RootMap · Preprint in submission
RootMap maps conditional dependencies between aging hallmarks and organ-aging patterns, from blood DNA methylation.
Patterns in blood-based scores — not proof of biological control.
Preprint · in submission 2026
332
gene modules
656 + 1,394
two cohorts — discovery + replication
22
hallmark–organ pairs passed four checks
17/20
kept direction in the replication cohort
Why
Protein–protein interaction (PPI) networks show which molecules are connected. They do not, by themselves, show which maintenance processes sit upstream of particular organ functions.
Aging-hallmark catalogues name processes shared across aging and age-related disease. They are used as a list of mechanisms — not as relationships between foundational processes and specific organ functions.
Longevity medicine has to decide what to measure, which candidate targets to investigate, and what to follow up in longitudinal and N-of-1 studies.
How
Every gene module gets a score in every blood sample: one minus the normalized module aging index (mIC, module-level intrinsic capability), with bounds frozen in a reference cohort. A higher mIC means a higher estimated capability on that reference scale.
Split participants into higher and lower age-adjusted hallmark-score groups, then compare the organ module's association with age between the two groups. A positive Δz means the association is more positive in the higher-score group; a negative Δz means it is more negative.
Multiple-testing correction; random-direction null gene sets; an age-proxy check, so the split does not simply separate younger from older participants; and random gene-set controls. Cross-cohort replication is graded separately.
Evidence
Tier 1 — cross-cohort direction replication
Tier 2 — single-cohort dose shapes
Whether a relationship changes steadily across low, middle and high groups is reported as single-cohort evidence unless independently tested.
Every number on this page traces to the preprint (Tables S1–S2; Figs. 2–3) and is kept in a single facts file.
Not proof of causation
A dependency is a measured difference in associations. It does not show that one module causes change in another.
Not a prediction of individual response
RootMap does not predict an individual's response to an intervention, and does not recommend one.
Not a substitute for controlled studies
These are observational patterns in blood-based scores. They do not replace controlled studies.
Findings
Direction asymmetry
Forward (hallmark → organ) links were more concentrated than reverse links: ΔC = +0.18, p = 0.030. Hallmark modules more often marked differences in organ–age associations than organs marked differences in hallmark–age associations.
ΔC = +0.18 · p = 0.030
Stem-cell maintenance × bone marrow
Among people with higher age-adjusted stem-cell-maintenance mIC, bone-marrow mIC was more negatively associated with age (r = −0.241); among those with lower mIC, the association was close to zero (r = +0.005). The same comparison for senescence mIC and kidney mIC: −0.281 versus −0.023.
bone marrow: −0.241 / +0.005 · kidney: −0.281 / −0.023
Carrier reversal
With Gene Ontology pathway sets, none of the 15 pairs that were significant with disease-anchored sets remained significant; PPI-expanded versions of the same sets recovered 8 of 11. In this blood-based analysis, the sets that looked most organ-specific carried the least detectable dependency signal.
GO sets: none of 15 survived · PPI-expanded: 8 of 11 recovered
TCM in the same framework
Essence was the largest hub among TCM fundamental substances, contributing 5 of 10 significant pairs. The TCM concept of liver corresponded more strongly to lymph and immune modules than to the anatomical liver (r = −0.15).
essence: 5 of 10 pairs · TCM liver r = −0.15
Measurable patterns in module scores, not validation of traditional theory.
For practice
A capability-state panel: mIC scores for the 14 foundational hallmark modules, plus organ, metabolic and immune modules — one frozen reference scale, every module scored the same way.
In a screening benchmark, filtering a PPI-ranked anti-aging candidate list by functional-layer labels made the list 2.8× more compact and raised hit enrichment from 3.19-fold to 8.87-fold (q = 0.033) — not distinguishable from draws matched for the number of drug targets (p = 0.18).
A navigation chain from a clinical problem: phenotype → organ module → hallmark regulators → related TCM concepts → compounds and herbs. These are candidate-retrieval links, not independent validation.
Candidate retrieval — not independent validation.
Across eight intervention series, functional modules changed more often than hallmark modules (25.4% versus 14.4% of tests; Fisher p = 0.020) — in the studied series, the functional layer was the one that moved more often.
Limits & roadmap
A dependency here means an association differs across levels of another module; no causation is claimed.
Module-level correlations with the broad cell-mixture signal reach |r| = 0.65–0.84. We correct for it, but blood is not organ tissue.
GO pathway sets lost the organ signal; PPI-expanded sets recovered it. Each organ claim must be read together with its gene-set construction.
Among participants older than 65, 13 of 20 pairs agreed in direction.
The frozen reference scale does not yet calibrate absolute values across platforms; we compare directions and ranks, not absolute levels.
A third cohort and cross-modal readouts (protein clocks, imaging)
Calibration studies for absolute mIC values
An interactive RootMap tool, offered as a Research Preview through steeramed.com
Paper & access
Jianghui Xiong · DeepoMe Inc.
Abstract, first sentence
“Longevity medicine needs ways to identify a small number of candidate intervention targets that might affect several functions at once.”
207 gene-set constructions, module scores, and per-pair statistics (r_high/r_low, Δz, q) underlying the figures — available from the corresponding author.
Research collaboration and independent validation are welcome.
Boundary
Patterns in blood-based scores — not proof of biological control.