Technical Robustness of Transcriptomic Senescence Classification in Human Microglia: An Evaluation of SenFlag and ICE
- 1Kocaeli Vocational School of Health Services, Kocaeli University, Aydin 41100, Türkiye
- 2Eureka Patent, Kocaeli Technopark, Kocaeli 41275, Türkiye
Abstract
The dependence of senescence labels on technical choices becomes apparent when they are used with sparse data from the human brain. We have assessed the SenFlag rules together with five different ICE initialisations in publicly available microglial single-nucleus RNA sequencing datasets. The discovery cohort included 12,220 nuclei obtained from 17 donors; the full ICE perturbation analyses were carried out on a fixed subset of 1823 nuclei. The external cohort consisted of 9371 nuclei from ten donors. All 85 complete-pipeline donor-removal fits were carried out. The median positive-label Jaccard agreement for SenMayo-initialised ICE was 95.3%, as compared with 80.0%, 29.4%, 17.6% and 1.7% for the ATF3, CDKN1A, CDKN2A and MX1 initialisations, respectively. Out of 40 attempts, 39 estimable chains were obtained under six matched molecule-thinning conditions and two MAGIC initialisation controls. The median agreement for SenMayo-initialised ICE was 95.5% at 50% count retention and 0.0% at 25%. The positive labels from the discovery cohort differed when using example constants as opposed to population-relative thresholds (91 versus 12) and decreased following thinning. In the external cohort, three ICE chains completed before depth filtering and one afterwards; no initialisation completed in both conditions. Technical stability, estimability and biological validity are separate concepts. Before biological interpretation, senescence-associated labels must have explicit operational definitions and robustness reporting.
Related
- In Silico Single-Cell Transcriptomic Analysis Reveals Senescence-Associated Programs’ Progress Along the Normal–Adenoma–Carcinoma Sequence in Colorectal Cancer
- Strategic roadmap for FAIR-compliant transcriptomic workflows: from technical challenges to bioinformatics democratization
- Systematic evaluation of robustness of deconvolution methods for spatial transcriptomics data in case of cell type mismatch.
- Senomorphic miR-3155b suppresses the senescence-associated secretory phenotype in therapy-induced senescence
- G3BP1 reduction during cellular senescence promotes SVIL expression and selected senescence-associated phenotypes
BibTeX
@article{yuksel2026technical,
title = {{Technical Robustness of Transcriptomic Senescence Classification in Human Microglia: An Evaluation of SenFlag and ICE}},
author = {Burcu Yuksel and Levent Aydin},
journal = {International Journal of Molecular Sciences},
year = {2026},
month = nov,
doi = {10.3390/ijms272210273},
url = {https://www.mdpi.com/1422-0067/27/22/10273},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Technical Robustness of Transcriptomic Senescence Classification in Human Microglia: An Evaluation of SenFlag and ICE
AU - Burcu Yuksel
AU - Levent Aydin
PY - 2026
DA - 2026/11/16
JO - International Journal of Molecular Sciences
DO - 10.3390/ijms272210273
UR - https://www.mdpi.com/1422-0067/27/22/10273
AB - The dependence of senescence labels on technical choices becomes apparent when they are used with sparse data from the human brain. We have assessed the SenFlag rules together with five different ICE initialisations in publicly available microglial single-nucleus RNA sequencing datasets. The discovery cohort included 12,220 nuclei obtained from 17 donors; the full ICE perturbation analyses were carried out on a fixed subset of 1823 nuclei. The external cohort consisted of 9371 nuclei from ten donors. All 85 complete-pipeline donor-removal fits were carried out. The median positive-label Jaccard agreement for SenMayo-initialised ICE was 95.3%, as compared with 80.0%, 29.4%, 17.6% and 1.7% for the ATF3, CDKN1A, CDKN2A and MX1 initialisations, respectively. Out of 40 attempts, 39 estimable chains were obtained under six matched molecule-thinning conditions and two MAGIC initialisation controls. The median agreement for SenMayo-initialised ICE was 95.5% at 50% count retention and 0.0% at 25%. The positive labels from the discovery cohort differed when using example constants as opposed to population-relative thresholds (91 versus 12) and decreased following thinning. In the external cohort, three ICE chains completed before depth filtering and one afterwards; no initialisation completed in both conditions. Technical stability, estimability and biological validity are separate concepts. Before biological interpretation, senescence-associated labels must have explicit operational definitions and robustness reporting.
SN - 1422-0067
ER -