Answers are grounded in this paper and cite their sources.
Background/Objectives: Public blood transcriptome datasets have been used to argue that mild cognitive impairment (MCI) resembles Alzheimer’s disease (AD). We asked where MCI lies between cognitively normal controls (CTLs) and AD once the technical structure is handled. Methods: GSE140829 was analysed within the 15 processing batches holding all three groups (243 CTL, 131 MCI, and 201 AD) and the AddNeuroMed batches, GSE63060 and GSE63061, separately, without pooling. Broad separation was tested with 300 random sets of 200 genes or probes. AD-versus-CTL axes were built using a leave-one-batch-out approach, Hedges’ g was pooled by random effects, and the MCI position and its interval were estimated using bootstrap resampling. An elastic-net classifier and eight leukocyte composition scores were used as checks. Results: In two AddNeuroMed batches, random 200-probe sets separated AD from CTL (median AUCs of 0.73 and 0.65), and an all-centre axis placed MCI at 0.73–1.04 on a CTL (0) to AD (1) scale. In GSE140829, random sets gave an AUC of 0.59, falling to 0.52 within batches. Inside batches, with each batch scored by an axis built without it, AD differed from CTL by Hedges’ g = 0.32 [95% CI 0.12, 0.52], and a leave-one-batch-out classifier reached an AUC of 0.57 (p = 0.033 against a null of 5000 label permutations). MCI differed from CTL (g = 0.25 [0.02, 0.48]; 0.17 [−0.06, 0.40] with APOE ε4 adjustment) and not from AD (g = 0.00 [−0.23, 0.23]). Its position was 0.78 [−0.30, 1.81] on genes and 0.44 [0.00, 0.84] on leukocyte composition scores. Both intervals contain the midpoint (0.5), so the direction of MCI could not be determined. Conclusions: Within batches, blood gene expression carries a small AD signal, but the position of MCI between controls and AD is not determined by these data: the interval covers controls, the midpoint, and AD alike, so the data are neither evidence that MCI resembles AD nor evidence that it resembles cognitively normal ageing.
No takes yet. Share an insight, caveat, or question.
Lee et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: