候选位点附近有什么基因，这能说明什么？Which genes lie near the candidates, and what does that show?
在共同门槛的工作口径下，1a 没有一对达到门槛。Under the working setting of the common bar, no pair reached it in 1a.
按 p 值事先选定：B 的第 1 名和 A2 的第 1 名。选择时没有看注释。Chosen in advance by p-value: B's top pair and A2's top pair. No annotation was used in choosing them.
GO、KEGG、STRING、FLOR-ID。注释只用来描述，不作为支持某一对的证据。GO, KEGG, STRING and FLOR-ID. Annotations only describe; they are not evidence for a pair.
还没做：Not yet done: 富集检验（1a 没有稳健的对，按工作流改为逐个写成案例）；逐篇核对文献；在独立的另一批品系中验证。enrichment testing (1a has no robust pair, so per the workflow the examples are written up as cases); checking the literature paper by paper; validation in an independent set of lines.
注释说明"附近有什么"，不说明"互作是真的"Annotation shows what lies nearby, not that an interaction is real
这一步的规则在运行前写进了 config：结果分成三层，每层能说的话不同。第 1 层是达到共同门槛的对；第 2 层是排名靠前但没有达到门槛的探索性候选；第 3 层是基因名称、功能描述、是否已知开花基因等注释。第 3 层只用来描述。挑选候选时用过的信息，更不能再拿来当证据。The rules for this step were written into the config before the run: results fall into three levels, and each level supports different statements. Level 1 is pairs that reach the common bar; level 2 is exploratory candidates that rank high but do not reach it; level 3 is annotation such as gene names, function descriptions and whether a gene is a known flowering gene. Level 3 only describes. Information already used to choose a candidate can never count as evidence for it.
把 SNP 对应到基因用了三种方式：落在哪个基因 ± 2 kb 的区间里；最近的基因及其方向（按链判断上下游）；± 20 kb 内与它 r² ≥ 0.5 的标签 SNP 落在哪些基因里。SNPs were mapped to genes in three ways: which gene's ± 2 kb interval they fall in; the nearest gene and its direction (upstream or downstream judged by strand); and which genes contain the tag SNPs within ± 20 kb with r² ≥ 0.5.
2:8009643 × 4:4737947
这一对只作为假象检查的示范，不作为值得进一步验证的候选。This pair serves only as a demonstration of artefact checks, not as a candidate worth validating.
四格的品系数和平均现蕾时间。在 Step 2 的位点对浏览器里可以看到每个品系。Line counts and mean time to buds per cell. Every line is shown in the Step 2 pair browser.
- 工作口径 pWorking p
- 5.16 × 10⁻¹¹ （门槛 2.87 × 10⁻¹¹，未达到）(bar 2.87 × 10⁻¹¹, not reached)
- 逐对重估方差比Per-pair variance ratio
- 9.0 × 10⁻¹⁴ （敏感性分析）(sensitivity)
- 秩变换 / 逐个剔除Rank / leave-one-out
- 1.3 × 10⁻⁸ / 3.8 × 10⁻⁹
- 最小格Smallest cell
- 13 个品系，6 个（46%）来自意大利；意大利品系占全体 7.8%13 lines, 6 (46%) from Italy; Italian lines are 7.8% of all
- 两端Ends
- 不同染色体，r² ≈ 0；MAF 0.11、0.12different chromosomes, r² ≈ 0; MAF 0.11, 0.12
- 出现次数Reuse
- 4:4737947 在 B 复检的 1,000 对中出现 173 次，在前 10 对中 1 次4:4737947 appears in 173 of B's 1,000 re-tested pairs and once in the top 10
- Step 3
- 只在 2/5 个训练折进入 B 的前 20 对；这两折里前 20 对整组的 Δr < 0，不能归因于这一对In B's top 20 in only 2 of 5 training folds; in those folds the top 20 as a whole had Δr < 0, which cannot be pinned on this pair
附近的基因（第 3 层）Nearby genes (level 3)
2:8009643 在 AT2G18480（主要易化子超家族蛋白）的基因体内，也在 PERK4（AT2G18470）的 ± 2 kb 区间内。4:4737947 不在任何基因的 ± 2 kb 区间内，最近的 AT4G07932 功能未知，在它 3′ 下游 4.5 kb。这几个基因都不在 FLOR-ID 的 2024-05-09 存档中。2:8009643 lies inside AT2G18480 (a major facilitator superfamily protein) and within the ± 2 kb interval of PERK4 (AT2G18470). 4:4737947 lies in no gene's ± 2 kb interval; the nearest gene, AT4G07932, has unknown function and lies 4.5 kb 3′ downstream. None of these genes is in the 2024-05-09 archive of FLOR-ID.
