kahkengwong/cepip_example

cepip Example Data: SEA-AD Microglia

Author: Kah Keng Wong

• Example data of the R package cepip, loaded with cepip_example().
• 24,463 Micro-PVM_3-SEAAD cells from 10 brain regions, the table of 84 donors, and the DEG matrices.
• With these files, the PVM_3 and HKG8 control arms of the CEP-IP with CMRF analysis of SEA-AD microglia can be reproduced.

Quick Start

# install.packages("remotes")
remotes::install_github("kahkengwong/cepip")
library(cepip)

pvm3 <- cepip_example("pvm3")     # downloaded once, then read from the cache
d    <- pvm3[pvm3$region_set == "NeoC", ]
mt8  <- c("MT-ATP6", "MT-CO2", "MT-ND3", "MT-ND4", "MT-CYB", "MT-ND1", "MT-ND2", "MT-CO1")
y    <- cepip_composite(d, mt8)  # mean of the eight partner genes
res  <- cepip(d[["MT-CO3"]], y, donor = d$donor, id = d$barcode, cmrf = TRUE, workers = 4)
res

The Quick Start uses the default settings, so its values can differ from the published ones. The published values (IP_x 2.4953, IPRS 0.791) come from ip_k = 10 and y_spine = d$composite_part8, as in the worked example of the cepip README.

cepip_example() keeps the files in the R user cache folder, and cepip_example_clear() deletes them. The full reproduction script of the paper, which checks 698 published values, is described in the cepip README.

Example Output

Output of cepip on the 13,331 PVM_3 NeoC cells with the settings of the paper, drawn by readme_figure.R of the cepip repository. The aims, methods, code, and results of each pair of panels are in the worked example of the cepip README. The notes below explain what each pair of panels means.

Panels A and B: SKCS-core Cells and the Inflection Point

8,407 cells pass both SKCS gates (A). The residual GAM of the best-explained cells crosses zero at IP_x = 2.4953, with an IPRS of 0.791 (B).

(i) TREP and non-TREP cells.

• The number of TREP cells follows the deviance explained (DE) of the curve: they are the top round(n x DE) cells by EP.
• TREP cells are the cells the curve explains best. Non-TREP cells (blue and red in A) follow the same trend but lie farther from the curve.

(ii) Why a second GAM.

• The first GAM is one smooth curve through all cells, so it averages over a change in the relationship.
• The residuals of the best-explained cells show the change: these cells lie below the curve before the IP and above it after.
• The second GAM gives these residuals a standard error, so the IP gets a 95% CI and IPRS can score how clear the crossing is.

Panels C and D: the CMRF Band and the RSk-fCV Gate

1,994 non-SKCS-core cells fall inside the band (C), and the gate keeps 1,351 of them as Recovered cells (D).

(iii) Why the middle of the curve is lost.

• S_i is the product of the two centered ranks of a cell. A cell near the middle rank of either gene therefore has S_i close to 0, even when it lies on the curve.
• The fixed S_i gate removes this part of the curve (the gap in panel A). CMRF tests these cells by their distance to the curve instead.

(iv) What the gate protects against.

• The candidates are kept only when the monotone curve fitted with them predicts held-out cells no worse than the curve of the reference cells alone, within its background variability.
• Cells are removed from the farthest to the nearest, so the 643 removed candidates are the cells farthest from the curve.

Panels E and F: the Final Module and the Pitcher Plot

9,454 cells form the Module (E). In the pitcher plot, most Recovered cells lie left of the S_i gate line (F).

(v) Use of the Module.

• The Module, split at the IP into Pre-IP and Post-IP cells, is the unit of the downstream steps: the share of each donor's cells in each group (cepip_donor()), the DEGs between groups (cepip_deg()), and the association of a donor share with outcomes (cepip_associate()).

(vi) Shape of the pitcher plot.

• S_i and K_i are both set by the same two ranks of a cell, so they agree at the extremes.
• A cell with high ranks in both genes, or low ranks in both, has the highest S_i and K_i. The Post-IP (green) and Pre-IP (amber) branches therefore meet at the top-right tip.
• A cell high in one gene and low in the other has the lowest of both scores and lies at the bottom-left tip.
• Between the tips, a cell near the middle rank has S_i close to 0 whatever its K_i, which widens the body of the pitcher. The Recovered cells are such cells.

