The Evidence Layer for Scientific AI

 Quick Analysis Plan Copied

 Long Analysis Plan

Financial and Structural Conflicts of Interest

The clearest financial COI in this evidence base is a NIH–industry CRADA (Cooperative Research and Development Agreement) with Novartis and SomaLogic, whose platform (SomaScan, 4,789 aptamers / 4,114 proteins) produced the data itself — a structural conflict where the assay vendor co-funds studies validating its own technology (Novartis/SomaLogic CRADA in CSF proteomics). Similarly, the Framingham biomarker study acknowledges a Gilead Sciences Research Scholar Award to Dr. Ho — a direct industry payment to the lead author (Gilead scholar award in Framingham biomarker study). Two reviews explicitly declared no competing interests (Self-declared no-COI aging biomarker review), but self-declaration alone cannot exclude indirect bias (academic incentive to promote 'omics clocks).

Methodological Bias Risks in Sex-Specific Biomarker Research

• Small-sample overfitting: The AD plasma biomarker study used only n=119 with ~79% accuracy and no external validation — a classic positive-result/overfitting risk (Small-sample AD sex-biased biomarker study).
• Platform selection bias: The Framingham panel (71 biomarkers from 85 candidates) was pre-enriched for known CVD pathways — not unbiased proteomics (Preselected CVD biomarker panel limitation).
• Population homogeneity: Framingham is predominantly white, limiting generalizability of sex-specific biomarker claims (Framingham cohort demographic limitation).
• Mouse-to-human extrapolation bias: Mouse strain studies (C57BL/6N vs BALB/c, n=9/group, 8-week-old only) show conserved sex biases (LPL, GHRL) but cannot validate human aging clocks (Mouse strain sex-differential proteomics study).
• Missing external validation in high-profile claims: The MS longitudinal CSF study (n=394, 1,040 samples) explicitly lacks an external validation cohort, and sex-differences in disability progression remain unreplicated (Unvalidated MS sex-difference proteomic findings).

Quantitative Sex-Difference Effect Sizes (Framingham, β values)

Blue bars = female-enriched; red = male-enriched. Data as reported (Reported sex β effect sizes).

Known Unknowns and What Would Change These Conclusions

• No independent replication of sex-specific biomarker-clock interactions in diverse populations — replication would either confirm or collapse the sex-specific claims.
• Cross-omics clock concordance is modest, and clocks built on cross-sectional data risk HARKing and p-hacking without longitudinal interventional tests (Modest cross-omics clock concordance).
• Publication bias toward positive sex-specific findings — null sex-interaction studies are underreported across the aging biomarker literature.
• Commercial incentives: proprietary platforms (SomaScan, Olink, Luminex) create vendor-driven incentive to overstate assay sensitivity and clinical utility before independent validation.