As the year comes to its end, our epigenetic clocks keep on ticking, and hot new DNA methylation-based studies keep rolling in. Here, we gaze into two deeply interesting studies that take us on an epigenetic journey from deep space to deep learning by reporting how space travel can influence epigenetic aging and how neural networks can make epigenetic age predictions more accurate.
The Epigenetic Clock Ticks Slower on the Way to the Stars
Jaunts into the far-flung corners of deep space are likely to expose astronauts to a certain amount of “stress” (The Martian, Apollo 11, Alien – have you seen a movie where things go right on a spaceship!?); therefore, many have surmised that longer journeys should lead to increased epigenetic aging and the prevalence of potentially harmful consequences. But just how quick will the epigenetic clock tick for future space crews?
A recent article led by Andres Cardenas (University of California, Berkeley) evaluated the potential acceleration of epigenetic aging in blood samples taken from the six “astronauts” that formed the Mars-500 mission – a 520-day psycho-social isolation experiment (sounds fun!?) that aimed to prepare for a crewed spaceflight to Mars. However, in keeping with most other deep space adventures (2001: A Space Odyssey, Aliens, Interstellar – do I need to go on?!), the results of this epigenetic mission threw up a few surprises.
So, what deep insight did Nwanaji-Enwerem and colleagues discover about our epigenetic clocks as we make our way to the stars?
- Unexpectedly, analyses using five different DNA-methylation-based metrics discovered significant decreases to Mars-500 participant’s epigenetic age during the mission
- Decreases in epigenetic aging according to DNAm PhenoAge, a marker of disease risk that strongly correlates with chronological age, remain significant after the mission
- The six astronauts also displayed significant alterations to the proportions of plasmablast, CD4T, CD8 naive, and natural killer cells
- However, only decreases to natural killer cell levels (a type of cytotoxic lymphocyte) remain significant after the mission
The authors hope that their far-out findings may aid the future understanding of the health risks posed by those brave souls undertaking missions into the darkest corners of our galaxy. Could we return to earth (epigenetically) younger?
Deep Learning Improves Epigenetic Clock Accuracy
Estimating age using DNA-methylation-based metrics relies on the unerring accuracy of the cogs, levers, and pivots of the epigenetic clock, which are represented by 353 CpG sites on Illumina DNA methylation array platforms in the case of Horvath’s widely used clock. In a profoundly interesting new study, researchers led by Alex Zhavoronkov (Deep Longevity Limited/Insilco Limited, Hong Kong) explored “deep learning” approaches (specifically, neural networks) to create a DNA methylation-based metric with more accuracy than the 353 CpG clock. Of note, deep learning has already been employed to accurately diagnose brain damage and measure blood flow parameters from magnetic resonance imaging scans and may play a key role in health-based science in the future.
So, let’s take a deep dive into this unfathomably exciting study from Galkin and colleagues that describes a neural network called DeepMAge as a more accurate means to measure epigenetic age:
- The DeepMAge neural network, which uses a set of 1,000 CpG sites, was trained on 4,930 blood DNA methylation profiles from seventeen studies to predict age with a median absolute error (MedAE) of 2.24 and 3.29 years for the control- (healthy-) and case-cohort, respectively
- DeepMAge analysis of an independent verification set of 1,293 samples from fifteen studies provided a MedAE of 2.77 and 4.18 years for the control- (healthy-) and case-cohort, respectively
- DeepMAge assigns patients with ovarian cancer (+1.7 years), irritable bowel disease (+1.23 years), and multiple sclerosis (+2.10 years) with a higher predicted age
- In comparison, the classic 353 CpG clock provides significantly greater MedAE values of 3.56 years (original publication) and 3.51 (control cohort)
- DeepMAge outperforms the 353 CpG clock in thirteen of the fifteen verification datasets
- The 353 CpG clock also didn’t predict an increase in age in donors with irritable bowel disease
The authors hope that DeepMAge, as the first deep learning epigenetic clock, will allow for an in-depth understanding of how epigenetic landscapes shift over time and potentially function as a health marker following further research to ensure reproducibility and robustness in independent longitudinal studies.
The first author of the paper, Fedor Galkin, comments “Aging clocks have come a long way since the first works by Horvath and Hannum in 2013. We are happy to contribute to this research field. Now, we are going to explore how epigenetic aging can be slowed down with the interventions available to consumers.”
The Epigenetic Clock Ticks into the Future
As the clock ticks down towards the end of 2020, truly an annus horribilis, we can only look forward to all the deeply exciting innovative studies on DNA methylation-based epigenetic clocks that 2021 holds in promise!
For now, see how our epigenetic clocks may tick slower on the way to the stars at Cell Reports, December 2020, and how deep learning can help to predict epigenetic age with enhanced accuracy at Aging and Disease, December 2020.