解读Interpretation
两个少数等位基因同时出现的 13 个品系比其他三格晚约 19–27 天，其中 6 个来自意大利；秩变换和逐个剔除都让信号明显变弱。一个更简单的解释是：一小群晚花品系同时带着两个少数等位基因，K 没有完全吸收这种结构。目前不能区分这是真实的上位性，还是群体结构或极少数品系造成的假象。如果以后要追查，先看这 13 个品系的来源和亲缘关系。The 13 lines carrying both minor alleles are about 19–27 days later than the other three cells, and 6 of them come from Italy; rank transformation and leave-one-out both weaken the signal clearly. A simpler explanation is that a small group of late lines carries both minor alleles and K does not fully absorb this structure. It cannot yet be told apart from true epistasis or from an artefact of structure or a handful of lines. Any follow-up should start with the origin and relatedness of these 13 lines.
5:3126806 × 5:18590741
这一对说明的是：边际筛选主要找回已知的主效位点。两端靠近已知开花相关基因，解释的是它为什么被选中，不是互作的证据。This pair shows that marginal screening mainly recovers known main-effect loci. That its ends lie near known flowering-related genes explains why it was selected; it is not evidence of an interaction.
在 Step 2 的位点对浏览器里可以看到每个品系。Every line is shown in the Step 2 pair browser.
- 工作口径 pWorking p
- 5.7 × 10⁻³ （远未达到门槛；也没达到 54 对内部的 Bonferroni 9.3 × 10⁻⁴）(far from the bar; also short of the within-54-pairs Bonferroni 9.3 × 10⁻⁴)
- 稳健性Robustness
- 加入邻近边际 SNP 0.033；逐对重估 0.0051；秩变换 0.089；逐个剔除最差 0.0091adding nearby marginal SNPs 0.033; per-pair ratio 0.0051; rank 0.089; worst leave-one-out 0.0091
- 单 SNP 信号（MLM）Single-SNP signal (MLM)
- 5:3126806 p = 2.7 × 10⁻⁵；5:18590741 p = 8.9 × 10⁻⁸5:3126806 p = 2.7 × 10⁻⁵; 5:18590741 p = 8.9 × 10⁻⁸
- 两端Ends
- 同在 5 号染色体，相距 15.5 Mb，r² = 0.02both on chromosome 5, 15.5 Mb apart, r² = 0.02
- 出现次数Reuse
- 两端在 A2 的 54 对中各出现 10 次each end appears in 10 of A2's 54 pairs
附近的基因（第 3 层）Nearby genes (level 3)
5:18590741 在 DOG1（AT5G45830）的基因体内。DOG1 是已知的种子休眠基因，原始研究报告它也通过 miR156 / miR172 途径影响开花时间（Huo et al. 2016）。5:3126806 在 FD4（铁氧还蛋白 4）的基因体内，离 FLC 约 47 kb，处在 Step 1 的 FLC 附近主效峰所在区域；± 20 kb 的标签窗口没有到达 FLC，所以它是否标记 FLC 的变异没有确立。5:18590741 lies inside DOG1 (AT5G45830). DOG1 is a known seed-dormancy gene, and the original study reports that it also affects flowering time through the miR156 / miR172 pathway (Huo et al. 2016). 5:3126806 lies inside FD4 (ferredoxin 4), about 47 kb from FLC, in the region of Step 1's main-effect peak near FLC; the ± 20 kb tag window does not reach FLC, so whether it tags an FLC variant is not established.
解读Interpretation
按四格均值，DOG1 端的效应在另一端的两种背景下分别约 +27.5 天和 +14.2 天，方向上像"小于相加"，但检验远没有达到门槛。按工作流的写法，它最多算证据很弱的候选。要推进，需要在另一批独立品系中预先指定这一对再检验，同时排除 FLC 区域的等位异质性。FT10 与 DTF1 共有 935 个品系，只能算另一种条件下的稳健性检查，不是独立验证。From the cell means, the DOG1 end's effect is about +27.5 days in one background of the other end and +14.2 days in the other, which looks "less than additive", but the test is far from the bar. By the workflow's wording it is at best a weakly supported candidate. Taking it further would need a pre-specified test in an independent set of lines, while ruling out allelic heterogeneity in the FLC region. FT10 shares 935 lines with DTF1, so it is only a robustness check under another condition, not independent validation.
