外部复制

步骤 4 · 只分流 · 图 1 张 · 发表版表格 4 个　（流程见 v5 定稿）

4a 换暴露平台 · 4b 换结局队列 · 4c 两侧都换，各自走三分法。★ 不否决，只把每一对分到「通过 / 送 4X / 不可评估」。

---

图

Figure F4. External replication of the 53 complication associations that survived the directionality check, under three different substitutions

完整图注（英文，与投稿版一致）

Figure F4. External replication of the 53 complication associations that survived the directionality check, under three different substitutions.

The three checks answer different questions and are deliberately not combined into a single pass/fail. 4a replaces the exposure measurement: the same protein is instrumented from an independent proteomic platform (deCODE SomaScan instead of UKB-PPP Olink), holding the outcome fixed. 4b replaces the outcome: the same instrument is tested against an independent outcome cohort, holding the exposure fixed. 4c replaces both at once.

(a) Every bar uses the SAME denominator, the 53 associations that reached this step, so the three checks can be read against one another. Each association falls into one of three states: the alternative data agree in direction, disagree in direction, or CANNOT BE JUDGED.

The not-assessable share is large by design of the available data, not because the findings failed: 27 of 53 for 4a, 19 for 4b and 36 for 4c. Reporting these as replication failures would be wrong, and computing a replication rate from the remainder would be misleading; that is why no percentage is shown.

(b) Why an association could not be judged, which in every case is a property of the available data rather than of the association. For 4a the protein is either absent from the alternative platform or has no cis instrument there. For 4b no independent cohort exists for that outcome at all: every not-assessable association in 4b is a diabetic maculopathy association, for which no external GWAS was available; the other three complications were tested against MVP diabetic retinopathy (GCST90475689), Salem et al. 2019 diabetic nephropathy, and MVP neuropathy (GCST90475676). For 4c an association can only be examined if it satisfies both of the preceding requirements simultaneously.

Direction concordant combines associations that reached significance in the alternative data with those that agreed in direction without reaching significance; direction opposite likewise combines significant and non-significant sign reversals. The finer breakdown is given in the supplementary tables.

Unit. Counts are protein-outcome associations. The underlying tables for 4a and 4c contain one row per SomaScan aptamer, so they have more rows than associations: MICB_MICA is represented by two aptamers. Where the aptamers for one protein disagree, the association is classified as direction-opposite and routed to the epitope investigation; this happened for 2 associations. That disagreement is itself informative. For MICB_MICA the two aptamers give exposure effects of opposite sign for the SAME variant (rs3132467): +0.21 for aptamer 5102_55 and -1.01 for aptamer 2730_58, against -1.10 measured by the Olink assay. Two aptamers targeting one protein cannot both be measuring the same quantity; this is the signature of an aptamer-binding artifact rather than of a failed replication, and MICB_MICA lies in the highly polymorphic MHC region where protein-altering variants affecting binding are common. The association is therefore sent to Figure F4X rather than judged here.

Method and threshold. The same Wald ratio estimator is applied in the alternative data and the two estimates are compared after harmonising to the same effect allele. Direction agreement is judged on the SIGN of the point estimate and deliberately does NOT require significance: demanding p < 0.05 in a smaller replication dataset would turn insufficient power into evidence against a correctly-signed result. Where significance is referred to, it is nominal p < 0.05 in the alternative dataset. Role in the analysis pipeline: this step TRIAGES, it does not veto. No association is removed on the basis of Figure F4. A direction disagreement here routes the association to the epitope and protein-altering-variant investigation (Figure F4X), which is the only step in the workflow that can veto on this evidence.

Data sources and sample sizes. Exposure: UKB-PPP plasma proteome (Olink Explore 3072), 34,557 European participants, 1,954 proteins with a cis-pQTL. Outcomes: FinnGen R9 -- Diabetic retinopathy 10,413 cases / 308,633 controls; Diabetic maculopathy 3,572 cases / 308,547 controls; Diabetic nephropathy 4,111 cases / 308,539 controls; Diabetic neuropathy 2,843 cases / 271,817 controls.

---

发表版表格

文件	下载
replication_4a_alt_platform.csv	下载
replication_4b_alt_outcome.csv	下载
replication_4c_both_sides.csv	下载
replication_outcome_cohorts.csv	下载

本页图为网页版 PNG（最宽 1600 px）。投稿用的矢量 PDF 与 600 dpi TIFF 体积较大，留在仓库 results/external_replication/figures/，不随文档站分发。