NF04 · Real multiple-testing correction on the CP↔motility associations¶

nb06's callout states "41 associations survive Bonferroni... 87 survive Benjamini-Hochberg FDR 5%", but auditing the actual notebook found no multipletests/fdr_bh code anywhere — only raw Spearman p-values are computed (assoc in nb06 cell 7). Bonferroni is trivial to check by eye (p < 0.05/640), but the BH-FDR number was apparently never actually run. This notebook adds real, reproducible statsmodels.stats.multitest.multipletests(method="fdr_bh") on the existing 640-test assoc table and reports what it actually finds.

640 tests (160 CP features x 4 motility metrics)
Bonferroni (p < 7.81e-05): 41 significant
BH-FDR 5%: 87 significant
(nb06's callout previously *stated* 41 / 87 without a code cell computing them)

BH-significant associations by motility metric:
metric
mean_step_um           69
net_displacement_um     7
straightness            7
path_length_um          4
dtype: int64

Top 10 by adjusted p-value:

	feature	metric	rho	p	p_adj	bh_significant
575	RadialDistribution_RadialCV_OrigActin_Golgi_Me...	mean_step_um	0.158658	4.224915e-11	2.703946e-08	True
571	RadialDistribution_RadialCV_OrigActin_Golgi_Me...	mean_step_um	0.148550	6.792659e-10	2.173651e-07	True
183	Intensity_MassDisplacement_OrigActin_Golgi_Mem...	mean_step_um	0.141022	4.773415e-09	9.228479e-07	True
455	RadialDistribution_FracAtD_OrigMito_2of4	mean_step_um	-0.140270	5.767799e-09	9.228479e-07	True
439	RadialDistribution_FracAtD_OrigER_2of4	mean_step_um	-0.138113	9.871156e-09	1.263508e-06	True
451	RadialDistribution_FracAtD_OrigMito_1of4	mean_step_um	-0.137073	1.275226e-08	1.360241e-06	True
343	Intensity_StdIntensityEdge_OrigActin_Golgi_Mem...	mean_step_um	0.134430	2.424637e-08	2.017934e-06	True
463	RadialDistribution_FracAtD_OrigMito_4of4	mean_step_um	0.134265	2.522417e-08	2.017934e-06	True
435	RadialDistribution_FracAtD_OrigER_1of4	mean_step_um	-0.133118	3.319727e-08	2.360694e-06	True
447	RadialDistribution_FracAtD_OrigER_4of4	mean_step_um	0.131863	4.471434e-08	2.861718e-06	True

All results that survive BH-FDR¶

The full set of 87 BH-significant CP<->motility associations -- not just the top 10 -- as a browsable table and a single overview plot, so every surviving result (not only the strongest) is visible in one place. (A scrubber view that highlights the specific cells driving each of these -- e.g. high/low mean_step_um colored on the motility panel -- is planned as a follow-up investigation page, not built here.)

87 associations survive BH-FDR 5% -- full list:

	feature	metric	rho	p	p_adj
0	RadialDistribution_RadialCV_OrigActin_Golgi_Me...	mean_step_um	0.158658	4.224915e-11	2.703946e-08
1	RadialDistribution_RadialCV_OrigActin_Golgi_Me...	mean_step_um	0.148550	6.792659e-10	2.173651e-07
2	Intensity_MassDisplacement_OrigActin_Golgi_Mem...	mean_step_um	0.141022	4.773415e-09	9.228479e-07
3	RadialDistribution_FracAtD_OrigMito_2of4	mean_step_um	-0.140270	5.767799e-09	9.228479e-07
4	RadialDistribution_FracAtD_OrigER_2of4	mean_step_um	-0.138113	9.871156e-09	1.263508e-06
...	...	...	...	...	...
82	Intensity_MinIntensity_OrigDNA	path_length_um	-0.068014	4.909864e-03	3.785919e-02
83	AreaShape_BoundingBoxArea	net_displacement_um	-0.066704	5.805065e-03	4.416094e-02
84	Intensity_MinIntensityEdge_OrigDNA	path_length_um	-0.066623	5.865125e-03	4.416094e-02
85	AreaShape_MinFeretDiameter	straightness	-0.065864	6.454097e-03	4.763719e-02
86	RadialDistribution_FracAtD_OrigActin_Golgi_Mem...	net_displacement_um	0.065837	6.475680e-03	4.763719e-02

87 rows × 5 columns

No description has been provided for this image

saved p_adj / bonferroni_significant / bh_significant back to cp_motility_spearman.parquet

⭐ Key finding¶

The previously-stated numbers check out: 41 survive Bonferroni (p < 7.81e-05) and 87 survive BH-FDR 5%, exactly matching nb06's callout — so the earlier claim was correct, it just had no reproducible code behind it (now fixed: p_adj/bonferroni_significant/ bh_significant are saved back to cp_motility_spearman.parquet).

Breaking the 87 BH-significant hits down by metric: 69 are with mean_step_um, only 7 with net_displacement_um, 7 with straightness, 4 with path_length_um — reinforcing nb06's finding that speed, not displacement/path/straightness, is what CP morphology tracks. Top hits are unchanged from nb06: RadialDistribution_RadialCV_OrigActin_Golgi_Membrane_* and RadialDistribution_FracAtD_Orig{Mito,ER}_*.