Methylation Clocks Determine Biological Age
22nd September 2021 - Last modified 3rd November 2023
Shifting Sands
The idea of lifespan prediction is as old as life itself, with humans attempting to foretell their own deaths, in various ways, in virtually every known culture across time. There’s something wildly uncomfortable about the conflict between the certainty of death and the uncertainty of its timing. We are willing to try anything to avoid the uncertainty of the future, and even in science-based western cultures, ancient practices such as tarot and palm reading are still popular despite widespread scepticism. So, we are always on the path to discover ever-more modern and sophisticated techniques to pacify our need for certainty. In today’s technological age, a popular option for this is a prediction from artificial intelligence (AI) models.
In my PhD research, I use AI to predict the cellular lifespan of yeast. Yeast cells are eukaryotic and insights into the ageing of yeast cells can translate to understanding of the ageing of all eukaryotic cells, including human cells. But AI lifespan predictions are not restricted to single cells and the concept is not a new idea.
With a big rise in health consciousness in recent years, it has become almost a generic term to discuss people’s ‘biological’ or ‘real’ age as opposed to their chronological age. We have developed a widespread understanding that different factors in our lives can contribute to the global health or ‘age’ of our bodies. This can be measured by many different markers from blood test results to brain scans. But the last decade has seen a big rise in research attempting to determine biological age from these markers using AI.
Can DNA hold the key to predicting the sands of time?
In a bind
Ageing is the result of a complex and poorly understood interplay between many different biological systems. Because of this, it is understandably difficult to define and research, so often single elements of the ageing process are researched individually. But how do these relate to the bigger picture that is ageing?
DNA methylation – the addition of a methyl group (CH3) directly to the DNA strand, usually a cytosine (C) base – is a very long-standing field of ageing research. It’s a normal and necessary part of the epigenetic control of gene expression used to ‘switch’ genes on/off [1].
Areas rich in cytosine and guanine (G) bases, known as CpG islands are particularly interesting with regards to DNA methylation and ageing. Studies have shown that as we age, the level of DNA methylation at our CpG islands increases, causing hypermethylation of these areas[2]. The natural question following this discovery was: can we use DNA methylation levels to determine biological age?
Methylation of the DNA base cytosine by the addition of a methyl group.
Tick-tock goes the ageing clock
The original Ageing Clock was published in 2013 by Professor Steve Hovarth at UCLA[2]. It was a hugely successful attempt to use DNA methylation levels to determine biological age, which is now more commonly known as the Hovarth Clock.
Hovarth leveraged huge pre-existing datasets to build his model, taking advantage of one of AI’s biggest selling points within biology – breathing new life into old data. In many fields of biological research, there are vast quantities of data available, the bottleneck is how to analyse them in a way that gets the best insight from the data. Hovarth built his clock using 8,000 DNA methylation samples which included 51 healthy tissue and cell types to ensure that the model was not limited to one tissue type. From this, he was able to identify 353 CpG regions that are predictive of age and could be used to create the clock.[2]
The Hovarth Clock relies entirely on linear regression. This means that it looks for linear relationships between the predictive factors (DNA methylation levels) and age. Using his model, Hovarth was able to predict chronological age from the DNA methylation levels at 353 CpG sites, known as the DNAm Age, with strong statistical significance across a range of tissue and cell types[2]. Linear regression is the logical, popular first step in machine learning and AI, but there is so much more to explore…
The Hovarth Clock is based on levels of DNA methylation.
Digging deeper
While the Hovarth Clock was a jaw-dropping success, AI has since taken great strides – notably in neural networks and deep learning . Unlike the simplicity of linear models, neural networks are a type of model with a structure loosely designed to mimic the human brain with layers of interconnected nodes. Deep learning takes this one step further by expanding this layered structure and increasing the number of layers. This allows the network to ‘train itself’ and make deeper, more complex connections from the data.[3]
Published just last month (August 2021), the DeepMAge clock is the first attempt at a deep learning DNA methylation ageing clock that out-performs linear models such as the Hovarth clock[4]. Not only can DeepMAge more accurately predict chronologoical age, but it is also able to account for the effect of disease by increasing the predicted age for patients with diseases such as ovarian cancer, irritable bowel diseases and multiple sclerosis[4]. This is another exciting step forward in ageing modelling as diseases are notoriously difficult to model but have huge influence over health, age and lifespan.
Deep learning is a type of neural network with many layers which trains itself.
But… what is ageing?
So, what does all of this mean for lifespan prediction? While DNA methylation clocks are becoming more and more accurate using newer AI techniques, how can we be sure that AI predictions determine biological age?
The DeepMAge clock is a step in the right direction in its ability to augment its predicitions for patients with diseases, but we are still a long way from predicting true biological age and lifespans. Deep learning, with its ability to handle enormous quantities of input data, offers the chance to continue incorporating new aspects of ageing into models until they can finally reach the goal of lifespan prediction.
Blog Author
Olivia Hillson is a LIDo BBSRC funded PhD student at the Bähler lab, UCL, UK. Her PhD research uses a combination of molecular biology and machine learning techniques to look into cellular ageing.
She is currently undertaking a 12-week internship in science writing and scientific content for SEO, with Alto Marketing – the healthcare and life science marketing agency.
To find out more about how Alto Marketing can help you with your scientific content marketing – Contact us
References
1. Gregory Hannum, et al., Genome-wide methylation profiles reveal quantitative views of human aging rates. Molecular cell, 2013. 49(2): p. 359-367.
2. Steve Horvath, DNA methylation age of human tissues and cell types. Genome biology, 2013. 14(10): p. R115-R115.
3. Jason Brownlee, What is Deep Learning? 2019: Macine Learning Mastery.
4. Fedor Galkin, et al., DeepMAge: A Methylation Aging Clock Developed with Deep Learning. Aging and disease, 2021. 12(5): p. 1252-1262.
