Unlocking secrets of aging: ENABL age offers detailed, individualized aging profiles using cutting-edge AI techniques

By Pooja Toshniwal PahariaNov 9 2023Reviewed by Lily Ramsey, LLM

In a recent study published in The Lancet Healthy Longevity, researchers introduced a computational ExplaiNAble BioLogical Age (ENABL Age) estimation framework combining machine learning with explainable artificial intelligence (XAI) to predict biological age with personalized explanations.

Study: ExplaiNAble BioLogical Age (ENABL Age): an artificial intelligence framework for interpretable biological age. Image Credit: Nan_Got/Shutterstock.com

Background

Aging exacerbates various age-related disorders, such as heart disease, neurodegeneration, and cancer. The period since birth denotes the chronological age, whereas aging is the steady reduction in a biological function that increases disease or mortality risk.

Measuring an individual's aging state (i.e., biological age prediction) is crucial to understanding and treating age-related disorders and increasing lifespans. Although existing clocks used for biological age estimation are valuable, they frequently compromise interpretability and accuracy.

Studies have concentrated on first-generation biological age clocks aimed at predicting chronological age. These clocks have fewer connections with mortality risk than second-generation clocks, and their relationships with other aging outcomes are varied.

A limited number of second-generation biological age clocks, such as PhenoAge and GrimAge, were constructed directly from aging results. However, these investigations used linear models and did not give personalized explanations.

About the study

In the present study, researchers introduced the ENABL Age, an advancement over existing biological age estimation methods.

In contrast with other methods, NABL Age used an approach combining complicated machine learning with XAI techniques to directly estimate age-associated outcomes and transform the estimations into biological age.

Enhanced XAI techniques were used to derive the Shapley additive explanations (SHAP) values and determine the extent to which the input features contributed to ENABL Age estimation.

To develop the ENABL Age biological clock, the team estimated an age-associated outcome (e.g., cause-specific or all-cause mortality) using Cox proportional gradient-boosted trees (GBTs) and hazard ratios (HRs) and subsequently rescaled the estimations to predict biological age analyzing the National Health and Nutrition Examination Survey (NHANES) and United Kingdom Biobank (UKBB) data.

Individual ENABL-estimated ages were broken down into risk variables using existing XAI methodologies.

Further, the team presented practical uses of the model through two ENABL Age clock variants, i.e., ENABL Age clock-Q (using questionnaire features) and ENABL Age-L (using regular blood tests). Lastly, the aging mechanisms elucidated by the ENABL Age clocks were validated by performing association analysis using genome-wide association studies (GWAS) data.

The researchers re-estimated BioAge and PhenoAge weights (second-generation biological age estimation clocks constructed with phenotypic characteristics) on the NHANES and UKBB datasets.

Eighty percent of the data were used for training, whereas 20% were used for validation. The UKBB dataset included 501,366 samples from individuals aged between 40 and 70 years enrolled from 2007 to 2014 across Scotland, Wales, and England, and the NHANES dataset included 47,084 samples from United States (US) residents aged between 18 and 80 years enrolled from 1999 to 2014.

Features missing in most samples, highly correlated features, and individuals who died due to external reasons were excluded.

Results

The ENABL Age-estimated biological age showed significant correlations with the participants' chronological age (r values of 0.8 and 0.7 for the UKBB and NHANES datasets, respectively).

The clocks could distinguish healthy individuals (i.e., chronological age exceeding ENABL-estimated biological age) from their unhealthy counterparts (i.e., ENABL Age exceeding the chronological age), estimating mortality more efficiently than the existing age estimation clocks.

Unhealthy individuals showed 3.0- to 12-fold higher log HRs than healthy individuals in the ENABL Age estimation method.

ENABL Age attained high mortality estimation power, as indicated by the area under the receiver operating characteristic (ROC) curve (AUC) values of 0.8 for five- and ten-year mortality among UKBB participants and 0.9 for the corresponding mortality estimations among NHANES participants.

ENABL Age outperformed BioAge and PhenoAge in the validation analyses. The individualized explanations revealing the contributions of particular characteristics to the ENABL Age yielded valuable insights into aging.

Association analyses with aging-associated morbidities and risk determinants and genome-wide association studies results on the ENABL Age estimation clocks developed from several causes of mortality showed that ENABL Age captured distinct mechanisms of aging.

Moreover, ENABL Age captured more comprehensive aging-related pathways than BioAge and PhenoAge using any available feature, providing a more realistic picture of an individual's health state, with rapid ENABL Age acceleration indicating a significant rise in mortality risk.

The method's main contribution was its high interpretability. The tree models performed better than linear models for the UKBB dataset, significantly improving nearly all mortality estimation tasks (circulatory, respiratory, digestive, all-cause, neoplasms, and other causes).

Likewise, for the NHANES dataset, GBTs outperformed linear models in seven of ten mortality estimation tasks, with significant improvements in three tasks.

The superior estimation performance of the GBTs indicated that they could effectively capture mortality-related signals, which are also strongly associated with aging.

Conclusion

Based on the study findings, ENABL Age is a significant advancement in using machine learning and XAI for estimating biological age. This advancement broadens the scope of biological age estimation studies.

ENABL Age has also shown remarkable promise in clinical contexts, where it might assist health professionals in unraveling the complexities of aging-related systems, thereby proving beneficial to informed clinical decision-making.