"没有记录"只是查询结果，不是否定功能的证据"No record" is a query result, not evidence against a function
| 基因Gene | 端点End | FLOR-ID | 有实验证据的 GO 生物过程GO biological processes with experimental evidence | KEGG | STRING ≥ 700 伙伴中的 FLOR-ID 基因FLOR-ID genes among STRING ≥ 700 partners |
|---|---|---|---|---|---|
| PERK4 | 2:8009643 | 否no | 4（如脱落酸信号、种子萌发）4 (e.g. abscisic acid signalling, seed germination) | — | — |
| AT2G18480 | 2:8009643 | 否no | 1（侧根形成）1 (lateral root formation) | — | AP2, AGL15 |
| AT4G07932 | 4:4737947 | 否no | 0（只有 3 个 ND 根条目）0 (only 3 ND root entries) | — | — |
| FD4 | 5:3126806 | 否no | 0 | 光合作用photosynthesis | — |
| AT5G10010 | 5:3126806 | 否no | 5（如热响应、表观调控）5 (e.g. heat response, epigenetic regulation) | — | — |
| CIPK20 | 5:18590741 | 否no | 1 | — | — |
| DOG1 | 5:18590741 | 不在本快照中（原因未核实）not in this snapshot (reason not checked) | 5（如脱落酸响应、种子休眠）5 (e.g. abscisic acid response, seed dormancy) | — | GA3OX1, LDL1, HUB1, RGL2, LDL2 |
两对的两端之间，在 STRING 12.5 里都没有检出关联记录。FLOR-ID 用的是 2024-05-09 的网页存档，存档日期不等于内容的更新日期。DOG1 不在这份快照里，但原始研究报告它影响开花，所以"不在表中"不能理解为"与开花无关"。In STRING 12.5 no association record was found between the two ends of either pair. FLOR-ID is the web archive of 2024-05-09, and the archive date is not the date the content was last updated. DOG1 is missing from this snapshot although the original study reports that it affects flowering, so "not in the list" cannot be read as "unrelated to flowering".
复核中发现、已更正的三个问题Three problems found in review and corrected
把整组的预测下降算到一对头上A whole group's prediction loss pinned on one pair
最初写示例 1"在测试折没有提高预测"。Step 3 只报告前 20 对整组的 Δr，单独一对的贡献并没有算过。Example 1 was first described as "not improving prediction in the test folds". Step 3 reports Δr only for the top 20 pairs as a group; the contribution of a single pair was never computed.
处理：改为"只在 2 折进入前 20 对，这 20 对整组 Δr < 0，不能归因于这一对"。Correction: now "entered the top 20 in only 2 folds; that group of 20 had Δr < 0, which cannot be pinned on this pair".
6/13 被写成"多数"6 of 13 written as "most"
最小格里来自意大利的品系是 6 个（46%），最初被写成"多数"或"为主"。Six (46%) of the smallest cell's lines are from Italy; this was first written as "most" or "mainly".
处理：直接写数字 6/13，并给出意大利品系占全体的 7.8% 作对照。Correction: state the number, 6 of 13, with Italy's 7.8% share of all lines for comparison.
出现次数没有注明来自哪个策略、哪个范围Reuse counts without their strategy and scope
示例 2 的出现次数最初直接取了另一张表里第一条匹配的记录，那一条来自 A1 严格版，端点顺序也相反，所以既不是 A2 的计数，也没有对齐。Example 2's reuse counts were first taken from the first matching row of another table; that row came from A1 strict with the ends in reverse order, so it was neither A2's count nor aligned.
处理：出现次数按策略和范围分列（例如"在 A2 的 54 对中各 10 次"），并按端点对齐。Correction: reuse counts are now given per strategy and scope (for example "10 times each among A2's 54 pairs") and aligned by end point.
算一算最小格的富集倍数Work out the enrichment in the smallest cell
python3 -c "
from scipy.stats import binomtest
k, n, p0 = 6, 13, 0.078 # Italian lines in the smallest cell, cell size, share of all lines
print('fold enrichment', round(k / n / p0, 1))
print('binomial p (one-sided)', f'{binomtest(k, n, p0, alternative=\"greater\").pvalue:.2g}')
"应得到约 5.9 倍。这个 p 值能说明什么、不能说明什么？提示：13 个品系是在看过结果后才挑出来的。You should get about 5.9-fold. What can this p-value show, and what not? Hint: the 13 lines were picked out after the result was seen.
示例 2 的两端分别靠近 DOG1 和 FLC 区域。这能作为这对互作的证据吗？Example 2's ends lie near DOG1 and the FLC region. Is that evidence for this interaction?
富集检验Enrichment testing
等有了稳健的对再做。背景要从各策略自己的搜索空间里抽，并匹配 MAF、LD 和四格人数，否则富集只反映搜索空间的偏向。To be done once robust pairs exist. The background must be drawn from each strategy's own search space and matched on MAF, LD and cell sizes; otherwise enrichment only reflects the bias of the search space.
逐篇核对文献Checking the literature
目前只核对了 DOG1 的一篇原始研究，其他基因的文献还没有逐篇读过。So far only one original study on DOG1 has been checked; the literature for the other genes has not been read paper by paper.
写下你自己的解释Write your own explanation
为示例 2（A2 第 1 名）写一段有边界的说明：它属于第几层证据，两端靠近哪些已知区域，这能说明什么、不能说明什么，下一条需要的证据是什么。Write a bounded description of example 2 (A2's top pair): which level of evidence it belongs to, which known regions its ends are near, what that does and does not show, and what the next piece of evidence would be.
只保存在这个浏览器里，换设备或清除浏览器数据后会丢失。所有页面的记录可以在总览页的“我的学习记录”里一起导出。Saved in this browser only; it is lost on another device or after clearing browser data. The records of all pages can be exported together under “My learning record” on the overview page. 去导出Go to export