Blog

Aging clocks are typically trained on pooled multi-tissue data, implicitly assuming that aging is uniform across organs. Here we challenge this assumption by constructing 49 transcriptomic clocks across 47 tissue types and 2 cell-line categories from GTEx v8 (948 unique donors contributing to the retained clock categories) using donor-grouped cross-validation. Clocks achieved a median Pearson r of 0.543 (best: artery aorta, r = 0.855). We identified 10,253 unique clock genes, of which 69.2% appeared in only one tissue; however, null simulation confirmed that this low overlap is the expected consequence of sparse elastic-net selection rather than biological tissue-specificity. We then projected 3,926 LINCS L1000 compound-name entries onto each clock to build a drug × category age-reversal matrix. At a permissive threshold (|score|> 1.0), 94.2% of drugs showed mixed score directions (positive in some categories and negative in others). This permissive mixed-direction proportion was descriptive and did not itself exceed shuffled expectations. Under a more stringent exploratory criterion requiring |score|> 2.0 in at least three categories in each direction, 2.8% of compound entries showed pronounced bidirectional divergence, compared with approximately 0.1% under the shuffled null. Compound rankings remained stable when analysis was restricted to the 34 clocks with r ≥ 0.5 (Spearman ρ = 0.897 versus the full analysis). A Jaccard-based enrichment statistic yielded highly concordant compound rankings (median per-category Spearman ρ = 0.961), indicating that results were not artifacts of the enrichment method. After Benjamini–Hochberg FDR correction across all 192,374 drug-category pairs, only 0.79% reached FDR < 0.05, indicating that individual drug-category calls require caution. Rapamycin showed net pro-aging transcriptional signatures in our system; we explicitly emphasize that transcriptome-based scores and organismal lifespan are distinct endpoints. DepMap CRISPR analysis was repeated after GTEx-based standardization across 13 matched categories, although individual gene-level results were limited. Four statistical robustness tests confirmed model stability. These findings describe category-associated drug-score patterns and provide a tissue-aware resource for generating hypotheses about drug responses, while cautioning against over-interpretation of individual drug-category predictions without experimental validation.