Analyzing the multidimensionality of aging by using machine learning to predict age from diverse medical datasets
Author
Le Goallec, Alan
Metadata
Show full item recordCitation
Le Goallec, Alan. 2021. Analyzing the multidimensionality of aging by using machine learning to predict age from diverse medical datasets. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.Abstract
The world population is aging, leading to a rise in the prevalence of age-related diseases such as cardiovascular disease and cancer. In parallel to treating the diseases, an attractive idea is to address the problem at its root by slowing aging. In contrast to a person’s chronological age, which solely measures the time since their birth, biological age represents the state of the person’s body, is the true underlying cause of age-related diseases and could potentially be reduced by rejuvenating therapies. Training machine learning algorithms to predict chronological age from biomedical datasets yields biological age predictors. Such predictors, such as the DNA methylation aging clock, can be used to further our understanding of aging as a biological process, to identify new mechanisms or lifestyle changes susceptible to slow down aging, and to assess the efficiency of upcoming rejuvenating therapies. A large number of biological age predictors have already been built on various biomedical datasets such as brain MRI images, full body X-rays and blood samples, but it is currently largely unknown whether these predictors all capture a single, central biological aging process, or if each predictor captures a particular dimension of the complex and multi-faceted process that is aging. In other words, can a forty-year-old participant have the arteries of a fifty-year-old person, and the musculoskeletal health of a thirty-year-old person? In the following, we studied how the correlation structure between blood and anthropometric biomarkers changes with age (Chapter 1), and we predicted infant’s age from their gut microbiome with a R2 of 62.5±2.1% (Chapter 2). Finally, (Chapter 3), we leveraged 676,787 samples from 502,211 UK Biobank participants to build 331 age predictors on scalar biomarkers (e.g. blood samples), time series (e.g. electrocardiograms), images (e.g. full body X-rays) and videos (heart MRI). We conclude that, although the different aging dimensions are similarly associated with environmental and socioeconomic variables, they are both phenotypically and genetically largely uncorrelated with each other (respective average correlations of .139±.090 and .104±.149), highlighting the multidimensionality of the aging process.Terms of Use
This article is made available under the terms and conditions applicable to Other Posted Material, as set forth at http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAACitable link to this page
https://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37368352
Collections
- FAS Theses and Dissertations [6902]
Contact administrator regarding this item (to report mistakes or request changes)