot fully replace manual review.

Franklin (Genoox)

商業 web 平台,自動套 ACMG + ClinGen 細化規則,支援 manual override 與報告生成。臨床診斷實驗室主流選擇之一。

Commercial web platform applying ACMG + ClinGen-refined rules, with manual override and report generation. A mainstream choice for clinical diagnostic labs.

Varsome / VarSeq

類似的整合平台,內建 InterVar engine + 多 prediction tool 整合。Varsome 有免費社群版。

Similar integrated platforms with InterVar engine + multi-predictor integration. Varsome has a free community edition.

ClinGen Variant Curation Interface (VCI)

ClinGen 官方手動 curation 平台,給 expert panel 使用。判讀結果直接進入 ClinVar,是最權威來源。

ClinGen's official manual curation platform for expert panels. Decisions feed directly into ClinVar — the most authoritative source.

ClinVar

NCBI 維護的 variant-disease 公開資料庫。每個 entry 標註 review status (1-4 stars),4 stars = expert panel reviewed。臨床判讀時必查。

NCBI's public variant-disease database. Each entry has a review status (1–4 stars); 4 stars = expert panel reviewed. Always check during clinical interpretation.

gnomAD

~80,000 WGS + 730,000 WES 的 allele frequency 參考。BA1 / BS1 / PM2 都需要查 gnomAD popmax AF。v4 已是 GRCh38 standard。

Allele frequency reference from ~80,000 WGS + 730,000 WES. BA1, BS1, PM2 all rely on gnomAD popmax AF lookups. v4 is the GRCh38 standard.

七、FAQ

VUS 太多怎麼辦?
這是 ACMG 最大的痛點。應對策略:(1) 跑 trio 看 de novo / co-segregation; (2) functional study (mini-gene splicing assay, expression assay); (3) 等 ClinVar 累積更多 submitter; (4) 加入 Matchmaker Exchange 找其他病例; (5) 遵循 ClinGen 基因專屬指引重新計分。新工具如 AlphaMissense 也持續減少 missense VUS 比例。
VUS overload is ACMG's biggest pain point. Strategies: (1) trio analysis for de novo / co-segregation; (2) functional studies (mini-gene splicing assay, expression assay); (3) wait for more ClinVar submissions; (4) join Matchmaker Exchange to find other cases; (5) re-score using ClinGen gene-specific guidelines. New tools like AlphaMissense are reducing the missense VUS fraction over time.
ACMG 與癌症 somatic 變異的判讀一樣嗎?
不一樣。ACMG 2015 主要針對 germline 遺傳疾病。Somatic(cancer)有獨立的 AMP/ASCO/CAP 2017 指引,分四級 (Tier I–IV),著重「臨床可行性 (actionability)」而非「致病性」。Tier I = FDA-approved drug target;Tier IV = unknown clinical significance。
No. ACMG 2015 targets germline genetic disease. Somatic (cancer) variants have separate AMP/ASCO/CAP 2017 guidelines with four tiers (I–IV), emphasizing "clinical actionability" over "pathogenicity." Tier I = FDA-approved drug target; Tier IV = unknown clinical significance.
不同 reviewer 判讀結果不一致怎麼處理?
建議 dual review + consensus meeting:兩位獨立 curator 各自評分,差異點由團隊討論裁定。臨床診斷實驗室通常還有第三位 sign-off authority。ClinGen 的 VCEP (Variant Curation Expert Panel) 對重要基因會有共識指引。
Use dual review + consensus meeting: two independent curators score independently; differences resolved by team discussion. Clinical labs typically also have a third sign-off authority. ClinGen's VCEPs (Variant Curation Expert Panels) issue consensus guidelines for important genes.
incidental finding (IF) / secondary finding (SF) 是什麼?
在做 WES/WGS 找原本疾病基因的同時,意外發現其他疾病基因的致病變異。ACMG SF v3.2 (2023) 列出 81 個基因(如 BRCA1/2, MLH1, RYR1, KCNQ1)必須回報,因為「可採取臨床行動」。病人有 opt-out 權利。
Pathogenic variants in genes unrelated to the primary indication, found incidentally during WES/WGS. ACMG SF v3.2 (2023) lists 81 genes (e.g., BRCA1/2, MLH1, RYR1, KCNQ1) that must be reported because they are "clinically actionable." Patients have the right to opt out.

八、小測驗

Q1. PVS1 適用的條件是?

Q2. gnomAD popmax AF=12% 的 missense 變異最可能被分類為?

Q3. 為什麼 VUS 不能作為臨床決策依據?