Palmitoylation is a reversible lipid modification in which fatty acid chains are attached to proteins, anchoring them to cellular membranes and controlling where they travel and with whom they interact. Because the brain is packed with membrane-intensive machinery, from synaptic vesicles to myelin sheaths, palmitoylation is deeply involved in neural function. Yet its cell-type-specific behavior in Alzheimer’s disease had never been systematically characterized. The research team set out to fill that gap by asking a deceptively simple question: which cells in the Alzheimer’s brain show altered palmitoylation activity, and can the genes behind that change be distilled into a reliable diagnostic signature?

To answer it, the team assembled an unusually broad data foundation. They analyzed a single-cell RNA sequencing dataset containing 48,714 individual cells from human brain tissue, five bulk transcriptomic cohorts totaling 769 samples, and spatial transcriptomic data from 16 brain sections generated with the 10x Visium platform. The single-cell data allowed them to score palmitoylation activity in each cell individually using a curated set of 46 genes, while the bulk cohorts provided statistical power across many patients and the spatial data preserved the crucial question of where in the brain’s architecture the signals actually reside. This three-pronged design matters because gene expression changes only become biologically meaningful when they can be tied to specific cells in specific locations.

The first major finding was unambiguous: oligodendrocytes emerged as the cell population most strongly enriched for palmitoylation activity in the Alzheimer’s brain. Oligodendrocytes are the myelin-producing cells of the central nervous system, wrapping axons in fatty insulation that speeds electrical signaling. Their involvement in Alzheimer’s has gained traction in recent years, with evidence of myelin breakdown and white matter degeneration preceding or accompanying cognitive decline, but they remain far less studied than neurons, astrocytes and microglia. The new results, supported by per-cell-type scoring, odds ratio enrichment analysis and a 658-gene oligodendrocyte-anchored palmitoylation module, suggest that defective lipid modification within these insulating cells may be a previously underappreciated dimension of the disease.

From there, the researchers narrowed the field. By intersecting the oligodendrocyte palmitoylation module with genes known to be differentially expressed in Alzheimer’s disease, they produced a list of 89 candidate genes. That list then passed through a machine learning gauntlet: three complementary algorithms, LASSO regression, random forest and XGBoost, were each used to identify the genes with the greatest diagnostic value, and the overlapping winners formed a parsimonious seven-gene panel. The final panel comprised ANKIB1, SPAG9, HBP1, SLC25A29, FAM171A1, ATXN10 and NDUFV1. Using multiple algorithms in parallel is a deliberate safeguard against the overfitting that plagues many biomarker studies, since a gene must prove its worth under different mathematical assumptions to survive the cut.

One gene stood above the rest. ANKIB1, which encodes an ankyrin repeat and IBR domain-containing protein, consistently topped every layer of the analysis. It achieved the highest area under the receiver operating characteristic curve for distinguishing Alzheimer’s disease from healthy tissue, at 0.813 with a 95 percent confidence interval of 0.782 to 0.844, a level of discrimination that places it among the more promising transcriptomic candidates reported for the disease. Critically, ANKIB1 also showed persistent spatial colocalization with oligodendrocyte-enriched regions across both early-stage Alzheimer’s disease and established disease, meaning its expression pattern tracks with the geography of myelin-forming cells as the pathology advances.

The spatial analysis deserves particular attention because it employed four complementary metrics rather than a single test. The team calculated the Jaccard index, the overlap coefficient, the colocalization percentage and the odds ratio to evaluate whether gene expression signals and specific cell populations genuinely occupy the same tissue neighborhoods. Requiring agreement across all four measures reduces the risk that a colocalization signal is a statistical artifact. In the Alzheimer’s field, where spatial transcriptomics is rapidly becoming a standard tool, this kind of methodological rigor sets a useful benchmark for how candidate biomarkers should be validated before moving toward experimental models.

Beyond diagnosis, the study ventured into therapeutic territory using network pharmacology, a computational approach that maps the interactions between genes, proteins and known drugs. This analysis nominated AHR, the aryl hydrocarbon receptor, a ligand-activated transcription factor with roles in immune regulation and neuroinflammation, and TRIM11, a member of the tripartite motif protein family involved in protein quality control, as candidate upstream regulators of the palmitoylation-linked program. The team also catalogued compounds with reported interactions with the biomarker genes, creating a shortlist that could guide future drug screening. The authors are careful to frame these as hypotheses for further work rather than validated targets, but the framework provides a concrete starting point for experimental follow-up.

The broader significance of the study lies in how it reframes Alzheimer’s biology. Most biomarker research has centered on amyloid and tau, the proteins that define the disease pathologically, and on the neurons that die as the disease progresses. By demonstrating that a lipid modification program anchored in oligodendrocytes carries a robust disease signature, the work adds myelin biology and post-translational regulation to the list of processes that a complete model of Alzheimer’s must explain. It also fits a growing recognition that the brain’s supporting cells are active participants in neurodegeneration rather than passive bystanders, and that white matter changes may be as informative as the plaques and tangles that dominate conventional imaging and fluid biomarkers.

Caveats remain, as they do for any computational biomarker study. The analysis relied entirely on publicly available de-identified datasets from the Gene Expression Omnibus, so the findings await confirmation in independent, prospectively collected cohorts and in experimental systems where ANKIB1’s function in oligodendrocytes can be directly manipulated. The reported diagnostic performance reflects classification of tissue transcriptomes, not a blood test or clinical diagnostic, and translating a seven-gene panel into a practical clinical tool would require substantial further development. Still, the convergence of evidence is notable: a single gene emerging from single-cell scoring, bulk cohort validation, machine learning selection and spatial colocalization is a pattern few candidates achieve. If ANKIB1 and the oligodendrocyte-palmitoylation axis withstand that scrutiny, the study may mark the moment when the brain’s insulating cells, and the fatty chemistry that keeps their proteins in place, moved from the margins of Alzheimer’s research toward its center.

Subject of Research: Identification of the palmitoylation-associated gene ANKIB1 in oligodendrocytes as a candidate biomarker of Alzheimer's disease using single-cell, spatial and machine learning analyses

Article Title: Single-cell, spatial, and machine learning analyses identify ANKIB1 as an oligodendrocyte-anchored palmitoylation-associated candidate biomarker of Alzheimer’s disease

Article References: Liang, J., Xu, Z., Zhu, Z., Liang, F., Tang, S., Deng, Y., Zhang, X., Chen, H., Lin, Y., & Zhou, J. (2026). Single-cell, spatial, and machine learning analyses identify ANKIB1 as an oligodendrocyte-anchored palmitoylation-associated candidate biomarker of Alzheimer’s disease. Translational Psychiatry. https://doi.org/10.1038/s41398-026-04463-y

Image Credits: AI Generated

DOI: 10.1038/s41398-026-04463-y

Keywords: Alzheimer's disease, ANKIB1, oligodendrocytes, palmitoylation, single-cell RNA sequencing, spatial transcriptomics, machine learning, biomarkers, myelin, network pharmacology, transcriptomics, neurodegeneration