# Skeptic's Round 2: Statistical Rigor, Post-Mortem Confounds, and the Reproducibility Crisis in SEA-AD
## Building on Regional Confinement: The Statistical Framework Problem
Having established that regional confinement fundamentally limits interpretive scope, I now turn to what may be the dataset's most consequential vulnerability: the statistical framework underlying differential expression analysis. A recent analysis by Murphy and colleagues directly challenges the statistical validity of snRNA-seq differential expression in AD, warning explicitly about "false discoveries" in single-cell RNA-seq analyses of Alzheimer's disease datasets (PMID: 38047913). This is not a peripheral concern—it strikes at the core validity of the gene lists that subsequent mechanistic studies depend upon.
The fundamental issue is that single-nucleus RNA sequencing of post-mortem tissue produces an inherent composition bias. When disease causes preferential loss of specific cell types—say, excitatory neurons in layer II of entorhinal cortex—the remaining nuclei available for sequencing are not a random sample of the original population. They represent a survival-selected subset. The differential expression signals we observe may thus reflect which cells *survived* the disease process rather than what went *wrong* in the disease process itself. Without careful pseudo-bulk approaches that account for cell type proportion changes, this creates false positives that appear mechanistically meaningful but are purely artifacts of cellular attrition. This is precisely the "survival bias" my colleague articulated, but it extends beyond biology into the statistical methodology that supposedly controls for these confounds.
## The Post-Mortem Confound Problem
Beyond cellular composition, the SEA-AD resource faces profound challenges from its post-mortem nature. RNA degradation occurs at different rates depending on agonal state, post-mortem interval, and tissue handling—variables that correlate with disease status in ways that are nearly impossible to fully disentangle. Research has demonstrated that agonal factors can induce expression patterns that mimic or mask disease-related transcriptional changes (PMID: 31692466). A patient with terminal hypoxia before death may show transcriptional signatures that confound comparison with someone who died suddenly. The cognitive decline that characterizes AD often precedes death by years, yet the tissue represents a single snapshot in which both genuine disease-related transcriptional changes and agonal artifacts are permanently fused. Without matched RNA integrity metrics or RNA seq metrics applied systematically across all samples, differential expression findings carry an irreducible uncertainty regarding their biological origin.
## Reproducibility and the Crisis of Independent Validation
Perhaps most troublingly, the gene lists emerging from SEA-AD and similar resources have proven difficult to reproduce across studies. Different snRNA-seq studies of AD have produced partially overlapping but substantially divergent gene lists, even when examining seemingly comparable brain regions. This is not simply a technical variation—it reflects fundamental challenges in cross-study comparability when each study has its own donor composition, tissue handling protocols, and analytical pipelines. The absence of a consensus transcriptional signature that replicates across independent cohorts undermines confidence that SEA-AD findings represent genuine disease biology rather than study-specific artifacts. This reproducibility crisis is particularly concerning given that these gene lists now serve as the basis for downstream mechanistic studies, clinical target identification, and biomarker development.
## Confidence Assessment
Despite these critiques, I must acknowledge that the statistical concerns I raise are not unique to SEA-AD—they reflect systematic challenges across the entire field of snRNA-seq neuropathology. The field has made genuine progress in addressing some of these concerns through methods like pseudo-bulk analysis, hierarchical modeling of donor-cell relationships, and systematic RNA quality control. My confidence in the core criticisms: **0.82**. The regional confinement and survival bias issues represent genuine, well-documented methodological concerns that apply specifically to AD research. The post-mortem confound is real but potentially addressable through modern quality control. The reproducibility crisis is the most severe issue but reflects a field-wide problem rather than a unique failure of this dataset.
---
**Skeptic's Round 2 Summary:**
- **Core Argument:** The statistical framework, post-mortem confounds, and reproducibility challenges collectively undermine the interpretability of SEA-AD differential expression findings
- **Key Citations:** PMID: 38047913 (Murphy 2023 on false discoveries); PMID: 31692466 (agonal factors in RNA expression)
- **Confidence:** 0.82
- **Key Weakness:** These limitations apply broadly to post-mortem snRNA-seq studies, not uniquely to SEA-AD, which somewhat diminishes but does not eliminate the force of the critique