Files

File	Rows x columns	Contents
pvm3_cells.csv	24,463 x 23	Micro-PVM_3-SEAAD cells with MT-CO3 > 0 (NeoC 13,331; NonNeoC 11,132): barcode, region set, region, donor, MT-CO3, the eight mitochondrial partner genes, composite_part8 (the composite of the data export, single precision), ACTB, and the eight other housekeeping genes
hkg8_cells.csv	2,511 x 14	Cells of the HKG8 control arm (NeoC 1,432; NonNeoC 1,079): ACTB, the eight housekeeping genes, and composite_part8
donors.csv	84 x 18	Cognitive status, MMSE, CASI, CERAD, Braak, Thal, ADNC, CAA, APOE e4 dose, age at death, sex, PMI, RIN, brain pH, fresh brain weight, and the total Micro-PVM_3-SEAAD cells per donor and region set
pvm3_deg_NeoC.rds	20,654 genes x 13,331 cells	Sparse matrix of the NeoC PVM_3 cells for cepip_deg() (genes expressed in at least 40 cells)
pvm3_deg_NonNeoC.rds	19,449 genes x 11,132 cells	The same for the NonNeoC PVM_3 cells
build_deg_matrices.R		Script that built the two DEG matrices

The five data files are 82.8 MB in total. cepip_example() takes the names "pvm3", "hkg8", "donors", "deg_NeoC", and "deg_NonNeoC".

Data Details

• Expression values. ln(CP10K + 1) from the SEA-AD multiregion release of 22 June 2026.
• Region sets. NeoC (neocortical): DFC, MTG, AnG, STG, V1C, and ITG. NonNeoC (non-neocortical): HIP, MEC, LEC, and FI.
• Gene of interest. MT-CO3.
• Partner genes. MT-ATP6, MT-CO2, MT-ND3, MT-ND4, MT-CYB, MT-ND1, MT-ND2, and MT-CO1.
• HKG8 control arm. ACTB against eight other housekeeping genes, in the ACTB > 0 cells of the M4 arm of the paper.

Reproduced Values

The reproduction script of cepip runs the PVM_3 arm, the HKG8 control arm, and the same-cell housekeeping control on these files. All 698 published values agree within their tolerance, including 56 DEG values.

Value	PVM_3 NeoC	PVM_3 NonNeoC	HKG8 NeoC	HKG8 NonNeoC
Cells	13,331	11,132	1,432	1,079
IPx	2.4953	2.6572	1.5552	1.8130
IPRS	0.791	0.819	0.639	0.571
SKCS-core cells	8,407	6,941	250	199
OriRef / Recovered	8,103 / 1,351	6,646 / 1,089	239 / 287	191 / 240
Removed by the gate	643	619	65	22

Source Data

The files are derived from the SEA-AD immune snRNA-seq dataset (Multiregion release of 22 June 2026), available from the SEA-AD portal. The full processed data of the paper, for rerunning the whole pipeline, are in the Hugging Face dataset kahkengwong/CEP-IP_with_CMRF_in_SEA-AD.

Citation

If you use these data or cepip, please cite:

• K.K. Wong. CEP-IP: An explainable framework for cell subpopulation identification in single-cell transcriptomics. Computer Methods and Programs in Biomedicine, 282 (2026) 109372. https://doi.org/10.1016/j.cmpb.2026.109372
• M.I. Gabitto, K.J. Travaglini, V.M. Rachleff, E.S. Kaplan, B. Long, et al., Integrated multimodal cell atlas of Alzheimer's disease, Nat Neurosci, 27 (2024) 2366-2383. https://doi.org/10.1038/s41593-024-01774-5
• K.J. Travaglini, M.I. Gabitto, Y. Ding, A. Agrawal, N. Postupna, et al., Multiregional single-cell profiling reveals shared and specialized cellular vulnerability in Alzheimer's disease, bioRxiv, (2026). https://www.biorxiv.org/content/10.64898/2026.07.01.734821v1

License

The data are derived from the Seattle Alzheimer's Disease Brain Cell Atlas (SEA-AD) of the Allen Institute for Brain Science. Their use is subject to the Allen Institute Terms of Use (research and other noncommercial use) and Citation Policy. Commercial use needs the written permission of the Allen Institute. The cepip package code is licensed under the MIT License.