GrimAge
epigenetic method · published 2019 · measured from blood (DNA methylation, Illumina array)
epigenetic
Public coefficients: GrimAge version 1: Published; GrimAge version 2: Published
Output: GrimAge (years); DNAm-based mortality risk estimate
Facts
| Method type | epigenetic |
|---|---|
| Published | 2019 |
| Sample type | blood (DNA methylation, Illumina array) |
| Output | GrimAge (years); DNAm-based mortality risk estimate |
| Validated population | Version 1: Framingham Heart Study Offspring cohort, training (n = 1,731) and test (n = 625) split by pedigree; further cohorts are evaluated in the same paper |
| Model versions | GrimAge version 1; GrimAge version 2 |
This record keeps four things apart: the published method, its model versions, implementations that expose coefficients, and commercial products. Each statement cites one source. Unresolved means no adequate public source was found. Public availability of coefficients is not a free commercial licence, a clinical approval, or proof that a result can be calculated from routine blood tests.
Model versions
| GrimAge version 1 | Year: 2019. Status: Documented. Eight DNAm surrogates (seven plasma proteins and smoking pack-years) plus age and sex. Source: Lu et al. 2019, Aging (Albany NY) 11(2):303-327 (GrimAge version 1). |
|---|---|
| GrimAge version 2 | Year: 2022. Status: Documented. Adds DNAm surrogates for logCRP and logA1C; a separate version, not covered by the version 1 record. Source: Lu et al. 2022, Aging (Albany NY) 14(23):9484-9549 (GrimAge version 2). |
Implementations and public coefficients
Biolearn coefficients file GrimAgeV1.csv
| Kind | Public software or data file |
|---|---|
| Model version | GrimAge version 1 |
| Origin | Lu et al. 2019, Aging (Albany NY) 11(2):303-327 (GrimAge version 1) |
| Public coefficients | Published - Public CSV with the per-CpG weights of the eight surrogate sub-models, the final-model weights and the transform constants (1,141 rows). It exposes the weights; it does not make GrimAge computable from routine blood tests, because the inputs are methylation array values. Source: Biolearn coefficients file GrimAgeV1.csv. |
| Documented usage rights | Partly documented - The Biolearn maintainers state non-commercial use is free; commercial use requires contacting UCLA TDG (cosmetics, life insurance) or the Clock Foundation (other). This is a maintainer summary, not a licence from the rights holders. Source: Biolearn: Model IP and Usage. |
Biolearn coefficients file GrimAgeV2.csv
| Kind | Public software or data file |
|---|---|
| Model version | GrimAge version 2 |
| Origin | Lu et al. 2022, Aging (Albany NY) 14(23):9484-9549 (GrimAge version 2) |
| Public coefficients | Published - Public CSV for version 2 (1,364 rows); version 2 weights are separate from version 1. Source: Biolearn coefficients file GrimAgeV2.csv. |
| Documented usage rights | Partly documented - Same maintainer summary as version 1: non-commercial use free, commercial use requires contact. Not a licence from the rights holders. Source: Biolearn: Model IP and Usage. |
Commercial products
No commercial product is assessed in this record.
Required inputs
- Genome-wide DNA methylation array values (blood), age and sex
Original publication
Lu AT, Quach A, Wilson JG, Reiner AP, Aviv A, Raj K, Hou L, Baccarelli AA, Li Y, Stewart JD, Whitsel EA, Assimes TL, Ferrucci L, Horvath S. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY). 2019;11(2):303-327.
DOI: 10.18632/aging.101684 · PMID: 30669119
Limitations
- Public availability of the weights is not a free commercial licence, a clinical approval, or evidence that GrimAge can be calculated from routine blood tests; the inputs are DNA methylation array values.
- Version 1 (2019) and version 2 (2022) are different models with different weights; results from one must not be read as the other.
- Built from DNAm surrogates for plasma proteins and smoking pack-years via elastic net regression, not a direct biological measurement.
- Whether a given laboratory or portal reports version 1 or version 2 is a provider-specific question that this record does not resolve.
Sources and source limits
- Lu et al. 2019, Aging (Albany NY) 11(2):303-327 (GrimAge version 1)
Read: Article text (journal copy): eight DNAm surrogates plus age and sex; final elastic net Cox model listed in Supplementary Table 2; Framingham Offspring training (n = 1,731) and test (n = 625) split; conflicts statement.
Source limits: The article is open access under CC BY 3.0; this is the article licence, not a licence granted here for commercial use of the model. Supplementary Table 2 itself was not read; the conflicts statement mentions a University of California provisional patent application whose current status was not checked. - Lu et al. 2022, Aging (Albany NY) 14(23):9484-9549 (GrimAge version 2)
Read: Abstract and first page of the open-access copy: GrimAge2 adds DNAm logCRP and DNAm logA1C; evaluated in 13,399 blood samples across nine cohorts; CC BY 3.0 notice.
Source limits: Only the abstract and opening pages were read; the CC BY 3.0 notice covers the article, not the model. - Biolearn coefficients file GrimAgeV1.csv
Read: Header Y.pred,var,beta; 1,141 data rows: eight DNAm surrogate sub-models (GDF15, B2M, CystatinC, TIMP1, ADM, PAI1, Leptin, pack-years), ten final-model weights (the eight surrogates plus Age and Female) and four transform constants; SHA-256 c787e6ec602121d688e35a36b52b4d1b2b6af2e91e1ff7057cc900e9df308dad.
Source limits: File structure and row counts were checked; its values were not reconciled row by row against Supplementary Table 2 of the 2019 paper. Biolearn attributes the file to that paper. - Biolearn: Model IP and Usage
Read: Rows for GrimAgeV1, GrimAgeV2 and DunedinPACE: non-commercial use free; commercial use requires contact (UCLA TDG for cosmetics or life insurance, Clock Foundation for other commercial use; paper author for DunedinPACE).
Source limits: A summary written by the Biolearn developers to help users; it is not a licence granted by the rights holders and was not checked with them. - Biolearn coefficients file GrimAgeV2.csv
Read: Header Y.pred,var,beta; 1,364 data rows: ten DNAm surrogate sub-models (the eight of version 1 plus logCRP and logA1C), twelve final-model weights and four transform constants; SHA-256 437290aae18c45e808b2668535e917cf381d121ab050fb2097617567fdc7f0fd.
Source limits: File structure and row counts were checked; values were not reconciled against the 2022 paper's supplementary tables.