主题
定稿
步骤 7 · 图 4 张 · 发表版表格 1 个 （流程见 v5 定稿）
证据矩阵、措辞阶梯与候选状态表。4X 之后只能写「关联 / 候选」，只有共定位支持的子集才能升级措辞。
图
Figure F7b-forest. Wald ratio estimates for the 11 proteins that reached strong colocalisation with at least one diabetic complication, shown for all four complications in FinnGen R9
完整图注（英文，与投稿版一致）
Figure F7b-forest. Wald ratio estimates for the 11 proteins that reached strong colocalisation with at least one diabetic complication, shown for all four complications in FinnGen R9. Exposures are UKB-PPP plasma cis-pQTL sentinel variants; effects are per one standard deviation higher genetically predicted plasma protein. Bars are 95% confidence intervals on a logarithmic axis. Proteins are ordered by their highest posterior probability of a shared causal variant. Significance is Benjamini-Hochberg FDR within each outcome; strong colocalisation is coloc.abf PP.H4 >= 0.8 with priors p1 = p2 = 1e-4 and p12 = 1e-5 over a 1 Mb cis window. All 44 pairs carry an estimate, so no cell is missing or unassessable: of these, 13 are significant with strong colocalisation, 2 are significant without it (SIGLEC5 with retinopathy reaches only moderate colocalisation, and APOE with retinopathy was vetoed at the reverse Mendelian randomization step), and 29 are not significant. No candidate is significant for neuropathy.
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.
Figure F7b. Step-by-step evidence for the 11 proteins that reached strong colocalisation with at least one diabetic complication
完整图注（英文，与投稿版一致）
Figure F7b. Step-by-step evidence for the 11 proteins that reached strong colocalisation with at least one diabetic complication. Rows are all 15 Benjamini-Hochberg-significant protein-complication associations involving these proteins, so a protein appears once per complication; rows are ordered by colocalisation tier and then by PP.H4. Step 3 is reverse Mendelian randomization (a veto point). Steps 4a, 4b and 4c are replication with a different exposure platform (deCODE SomaScan), a different outcome cohort, and both sides changed; these triage findings and never veto them. Step 4X is the epitope and protein-altering-variant investigation, the only veto point in step 4, entered only on a direction alarm - "not tested" here means no alarm was raised. Step 5 is coloc.abf with priors p1 = p2 = 1e-4 and p12 = 1e-5 over a 1 Mb cis window (strong PP.H4 >= 0.8, moderate >= 0.5, weak < 0.5). Step 6 cross-layer counts, among retina, whole-blood, tissue and splicing quantitative trait loci layers with a computable posterior, how many reach PP.H4 >= 0.8. Colour is a three-level support grade whose meaning is column-specific. Supported: reverse Mendelian randomization passed; replication significant and in the same direction; strong colocalisation; at least one cross-layer locus colocalised. Partial support: replication direction-consistent but not significant; moderate colocalisation. Not supported: replication sign reversed but not significant - under the v4 rules this is not an alarm and does not enter step 4X; weak colocalisation; no cross-layer locus colocalised. Not assessable or not tested covers instruments absent from the comparison dataset, steps not run for that pair, and step 4X not entered because no direction alarm was raised - none of the 11 candidates raised one. The two right-hand columns are translational annotation rather than causal evidence and are therefore left uncoloured: approved drugs is the ChEMBL count of approved drugs with a mechanism record against the target (Open Targets returns 12 rather than 11 for IFNAR1; the ChEMBL count is used throughout this paper), and off-target traits is the number of Bonferroni-significant phenome-wide associations of the sentinel variant. Direction of the approved drugs is not shown here: for IFNAR1, 10 of the 11 are interferons acting on the IFNAR1-IFNAR2 complex in the direction implied by the Mendelian randomization estimate, while anifrolumab, the only IFNAR1-selective approved drug, is an antagonist and acts in the opposite direction.
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.
Figure F7a. Permitted wording at each stage of the v4 cis-pQTL drug-target Mendelian randomization pipeline
完整图注（英文，与投稿版一致）
Figure F7a. Permitted wording at each stage of the v4 cis-pQTL drug-target Mendelian randomization pipeline. Counts are pairs of protein and complication, and the distinct proteins they involve, across the four diabetic complications (retinopathy, maculopathy, nephropathy, neuropathy) in FinnGen R9. Exposures are UKB-PPP plasma cis-pQTL sentinel variants; 1587 proteins carried an instrument that harmonised against every complication. Step 2 applies Benjamini-Hochberg FDR < 0.05 within each outcome. The two veto points are step 3 (reverse Mendelian randomization) and step 4X (epitope or protein-altering-variant investigation); external replication in step 4 triages findings but never vetoes them. Step 5 is colocalisation by coloc.abf (Giambartolomei 2014) with priors p1 = p2 = 1e-4 and p12 = 1e-5 over a 1 Mb cis window; the upgrade threshold is posterior probability of a shared causal variant PP.H4 >= 0.8 outside the MHC region (chr6:25.5-34 Mb). The top rung is drawn as blocked because no association in this study reaches a drug-target claim: the primary results do not survive Bonferroni correction, so all wording remains exploratory.
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.
Figure F7c. Overview of the v4 cis-pQTL drug-target Mendelian randomization pipeline as applied to four diabetic complications (retinopathy, maculopathy, nephropathy, neuropathy) in FinnGen R9, with the count of surviving protein-complication pairs at each step
WARNING
⚠️ 这张总览图是按流程 v4 画的，其中 MR-Steiger 仍画在第 1 步。现行流程是 v5（Steiger 已并入第 3 步）。本图待随 v5 重画。
完整图注（英文，与投稿版一致）
Figure F7c. Overview of the v4 cis-pQTL drug-target Mendelian randomization pipeline as applied to four diabetic complications (retinopathy, maculopathy, nephropathy, neuropathy) in FinnGen R9, with the count of surviving protein-complication pairs at each step. Exposures are UKB-PPP plasma cis-pQTL sentinel variants within 500 kb of the encoding gene with F >= 10; 1,587 proteins carried an instrument that harmonised against every complication. Colour marks what each step is permitted to do to the candidate set. Only step 3 (directionality: variant-level MR-Steiger together with trait-level reverse Mendelian randomization) and step 4X (epitope and protein-altering-variant investigation) may remove a finding, and together they removed 5 of 57 associations. Step 4 replication is reported as a three-way split of consistent, opposite and not assessable, and never removes anything: an instrument absent from a comparison dataset cannot be judged and is not counted as a failure. Step 5 colocalisation (coloc.abf, priors p1 = p2 = 1e-4 and p12 = 1e-5, 1 Mb cis window) upgrades the wording permitted for the 13 pairs reaching PP.H4 >= 0.8 outside the MHC region but does not change membership. Step 6 annotates all 11 candidates and removes none. Pathway enrichment is a side branch that informs interpretation only. MHC region is chr6:25.5-34 Mb as defined in the UKB-PPP instrument table.
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 |
|---|---|
candidate_status.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/screen_R9_dm/figures/，不随文档站分发。